diff --git a/.editorconfig b/.editorconfig new file mode 100644 index 0000000000..d54392c567 --- /dev/null +++ b/.editorconfig @@ -0,0 +1,12 @@ +root = true + +[*] +indent_style = space +indent_size = 2 +end_of_line = lf +charset = utf-8 +trim_trailing_whitespace = true +insert_final_newline = true + +[*.sh] +indent_size = 4 diff --git a/.env.example b/.env.example index 963c0c58d1..2206a5d566 100644 --- a/.env.example +++ b/.env.example @@ -92,8 +92,28 @@ PORT=20128 # Comma-separated extra origins allowed to open a live WebSocket. The # loopback dashboard origins are already permitted by default; use this # var when fronting the server with a domain (e.g. https://omni.local). +# ⚠️ When using NEXT_PUBLIC_LIVE_WS_PUBLIC_URL or exposing the WS server +# beyond loopback, this MUST include the public origin(s) — otherwise +# the Origin allow-list check will reject all browser connections. +# Example: LIVE_WS_ALLOWED_ORIGINS=https://omni.local,https://dashboard.example.com,https://ws.my-ai.com # LIVE_WS_ALLOWED_ORIGINS=https://omni.local,https://dashboard.example.com +# Comma-separated extra hostnames allowed to open a live WebSocket (LAN/Tailscale). +# Unlike LIVE_WS_ALLOWED_ORIGINS (which matches full origin URLs), this matches +# only the host portion — useful for wildcard-ish LAN/Tailscale setups. +# Used by: src/server/ws/liveServerAllowList.ts +# Example: LIVE_WS_ALLOWED_HOSTS=omni.local,tailscale-host,192.168.1.50 +# LIVE_WS_ALLOWED_HOSTS=omni.local,tailscale-host,192.168.1.50 + +# Public URL for the live dashboard WebSocket (client-side, browser only). +# Set this when fronting the WS server with a reverse proxy or Cloudflare Tunnel. +# The browser will connect to this URL instead of ws://hostname:20129. +# The /live-ws path is already proxied from the main app (port 20128) to the +# live WS server (port 20129) by scripts/dev/standalone-server-ws.mjs. +# Used by: src/hooks/useLiveDashboard.ts +# Example: NEXT_PUBLIC_LIVE_WS_PUBLIC_URL=wss://ws.my-ai.com/live-ws +# NEXT_PUBLIC_LIVE_WS_PUBLIC_URL= + # Disable the standalone live WebSocket helper used by scripts/start-ws-server.mjs. # Used by: scripts/start-ws-server.mjs (CI/embedded harness toggle). # OMNIROUTE_DISABLE_LIVE_WS=0 @@ -239,10 +259,30 @@ ALLOW_API_KEY_REVEAL=false # Default: 10485760 (10 MB) # MAX_BODY_SIZE_BYTES=10485760 +# Heap-pressure-aware admission for POST /v1/chat/completions (#5152). A large +# coding-agent "compact" body amplifies into hundreds of MB of transient JS objects +# on the combo path; concurrent compacts can stack past the V8 heap ceiling and OOM +# the process. These shed a LARGE body with 503 (Retry-After) only while the heap is +# already under pressure — healthy heap admits every body untouched. +# Used by: src/shared/middleware/chatBodyAdmission.ts +# Bodies below this size skip the guard entirely (heap not even sampled). Default 262144 (256 KB). +# OMNIROUTE_CHAT_LARGE_BODY_BYTES=262144 +# Hard cap — bodies above this are rejected with 413 before any clone/parse. Default 52428800 (50 MB). +# OMNIROUTE_CHAT_HARD_MAX_BODY_BYTES=52428800 +# Shed large bodies once heapUsed/heap_size_limit reaches this ratio (0). # Set this when OpenWebUI or another relay reaches OmniRoute by an internal URL # but the user's browser must fetch images from a LAN, tunnel, or public origin. @@ -509,6 +550,14 @@ NEXT_PUBLIC_ENABLE_SOCKS5_PROXY=true # Allow OmniRoute to write CLI config files (token refresh, etc.). # CLI_ALLOW_CONFIG_WRITES=true +# Auto-sync CLI profile files after provider model discovery changes. OPT-IN, default OFF for +# both. When enabled, writes only the tool's profile files (~/.codex/*.config.toml or +# ~/.claude/profiles//settings.json); never changes the active/default config. Both also +# require CLI_ALLOW_CONFIG_WRITES (default on). Toggle from the CLI Code dashboard, or set here. +# Leave unset to disable. (Feature flags — a DB/dashboard override takes precedence over env.) +# OMNIROUTE_AUTO_SYNC_CODEX_PROFILES=true +# OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES=true + # Override binary paths for individual CLI tools. # CLI_CLAUDE_BIN=claude # CLI_CODEX_BIN=codex @@ -654,6 +703,25 @@ PROVIDER_LIMITS_SYNC_SPACING_MS=1500 # Used by: open-sse/services/compression/engines/rtk/filterLoader.ts. Default: 0. #OMNIROUTE_RTK_TRUST_PROJECT_FILTERS=0 +# T02 stacked-pipeline engine circuit-breaker (OPT-IN, default off). When enabled, a compression +# engine that throws repeatedly across requests is skipped (fail-open) for a cooldown. +# Used by: open-sse/services/compression/pipelineEngineBreaker.ts. +#COMPRESSION_PIPELINE_BREAKER_ENABLED=false # master switch (default false) +#COMPRESSION_PIPELINE_BREAKER_THRESHOLD=3 # consecutive failures before the engine opens +#COMPRESSION_PIPELINE_BREAKER_COOLDOWN_MS=30000 # ms the engine stays skipped before a probe + +# T08/H8 — CCR retrieval-feedback ramp factor. Each prior retrieval of a stored block raises its +# effective minChars linearly, so frequently-retrieved content is compressed progressively less +# (>= 3 retrievals = never compressed). 1 disables the ramp (binary skip at the threshold only). +# Used by: open-sse/services/compression/engines/ccr/index.ts. Default: 2. +#COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR=2 +# T08/H5 — usage-observed prefix freeze (OPT-IN, default off). When enabled, a system prompt seen +# >= THRESHOLD times is treated as a stable cacheable prefix and preserved from compression even +# for providers the static cache-aware heuristic does not recognize (freeze = preserve, never +# mutates). Used by: open-sse/services/compression/prefixFreeze.ts. +#COMPRESSION_PREFIX_FREEZE_ENABLED=false # master switch (default false) +#COMPRESSION_PREFIX_FREEZE_THRESHOLD=3 # observations before a prefix is frozen + # Skip the postinstall native-runtime warm-up (useful in CI / headless installs). Default: 0. # Used by: scripts/postinstall.mjs. #OMNIROUTE_SKIP_POSTINSTALL=0 @@ -756,7 +824,7 @@ GITHUB_OAUTH_CLIENT_ID=Iv1.b507a08c87ecfe98 # ── GitLab Duo ── # Register an OAuth app at: https://gitlab.com/-/profile/applications # Set redirect URI to: http://localhost:20128/callback (or your NEXT_PUBLIC_BASE_URL + /callback) -# Required scopes: api, read_user, openid, profile, email +# Required scopes: ai_features, read_user (matches GITLAB_DUO_CONFIG.scope in src/lib/oauth/constants/oauth.ts) # GITLAB_DUO_OAUTH_CLIENT_ID=*** # GITLAB_DUO_OAUTH_CLIENT_SECRET=*** # optional — PKCE flow does not require a secret # @@ -1035,6 +1103,17 @@ CURSOR_USER_AGENT="Cursor/3.4" # STREAM_IDLE_TIMEOUT_MS=600000 # Max silence between SSE chunks (default: 600000) # # Extended-thinking models rarely pause >90s. # STREAM_READINESS_TIMEOUT_MS=80000 # Time to receive the first non-ping SSE event +# STREAM_READINESS_MAX_TIMEOUT_MS=180000 # Cap for adaptive first-event extensions +# # (large/tool-heavy/high-reasoning requests). +# OMNIROUTE_AGENT_GOAL_POLICY_ENABLED=true # Kill-switch for the /goal heuristic below. +# # Set to false to fully disable detection — +# # readiness timeouts and stream recovery are +# # never elevated by request body/headers when off. +# OMNIROUTE_AGENT_GOAL_READINESS_MAX_TIMEOUT_MS=600000 # Auto cap for detected /goal agent runs +# OMNIROUTE_AGENT_GOAL_STREAM_RECOVERY=true # Auto early stream recovery for /goal runs. +# # NOTE: this can only ADD recovery on top of the +# # operator default — it never overrides an explicit +# # STREAM_RECOVERY_ENABLED / DB settings opt-out. # ── TLS client (wreq-js fingerprint proxy) ── # TLS_CLIENT_TIMEOUT_MS=600000 # Inherits from FETCH_TIMEOUT_MS by default @@ -1245,6 +1324,13 @@ APP_LOG_TO_FILE=true # Default: 86400 (24 hours) # MODELS_DEV_SYNC_INTERVAL=86400 +# Self-correcting context-window reconciler interval in seconds (feature 5004). +# Pins provider-declared windows from /models discovery as auto:discovery overrides +# when they diverge from the catalog. Set to 0 to disable. Never overwrites manual overrides. +# Used by: src/lib/contextWindowResolver.ts +# Default: 86400 (24 hours) +# CONTEXT_WINDOW_RECONCILE_INTERVAL=86400 + # ═══════════════════════════════════════════════════════════════════════════════ # 20. PROVIDER-SPECIFIC SETTINGS # ═══════════════════════════════════════════════════════════════════════════════ @@ -1731,6 +1817,11 @@ PLAYGROUND_COMPARE_MAX_COLUMNS=4 # MEMORY_VEC_TOP_K=20 # default top-K for vector search # MEMORY_RRF_K=60 # RRF k constant (sqlite-vec hybrid recipe) # HF_HUB_ENDPOINT=https://huggingface.co # override Hugging Face Hub base URL for static potion downloads +# TV6 typed memory decay (OPT-IN, default off — the sweep DELETES decayed memories) +# MEMORY_TYPED_DECAY_ENABLED=false # master switch for the destructive sweep (default off) +# MEMORY_TYPED_DECAY_EPISODIC_DAYS=30 # episodic TTL in days; 0 = episodic immune too +# MEMORY_TYPED_DECAY_ACCESS_IMMUNITY=3 # access_count >= N → immune; 0 disables access immunity +# MEMORY_TYPED_DECAY_SWEEP_INTERVAL=0 # periodic sweep interval (seconds); 0 = no periodic sweep # AgentBridge + Traffic Inspector (Group A) # AgentBridge diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml index d69f0921b9..906c02bb7f 100644 --- a/.github/workflows/ci.yml +++ b/.github/workflows/ci.yml @@ -624,7 +624,7 @@ jobs: cache: npm - uses: ./.github/actions/npm-ci-retry - run: npm run check:node-runtime - - run: node --max-old-space-size=4096 --import tsx --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit --test-concurrency=4 --test-shard=${{ matrix.shard }}/8 tests/unit/*.test.ts "tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts" + - run: node --max-old-space-size=4096 --import tsx --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit --test-concurrency=4 --test-shard=${{ matrix.shard }}/8 tests/unit/*.test.ts "tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,memory,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts" test-vitest: name: Vitest (MCP / autoCombo / UI components) @@ -676,7 +676,7 @@ jobs: cache: npm - uses: ./.github/actions/npm-ci-retry - run: npm run check:node-runtime - - run: node --max-old-space-size=4096 --import tsx --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit --test-concurrency=4 --test-shard=${{ matrix.shard }}/4 tests/unit/*.test.ts "tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts" + - run: node --max-old-space-size=4096 --import tsx --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit --test-concurrency=4 --test-shard=${{ matrix.shard }}/4 tests/unit/*.test.ts "tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,memory,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts" node-26-compat-build: name: Node 26 Compatibility Build @@ -729,7 +729,7 @@ jobs: cache: npm - uses: ./.github/actions/npm-ci-retry - run: npm run check:node-runtime - - run: node --max-old-space-size=4096 --import tsx --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit --test-concurrency=4 --test-shard=${{ matrix.shard }}/4 tests/unit/*.test.ts "tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts" + - run: node --max-old-space-size=4096 --import tsx --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit --test-concurrency=4 --test-shard=${{ matrix.shard }}/4 tests/unit/*.test.ts "tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,memory,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts" test-coverage-shard: name: Coverage Shard (${{ matrix.shard }}/8) @@ -770,7 +770,7 @@ jobs: --exclude=tests/** \ --exclude=**/*.test.* \ node --max-old-space-size=4096 --import tsx --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit --test-concurrency=4 \ - --test-shard=${{ matrix.shard }}/8 tests/unit/*.test.ts "tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts" + --test-shard=${{ matrix.shard }}/8 tests/unit/*.test.ts "tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,memory,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts" - name: Upload raw shard coverage if: always() uses: actions/upload-artifact@v7 @@ -889,12 +889,28 @@ jobs: else echo "SonarQube scan skipped because SONAR_TOKEN or SONAR_HOST_URL is not configured." >> "$GITHUB_STEP_SUMMARY" fi + - name: Read project version + if: ${{ github.event_name == 'pull_request' && env.SONAR_TOKEN != '' && env.SONAR_HOST_URL != '' }} + id: pkg + run: | + VERSION="$(node -p "require('./package.json').version")" + # Harden against output injection: only accept a plain semver-ish string. + if [[ ! "$VERSION" =~ ^[0-9]+\.[0-9]+\.[0-9]+([.-][0-9A-Za-z.-]+)?$ ]]; then + echo "Invalid package.json version: $VERSION" >&2; exit 1 + fi + echo "version=$VERSION" >> "$GITHUB_OUTPUT" - name: SonarQube Scan if: ${{ github.event_name == 'pull_request' && env.SONAR_TOKEN != '' && env.SONAR_HOST_URL != '' }} uses: SonarSource/sonarqube-scan-action@v8 env: SONAR_TOKEN: ${{ env.SONAR_TOKEN }} SONAR_HOST_URL: ${{ env.SONAR_HOST_URL }} + with: + # sonar.projectVersion advances the "previous_version" new-code baseline at + # each release; without it the baseline never moves (it stays pinned to the + # analysis where the version last changed) and legacy issues pile up as + # "new code" in the quality gate. + args: -Dsonar.projectVersion=${{ steps.pkg.outputs.version }} coverage-pr-comment: name: PR Coverage Comment diff --git a/.github/workflows/quality.yml b/.github/workflows/quality.yml index 4242fb6648..bc9a307d3b 100644 --- a/.github/workflows/quality.yml +++ b/.github/workflows/quality.yml @@ -52,6 +52,10 @@ jobs: - run: npm run check:route-guard-membership - run: npm run check:test-discovery - run: npm run check:test-runner-api + # Guards tap.testFiles drift: a covering unit test absent from stryker.conf.json + # tap.testFiles makes its module's mutants survive on a cold nightly-mutation run, + # false-failing the blocking mutationScore ratchet. See check-mutation-test-coverage.mjs. + - run: npm run check:mutation-test-coverage - run: npm run check:any-budget:t11 # Build-scope guard: fails if worktrees/cruft leak into the tsconfig include # scope (would OOM `next build`). Instant. See incident 2026-06-25 / #5031. diff --git a/.gitignore b/.gitignore index efdd25c7e5..bb05ae4c44 100644 --- a/.gitignore +++ b/.gitignore @@ -229,3 +229,6 @@ docs/prompts/AGENT-OWNERSHIP-PROTOCOL.md docs/prompts/AGENT-OWNERSHIP-PROTOCOL.omniroute-mim.md docs/prompts/AGENT-OWNERSHIP-PROTOCOL.omniroute-mid.md omniroute.md + +# mise configuration +mise.toml diff --git a/@omniroute/opencode-plugin/package-lock.json b/@omniroute/opencode-plugin/package-lock.json index a21cdf7c0d..b75338eb09 100644 --- a/@omniroute/opencode-plugin/package-lock.json +++ b/@omniroute/opencode-plugin/package-lock.json @@ -1,12 +1,12 @@ { "name": "@omniroute/opencode-plugin", - "version": "0.1.0", + "version": "0.2.0", "lockfileVersion": 3, "requires": true, "packages": { "": { "name": "@omniroute/opencode-plugin", - "version": "0.1.0", + "version": "0.2.0", "license": "MIT", "dependencies": { "zod": "^4.4.3" diff --git a/AGENTS.md b/AGENTS.md index 289ed00296..ce665709a4 100644 --- a/AGENTS.md +++ b/AGENTS.md @@ -3,12 +3,12 @@ ## Project Unified AI proxy/router — route any LLM through one endpoint. Multi-provider support -with **236 provider entries** (OpenAI, Anthropic, Gemini, DeepSeek, Groq, xAI, Mistral, Fireworks, +with **237 provider entries** (OpenAI, Anthropic, Gemini, DeepSeek, Groq, xAI, Mistral, Fireworks, Cohere, NVIDIA, Cerebras, Pollinations, Puter, Cloudflare AI, HuggingFace, DeepInfra, SambaNova, Meta Llama API, Moonshot AI, AI21 Labs, Databricks, Snowflake, and many more) with **MCP Server** (94 tools), **A2A v0.3 Protocol**, and **Electron desktop app**. -> **Live counts (v3.8.40)**: providers 236 · MCP tools 94 · MCP scopes 30 · A2A skills 6 · +> **Live counts (v3.8.40)**: providers 237 · MCP tools 94 · MCP scopes 30 · A2A skills 6 · > open-sse services 298 · routing strategies 17 · auto-combo scoring factors 12 · > DB modules 94 · DB migrations 106 · base tables 17 · search providers 11 · > i18n locales 42. **Refresh with `npm run check:docs-all`.** diff --git a/CHANGELOG.md b/CHANGELOG.md index 32c5c413f5..468959521e 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -4,6 +4,338 @@ --- +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/CLAUDE.md b/CLAUDE.md index b5b377e4bd..dcfda7c548 100644 --- a/CLAUDE.md +++ b/CLAUDE.md @@ -35,7 +35,7 @@ For full test matrix, see `CONTRIBUTING.md` → "Running Tests". For deep archit ## Project at a Glance -**OmniRoute** — unified AI proxy/router. One endpoint, 236 LLM providers, auto-fallback. +**OmniRoute** — unified AI proxy/router. One endpoint, 237 LLM providers, auto-fallback. | Layer | Location | Purpose | | ------------- | ----------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------ | @@ -47,7 +47,7 @@ For full test matrix, see `CONTRIBUTING.md` → "Running Tests". For deep archit | Services | `open-sse/services/` | Combo routing, rate limits, caching, etc | | Database | `src/lib/db/` | SQLite domain modules (94 files, 106 migrations) | | Domain/Policy | `src/domain/` | Policy engine, cost rules, fallback logic | -| MCP Server | `open-sse/mcp-server/` | 94 tools (34 base + memory/skill/agentSkill/pool/notion/obsidian/gamification/plugin modules), 3 transports (stdio / SSE / Streamable HTTP), 30 scopes | +| MCP Server | `open-sse/mcp-server/` | 95 tools (35 base + memory/skill/agentSkill/pool/notion/obsidian/gamification/plugin modules), 3 transports (stdio / SSE / Streamable HTTP), 30 scopes | | A2A Server | `src/lib/a2a/` | JSON-RPC 2.0 agent protocol | | Skills | `src/lib/skills/` | Extensible skill framework | | Memory | `src/lib/memory/` | Persistent conversational memory | @@ -480,7 +480,8 @@ list` shows worktrees you didn't create, leave them alone. End every session wit ## Environment -- **Runtime**: Node.js ≥22.0.0 <23 || ≥24.0.0 <27, ES Modules +- **Runtime**: Node.js ≥22.0.0 <23 || ≥24.0.0 <27, ES Modules. This is the **only** runtime for the published `omniroute` CLI, the server, and the test suites (`node:test` + vitest) — `engines.node` is authoritative and end users never need Bun. +- **Bun (build/dev script runner only)**: Bun `1.3.10` is pinned as an **exact devDependency** (provisioned through the existing `npm ci` via the lockfile's `@oven/bun-*` platform binaries — no `setup-bun`/ad-hoc install). It is used **only** to execute a small, allow-listed set of TypeScript **gate/generator scripts** (replacing `node --import tsx` for startup speed): the CI checks `check:provider-consistency`, `check:compression-budget`, `check:known-symbols`, and the non-CI `gen:provider-reference`, `bench:compression`. **Do NOT** widen Bun to `npm install`, the build (`build:cli*`), `check:pack-artifact`, the published runtime, or the test runners — those stay on Node. Any new Bun-invoking script must be validated byte-identical against its `node --import tsx` output first. After pulling the lockfile change, run `npm install` so `bun` resolves locally (a stale `node_modules` will fail those 5 scripts with `bun: not found`). - **TypeScript**: 6.0+, target ES2022, module esnext, resolution bundler - **Path aliases**: `@/*` → `src/`, `@omniroute/open-sse` → `open-sse/`, `@omniroute/open-sse/*` → `open-sse/*` - **Default port**: 20128 (API + dashboard on same port) @@ -539,7 +540,7 @@ the stale-enforcement added in Fase 6A.3. 18. Every bug fix must be validated before shipping: a failing-then-passing unit/integration test (TDD) OR a documented live test on the production VPS (192.168.0.15). A fix without either is not merged. See Testing → "Bug fix / issue triage protocol" for the full decision tree. 19. Never develop on the shared main checkout. Every development task runs in its own git worktree on its own dedicated branch, and you MUST confirm the base branch with the operator (e.g. via `AskUserQuestion`) before creating the worktree/branch — never assume `main` or the currently checked-out branch. A `git checkout` in the shared checkout silently destroys other sessions' uncommitted work. Tear down only the worktrees/branches you created (by name, never `fix/*`/`feat/*` wildcards), leave other sessions' worktrees untouched, and end on the branch you started on (the active `release/vX.Y.Z`, never `main`). See Git Workflow → "Worktree isolation". 20. PII redaction/sanitization is **opt-in — never on by default**. OmniRoute proxies for self-hosted/local LLMs where the operator owns the data, so mutating request/response payloads by default would silently corrupt legitimate traffic. The two data-mutating PII feature flags **MUST** keep `defaultValue: "false"` in `src/shared/constants/featureFlagDefinitions.ts`: `PII_REDACTION_ENABLED` (request-side) and `PII_RESPONSE_SANITIZATION` (response + streaming). All three application points — `src/lib/guardrails/piiMasker.ts` (request guardrail), `src/lib/piiSanitizer.ts` (response), `src/lib/streamingPiiTransform.ts` (SSE) — are gated on these flags; with both off the `pii-masker` guardrail still runs but never mutates payloads (data passes through untouched). Flipping either default to `"true"` requires explicit operator approval. The regression guard is `tests/unit/pii-opt-in-default.test.ts` (asserts both definition defaults + behavioral pass-through). Opt-in is per-operator via env or the settings/DB override (`src/lib/db/featureFlags.ts`), never a silent default. See `docs/security/GUARDRAILS.md`. -21. **Release-freeze — the release branch is frozen to campaign merges while a `/generate-release` is running.** `/generate-release` opens a marker issue labeled `release-freeze` at the start of reconciliation (Phase 0a) and closes it once the release PR squash-merges to `main`. Before merging **any** PR into the active `release/vX.Y.Z` branch, every campaign workflow (`/review-issues`, `/review-prs`, `/implement-features`, `/green-prs`, `/port-upstream-*`) **MUST** check `gh issue list --repo diegosouzapw/OmniRoute --label release-freeze --state open` — if a freeze is active, **HOLD the merge** (leave the PR ready and open; do NOT merge to the release branch), tell the operator, and resume once the freeze lifts. This is a **coordination signal, not a permission lock**: the release captain and the campaign sessions share the `diegosouzapw` identity, so a GitHub branch-protection lock cannot distinguish them — only this honored marker prevents the mid-release commit races that forced full CHANGELOG re-reconciliation in v3.8.40/v3.8.41 (a parallel campaign advanced `release/vX.Y.Z` by 34 commits mid-run). The release captain's own reconciliation/cycle-open pushes are exempt — they *are* the release. Fixes that must land during a freeze (a homologation finding) follow the post-merge read-only rule: land on `main` first via `fix/release-vX.Y.Z-*`. +21. **Release-freeze — the release branch is frozen to campaign merges while a `/generate-release` is running.** `/generate-release` opens a marker issue labeled `release-freeze` at the start of reconciliation (Phase 0a) and closes it once the release PR squash-merges to `main`. Before merging **any** PR into the active `release/vX.Y.Z` branch, every campaign workflow (`/review-issues`, `/review-prs`, `/implement-features`, `/green-prs`, `/port-upstream-*`) **MUST** check `gh issue list --repo diegosouzapw/OmniRoute --label release-freeze --state open` — if a freeze is active, **HOLD the merge** (leave the PR ready and open; do NOT merge to the release branch), tell the operator, and resume once the freeze lifts. This is a **coordination signal, not a permission lock**: the release captain and the campaign sessions share the `diegosouzapw` identity, so a GitHub branch-protection lock cannot distinguish them — only this honored marker prevents the mid-release commit races that forced full CHANGELOG re-reconciliation in v3.8.40/v3.8.41 (a parallel campaign advanced `release/vX.Y.Z` by 34 commits mid-run). The release captain's own reconciliation/cycle-open pushes are exempt — they _are_ the release. Fixes that must land during a freeze (a homologation finding) follow the post-merge read-only rule: land on `main` first via `fix/release-vX.Y.Z-*`. **⛔ ONLY `/generate-release` may raise a release-freeze, and ONLY at its Phase 0a (start of generating a new version) — lifted at Phase 12c after the squash-merge to `main`.** No campaign, session, or agent may open a `release-freeze` marker at any other time — a freeze is **never** a mid-development coordination tool. If a session ever believes a freeze is genuinely, unavoidably necessary outside the `/generate-release` flow, it **MUST first ask the operator (`diegosouzapw`) in chat, explicitly alert "estou criando um freeze" and get an explicit yes** — never open, extend, or re-open a `release-freeze` autonomously. Conversely, do **not** close/lift an active `/generate-release` freeze to unblock campaign merges: it protects the captain's single clean CI run and auto-lifts at Phase 12c — closing it early re-triggers the exact commit race it prevents. Verify a freeze is legitimate before acting on it: an open `release-freeze` whose title/body references an **OPEN** release PR (`gh pr view --json state`) is the authorized captain freeze — hold, don't touch. --- diff --git a/README.md b/README.md index 4f6d5abb86..8a3b54f398 100644 --- a/README.md +++ b/README.md @@ -6,7 +6,7 @@ # 🚀 OmniRoute — The Free AI Gateway -### Never stop coding. Connect every AI tool to **236 providers** — **50+ free** — through one endpoint. +### Never stop coding. Connect every AI tool to **237 providers** — **90+ free** — through one endpoint. **Plug Claude Code, Codex, Cursor, Cline, Copilot & Antigravity into FREE Claude / GPT / Gemini. Auto-fallback.**
@@ -19,8 +19,15 @@
-[![231 AI Providers](https://img.shields.io/badge/231-AI_Providers-6C5CE7?style=for-the-badge)](#-231-ai-providers--50-free) -[![50+ Free](https://img.shields.io/badge/50%2B-Free_Tiers-00B894?style=for-the-badge)](#-231-ai-providers--50-free) +

+ +⭐ Star the repo if OMNIROUTE helped you save money and make your work easier. [![Stars](https://img.shields.io/github/stars/diegosouzapw/OmniRoute?style=social)](https://github.com/diegosouzapw/OmniRoute) +

+ +diegosouzapw%2FOmniRoute | Trendshift + +[![237 AI Providers](https://img.shields.io/badge/237-AI_Providers-6C5CE7?style=for-the-badge)](#-237-ai-providers--90-free) +[![90+ Free](https://img.shields.io/badge/90%2B-Free_Tiers-00B894?style=for-the-badge)](#-237-ai-providers--90-free) [![1.6B Free Tokens/mo](https://img.shields.io/badge/1.6B-Free_Tokens%2Fmo-00B894?style=for-the-badge)](docs/reference/FREE_TIERS.md) [![Token Savings](https://img.shields.io/badge/up_to_95%25-Token_Savings-E17055?style=for-the-badge)](#%EF%B8%8F-save-1595-tokens--automatically) [![17 Strategies](https://img.shields.io/badge/17-Routing_Strategies-0984E3?style=for-the-badge)](#-combos--the-flagship) @@ -34,79 +41,78 @@ [![Telegram](https://img.shields.io/badge/Telegram-26A5E4?style=for-the-badge&logo=telegram&logoColor=white)](https://t.me/omnirouteOficial) [![WhatsApp Global](https://img.shields.io/badge/WhatsApp_Global-25D366?style=for-the-badge&logo=whatsapp&logoColor=white)](https://chat.whatsapp.com/JI7cDQ1GyaiDHhVBpLxf8b?mode=gi_t) [![WhatsApp Brasil](https://img.shields.io/badge/WhatsApp_Brasil-25D366?style=for-the-badge&logo=whatsapp&logoColor=white)](https://chat.whatsapp.com/BTGJXIyjeNIIgExvTMGGhI) +[![Website](https://img.shields.io/badge/Website-omniroute.online-blue?logo=google-chrome&logoColor=white)](https://omniroute.online) **Questions, provider tips, roadmap & support → [Discord](https://discord.gg/EkzRkpzKYt) · [Telegram](https://t.me/omnirouteOficial) · WhatsApp [🌍 Global](https://chat.whatsapp.com/JI7cDQ1GyaiDHhVBpLxf8b?mode=gi_t) / [🇧🇷 Brasil](https://chat.whatsapp.com/BTGJXIyjeNIIgExvTMGGhI)**
-diegosouzapw%2FOmniRoute | Trendshift - -[![npm](https://img.shields.io/npm/v/omniroute?logo=npm&style=flat-square)](https://www.npmjs.com/package/omniroute) -[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg?style=flat-square)](LICENSE) -[![Node](https://img.shields.io/badge/node-%E2%89%A522.0.0-brightgreen?style=flat-square)](package.json) -[![Stars](https://img.shields.io/github/stars/diegosouzapw/OmniRoute?style=social)](https://github.com/diegosouzapw/OmniRoute) - -
+### 🧩 Available [![npm version](https://img.shields.io/npm/v/omniroute?color=cb3837&logo=npm)](https://www.npmjs.com/package/omniroute) ![NPM Monthly](https://img.shields.io/npm/dm/omniroute?label=npm/month&color=cb3837&logo=npm) [![Docker Hub](https://img.shields.io/docker/v/diegosouzapw/omniroute?label=Docker%20Hub&logo=docker&color=2496ED)](https://hub.docker.com/r/diegosouzapw/omniroute) +[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg?style=flat-square)](LICENSE) ![Docker Pulls](https://img.shields.io/docker/pulls/diegosouzapw/omniroute?label=docker%20pulls&logo=docker&color=2496ED) ![Electron Downloads](https://img.shields.io/github/downloads/diegosouzapw/omniroute/total?style=flat&label=electron%20downloads&logo=electron&color=47848F) -[![Website](https://img.shields.io/badge/Website-omniroute.online-blue?logo=google-chrome&logoColor=white)](https://omniroute.online) -
- -
- -[**🚀 Quick Start**](#-quick-start) • [**🎯 Combos**](#-combos--the-flagship) • [**🌐 Providers**](#-231-ai-providers--50-free) • [**🔌 CLI & MCP**](#-full-cli--a2a--mcp) • [**🗜️ Compression**](#%EF%B8%8F-save-1595-tokens--automatically) • [**🌍 Website**](https://omniroute.online) +[**🚀 Quick Start**](#-quick-start) • [**🎯 Combos**](#-combos--the-flagship) • [**🌐 Providers**](#-237-ai-providers--90-free) • [**🔌 CLI & MCP**](#-full-cli--a2a--mcp) • [**🗜️ Compression**](#%EF%B8%8F-save-1595-tokens--automatically) • [**🌍 Website**](https://omniroute.online) [💥 The Promise](#-the-promise) • [🤔 Why](#-why-omniroute) • [🏆 What Sets Apart](#-what-sets-omniroute-apart) • [🤖 Compatible CLIs](#-compatible-clis--coding-agents) • [🖥️ Where It Runs](#%EF%B8%8F-where-omniroute-runs--anywhere) • [🔒 Private](#-private--local-first) • [🎬 In Action](#-omniroute-in-action) • [📚 Explore More](#-explore-more) • [📧 Support](#-support--community) @@ -138,18 +144,18 @@ -> One endpoint. **236 providers.** Never stop building — and let OmniRoute pick the cheapest one that works. +> One endpoint. **237 providers.** Never stop building — and let OmniRoute pick the cheapest one that works. - + - + - + - +
🚫 Never hit limits
Auto-fallback across 236 providers in milliseconds. Quota out? Next provider takes over — zero downtime.		🚫 Never hit limits
Auto-fallback across 237 providers in milliseconds. Quota out? Next provider takes over — zero downtime.		 💸 Save up to 95% tokens
RTK + Caveman stacked compression cuts 15–95% of eligible tokens (~89% avg on tool-heavy sessions).		🆓 $0 to start
50+ providers with a free tier, 11 free forever (Kiro, Qoder, Pollinations, LongCat…). No card needed.		🆓 $0 to start
90+ providers with a free tier, 11 free forever (Kiro, Qoder, Pollinations, LongCat…). No card needed.
🔌 Every tool works
16+ coding agents — Claude Code, Codex, Cursor, Cline, Copilot, Antigravity — through one config.		🔌 Every tool works
24+ coding agents — Claude Code, Codex, Cursor, Cline, Copilot, Antigravity — through one config.		 🧩 One endpoint
OpenAI ↔ Claude ↔ Gemini ↔ Responses API translation. Point any tool at /v1 and it just works.		🛡️ Production-grade
Circuit breakers, TLS stealth, MCP (87 tools), A2A, memory, guardrails, evals. 14,965 tests.		🛡️ Production-grade
Circuit breakers, TLS stealth, MCP (95 tools), A2A, memory, guardrails, evals. 21,000+ tests.
@@ -223,21 +229,56 @@ No combo to create. Set your model to `auto` (or a variant) and OmniRoute builds ### 🔀 Or build your own — 17 routing strategies -| Goal | Strategy / combo | -| --------------------------------------- | -------------------------------------------------- | -| 🥇 Drain my subscription before paying | `priority` / `fill-first` | -| ⚖️ Spread load across accounts | `round-robin` · `weighted` · `p2c` · `least-used` | -| 💸 Always cheapest viable model | `cost-optimized` · `auto/cheap` | -| 🧠 Hand off long context between models | `context-relay` · `context-optimized` | -| 🎲 Randomized / privacy routing | `random` · `strict-random` | -| 🧬 Fan out to a panel + judge synthesis | `fusion` | -| 📊 Route by remaining quota headroom | `reset-window` · `headroom` | -| 🤖 Just make it smart | `auto` (9-factor scoring) · `lkgp` · `reset-aware` | +All **17** strategies — mix & match per combo step: + +| # | Strategy | What it does | +| --- | ------------------- | ---------------------------------------------------------------- | +| 1 | `priority` | First-target ordered list — drain each before the next 🥇 | +| 2 | `fill-first` | Fill each target's quota fully before moving on | +| 3 | `weighted` | Weighted random by per-target weight | +| 4 | `round-robin` | Cycle through targets in order | +| 5 | `p2c` | Power-of-two-choices random load balancing | +| 6 | `least-used` | Pick the target with the lowest current load | +| 7 | `random` | Uniform random pick (deduplicated) | +| 8 | `strict-random` | Random without de-duplicating repeats 🎲 | +| 9 | `cost-optimized` | Minimize $ per request from live catalog pricing 💸 | +| 10 | `headroom` | Pick the target with the most remaining quota | +| 11 | `reset-window` | Prefer the target whose quota window resets soonest | +| 12 | `reset-aware` | Rank by quota reset time — short windows first 📊 | +| 13 | `context-relay` | Hand off context across targets for long conversations 🧠 | +| 14 | `context-optimized` | Pick the best fit for the current context size | +| 15 | `lkgp` | Last-Known-Good Path — sticky to the last successful target | +| 16 | `auto` | 9-factor live scoring across every connection 🤖 | +| 17 | `fusion` | Fan out to a panel of models + a judge synthesizes one answer 🧬 | The Auto-Combo engine scores every candidate on **9 factors** (health, quota, cost, latency, success rate, freshness…) — see [`docs/routing/AUTO-COMBO.md`](docs/routing/AUTO-COMBO.md). ## +### ⚖️ Quota-Share — split one subscription across a team ✨ NEW + +> Running several keys against the **same upstream account** (one Codex Pro plan, one Kimi key, one GLM Coding seat)? A burst on one key can burn the whole 5-hour / hourly quota and lock everyone else out. **Quota-Share** distributes a provider's time-based quota **fairly** across the keys in a pool — and it's _work-conserving_, so an idle member's slice is lent out instead of wasted. + +| Knob | What it controls | +| ------------------------ | ------------------------------------------------------------------------------- | +| ⚖️ **Allocation weight** | each key's slice of the pool — e.g. `50 / 30 / 20` | +| 📐 **Dimensions** | track `%` · requests · tokens · `$`, per **5h / 7d / per-model** window | +| 🚦 **Policy** | `hard` (block over share) · `soft` (deprioritize) · `burst` (use idle headroom) | +| 🧱 **Cap** | absolute ceiling per key, independent of mode | + +``` +Pool "team-codex" · 1 Codex Pro account · 3 keys · 5-hour window + ├─ alice weight 50 ██████████░░░░░░░░░░ ≤ 50% of the shared 5h quota + ├─ bob weight 30 ██████░░░░░░░░░░░░░░ ≤ 30% + └─ ci-bot weight 20 ████░░░░░░░░░░░░░░░░ ≤ 20% +Generous mode (<50% pool used) → idle shares are lent out +Strict mode (≥50% pool used) → each key held to its fair share +``` + +Enforced in the hot path **before** the request leaves OmniRoute, with per-(key, model) caps + session stickiness for prompt-cache integrity. 📖 [Quota Sharing Engine](docs/routing/QUOTA_SHARE.md) + +## + ### 🧱 Resilience is built in (3 independent layers) | Layer | Scope | What it does | @@ -267,15 +308,15 @@ Result: 4 layers of fallback = zero downtime | Feature | OmniRoute | Other routers | | -------------------------------------- | ------------------------------------------------------------------- | ------------- | -| 🌐 Providers | **231** | 20–100 | -| 🆓 Free providers | **50+ (11 free forever)** | 1–5 | +| 🌐 Providers | **237** | 20–100 | +| 🆓 Free providers | **90+ (11 free forever)** | 1–5 | | 🔀 Routing strategies | **17** (priority, weighted, cost-optimized, context-relay, fusion…) | 1–3 | | 🗜️ Token compression | **RTK + Caveman stacked (15–95%)** | None / 20–40% | -| 🧰 Built-in MCP server | **87 tools, 3 transports, 30 scopes** | Rare | +| 🧰 Built-in MCP server | **95 tools, 3 transports, 30 scopes** | Rare | | 🤝 A2A agent protocol | **6 skills, JSON-RPC 2.0** | None | | 🧠 Memory (FTS5 + vector) | **Yes** | Rare | | 🛡️ Guardrails (PII, injection, vision) | **Yes** | Rare | -| ☁️ Cloud agents | **Codex, Devin, Jules** | None | +| ☁️ Cloud agents | **Codex, Cursor, Devin, Jules** | None | | 🥷 TLS fingerprint stealth | **JA3/JA4 via wreq-js** | None | | 🖥️ Multi-platform | **Web · Desktop · Termux · PWA** | Web only | | 🌍 i18n | **42 locales** | 0–4 | @@ -290,18 +331,20 @@ Result: 4 layers of fallback = zero downtime -> Recent highlights from **v3.8.20 → v3.8.41**. Full history in [`CHANGELOG.md`](CHANGELOG.md). +> Recent highlights from **v3.8.20 → v3.8.43**. Full history in [`CHANGELOG.md`](CHANGELOG.md). +- **🗜️ Compression hardening** — a default-on **inflation guard** (discard the stacked result and send the verbatim original whenever compression would _grow_ the prompt), completed **Caveman rule packs** for German / French / Japanese (dedup + ultra) plus a new **Chinese (文言 / wényán) input pack** with zh-vs-ja auto-detection, and **RTK filters for Gradle & .NET (`dotnet`)** build output. → [Compression](docs/compression/COMPRESSION_ENGINES.md) +- **💸 Honest flat-rate cost** — subscription / coding-plan providers (ChatGPT Web, grok-web, the Minimax / Kimi / GLM / Alibaba Coding plans, Xiaomi MiMo…) now read **$0** in cost analytics instead of an inflated per-token estimate, while budget / quota / routing keep estimating unchanged. → [API Reference](docs/reference/API_REFERENCE.md) - **⚖️ Quota-Share routing** — a dedicated combo strategy that spreads load across accounts by _available quota_: Deficit-Round-Robin scheduling, per-connection `max_concurrent` with cooldown-wait queueing, multi-window usage buckets (5h / 7d / per-model), per-(key, model) caps, session stickiness for prompt-cache integrity, and proactive saturation from upstream token-usage headers. → [Resilience Guide](docs/architecture/RESILIENCE_GUIDE.md) - **🤖 One-command CLI/agent setup** — a dedicated `setup-*` command configures each coding tool to route through OmniRoute (Claude Code, Codex, Cline, Continue, Cursor, Roo Code, Kilo Code, Crush, Goose, Qwen Code, Aider, OpenCode); `omniroute launch` / `omniroute launch-codex` are zero-config launchers. → [CLI Integrations](docs/guides/CLI-INTEGRATIONS.md) - **🛰️ Remote mode** — drive a remote OmniRoute from any machine with scoped access tokens (`omniroute connect` / `omniroute contexts` / `omniroute tokens`), plus an `omniroute login antigravity` helper that runs Google "native/desktop" OAuth on your own machine and pastes a credential blob into a remote/VPS install (where the loopback redirect is unreachable). → [Remote Mode](docs/guides/REMOTE-MODE.md) - **🧭 Smarter auto-routing** — OpenRouter-style `auto/:` combos (e.g. `auto/coding:fast`, `auto/reasoning:pro`), a **Fusion** strategy (fan out to a panel of models in parallel, then synthesize via a judge), **task-aware routing** (best-fit connection per task type), per-request `X-Route-Model` override, live Arena-ELO + models.dev model intelligence, per-step account allowlists, provider-wildcard combo steps, nested combo-ref execution, sticky weighted selection, and `web_search`-aware routing. → [Auto-Combo](docs/routing/AUTO-COMBO.md) -- **🗜️ Pluggable compression** — an async pipeline of **9 composable engines** with Compression Studios, an LLMLingua-2 ONNX engine and a heuristic/SLM two-tier **Ultra**, RTK, delegated Anthropic Context Editing, **Output Styles** (output-axis steering: terse-prose / less-code / terse-CJK), an **adaptive context-budget dial** (escalate only as far as needed to fit the context window), per-request `x-omniroute-compression` control, an opt-in offline eval harness, one-click **Headroom** proxy lifecycle management from the dashboard (Docker sidecar supported), a synthetic **compression playground** (Play lanes + A/B Compare with USD-capped fidelity verdicts), an opt-in **per-step fidelity gate** that rejects a lossy engine before it degrades the prompt, a **best-of-N candidate encoder** (GCF vs TOON — keep whichever is shorter, with an A/B bytes/token table in the studio), **CCR ranged/grep/stats retrieval** (pull an exact byte/line slice or summary of a stored block instead of re-expanding it), and a unified panel with named profiles + an active-profile selector. → [Compression](docs/compression/COMPRESSION_ENGINES.md) +- **🗜️ Pluggable compression** — an async pipeline of **10 composable engines** with Compression Studios, an LLMLingua-2 ONNX engine and a heuristic/SLM two-tier **Ultra**, RTK, delegated Anthropic Context Editing, **Output Styles** (output-axis steering: terse-prose / less-code / terse-CJK), an **adaptive context-budget dial** (escalate only as far as needed to fit the context window), per-request `x-omniroute-compression` control, an opt-in offline eval harness, one-click **Headroom** proxy lifecycle management from the dashboard (Docker sidecar supported), a synthetic **compression playground** (Play lanes + A/B Compare with USD-capped fidelity verdicts), an opt-in **per-step fidelity gate** that rejects a lossy engine before it degrades the prompt, a **best-of-N candidate encoder** (GCF vs TOON — keep whichever is shorter, with an A/B bytes/token table in the studio), **CCR ranged/grep/stats retrieval** (pull an exact byte/line slice or summary of a stored block instead of re-expanding it), a unified panel with named profiles + an active-profile selector, an opt-in **per-engine pipeline circuit-breaker**, an opt-in **LLM-tier engine** (a model pass for higher-ratio semantic compression), a **read-lifecycle engine** that collapses superseded file reads, **usage-observed prefix freeze**, a graduated **CCR retrieval-feedback ramp**, a `preserveSystemPrompt` mode enum, and a **drag-reorder pipeline editor** in the studio. → [Compression](docs/compression/COMPRESSION_ENGINES.md) - **🕵️ Transparent MITM decrypt (TPROXY)** — capture & translate traffic from CLIs that ignore proxy env vars, with a per-SNI certificate authority and a trust-store installer. → [MITM/TPROXY](docs/security/MITM-TPROXY-DECRYPT.md) - **💸 Cost telemetry everywhere** — `X-OmniRoute-*` cost/usage headers on every endpoint (including media), a non-token cost engine, a cache-HIT `X-OmniRoute-Cost-Saved` header, and per-key USD spend quotas. → [API Reference](docs/reference/API_REFERENCE.md) -- **🧠 Memory you control** — opt-in int8 vector quantization (Qdrant + sqlite-vec), memory off by default, and a per-request `x-omniroute-no-memory` header. → [Memory](docs/frameworks/MEMORY.md) +- **🧠 Memory you control** — opt-in int8 vector quantization (Qdrant + sqlite-vec), opt-in **typed memory decay** (aged low-value memories fade on a per-type schedule), memory off by default, and a per-request `x-omniroute-no-memory` header. → [Memory](docs/frameworks/MEMORY.md) - **🛡️ Security** — a prompt-injection guard across every LLM route (backed by a red-team suite), plus a free DuckDuckGo last-resort web search. → [Guardrails](docs/security/GUARDRAILS.md) -- **🤝 More providers & agents** — Cursor Cloud Agent (a 4th cloud agent), CodeBuddy CN (`copilot.tencent.com`), a Google Flow video-generation provider, new gateways **DGrid** and **Pioneer AI** (Fastino Labs), inbound **xAI Grok** translators plus **Grok Build (xAI)** with an OAuth import-token flow, GPT-4 / GPT-4o-mini on the GitHub Copilot provider, multi-model **Factory Droid**, **ZenMux Free** (session-cookie free tier), **Alibaba DashScope** text-to-video (`wan2.7-t2v`), a refreshed 236-provider catalog (OrcaRouter, Wafer AI, OpenAdapter, dit.ai, TokenRouter, …), Vertex AI media generation (speech/transcription/music/video), and one-click account import from CLIProxyAPI (`~/.cli-proxy-api/`). → [Providers](docs/reference/PROVIDER_REFERENCE.md) +- **🤝 More providers & agents** — Cursor Cloud Agent (a 4th cloud agent), CodeBuddy CN (`copilot.tencent.com`), a Google Flow video-generation provider, new gateways **DGrid** and **Pioneer AI** (Fastino Labs), inbound **xAI Grok** translators plus **Grok Build (xAI)** with an OAuth import-token flow, GPT-4 / GPT-4o-mini on the GitHub Copilot provider, multi-model **Factory Droid**, **ZenMux Free** (session-cookie free tier), **Alibaba DashScope** text-to-video (`wan2.7-t2v`), a refreshed 237-provider catalog (OrcaRouter, Wafer AI, OpenAdapter, dit.ai, TokenRouter, …), Vertex AI media generation (speech/transcription/music/video), a first-class **Ollama** local-provider card, the **SenseNova** free Token Plan (chat + text-to-image), and one-click account import from CLIProxyAPI (`~/.cli-proxy-api/`). → [Providers](docs/reference/PROVIDER_REFERENCE.md) - **⚡ Local performance & infra** — a one-click local Redis launcher (`omniroute redis up`, plus a dashboard Redis panel), one-click **Cloudflare Workers** and **Deno Deploy** relay deployers wired into the proxy pool, and an optional Bifrost Go sidecar that offloads the hottest relay path (`BIFROST_BASE_URL`, with automatic fallback to the TypeScript path on timeout) — now with a relay-backend selector (`OMNIROUTE_RELAY_BACKEND=ts|bifrost|auto`) so the `/v1/relay` endpoint stays the stable surface while choosing the fastest backend internally. → [Environment](docs/reference/ENVIRONMENT.md)
@@ -336,7 +379,7 @@ Result: 4 layers of fallback = zero downtime ＋ also works with · Cline · Antigravity · Windsurf · AMP · Hermes · Qwen CLI · Roo · Continue · any OpenAI-compatible tool -📖 Per-tool setup for all 16+ tools → [`docs/reference/CLI-TOOLS.md`](docs/reference/CLI-TOOLS.md) · 🧩 OpenCode plugin → [`@omniroute/opencode-provider`](https://www.npmjs.com/package/@omniroute/opencode-provider) +📖 Per-tool setup for all 24+ tools → [`docs/reference/CLI-TOOLS.md`](docs/reference/CLI-TOOLS.md) · 🧩 OpenCode plugin → [`@omniroute/opencode-provider`](https://www.npmjs.com/package/@omniroute/opencode-provider) @@ -344,27 +387,60 @@ Result: 4 layers of fallback = zero downtime
-# 🌐 231 AI Providers — 50+ Free +# 🌐 237 AI Providers — 90+ Free
-> The most complete catalog of any open-source router: **236 providers**, **50+ with a free tier**, **11 free forever**. +> The most complete catalog of any open-source router: **237 providers**, **90+ with a free tier**, **11 free forever**.
+### 🏢 Every major lab — through one endpoint + + + + + + + + + + + + + + + + + + + + + + + + + + +
OpenAI
OpenAI		Anthropic
Anthropic		Gemini
Gemini		xAI Grok
xAI Grok		DeepSeek
DeepSeek		Mistral
Mistral
Qwen
Qwen		Meta Llama
Meta Llama		Groq
Groq		NVIDIA
NVIDIA		MiniMax
MiniMax		Cohere
Cohere
Perplexity
Perplexity		Hugging Face
HuggingFace		Together
Together		Fireworks
Fireworks		Cloudflare
Cloudflare		Baidu
Baidu
+ +…and 220+ more — every icon resolves live from the dashboard's provider catalog. 📖 [Provider Reference](docs/reference/PROVIDER_REFERENCE.md) + +
+ ### 🆓 Free Forever — $0, no card - - - - + + + + - - - + + +
AgentRouter
GPT-5, Claude, Gemini
$100 free credits		Qoder AI
Kimi-K2, DeepSeek-R1
Unlimited FREE		Pollinations
GPT-5, Claude, Llama 4
No key needed		LongCat
LongCat-2.0
10M tokens one-time (KYC) 🔑		AgentRouter
AgentRouter
GPT-5, Claude, Gemini
$100 free credits		Qoder AI
Qoder AI
Kimi-K2, DeepSeek-R1
Unlimited FREE		Pollinations
Pollinations
GPT-5, Claude, Llama 4
No key needed		LongCat
LongCat
LongCat-2.0
10M tokens one-time (KYC) 🔑
Cloudflare AI
50+ models
10K neurons/day		NVIDIA NIM
129 models
~40 RPM free		Cerebras
Qwen3 235B
1M tokens/day		Cloudflare AI
Cloudflare AI
50+ models
10K neurons/day		NVIDIA NIM
NVIDIA NIM
129 models
~40 RPM free		Cerebras
Cerebras
Qwen3 235B
1M tokens/day
@@ -420,7 +496,7 @@ Result: 4 layers of fallback = zero downtime
-> OmniRoute isn't just a server — it's a **full command-line cockpit** with **60+ commands**, plus open agent protocols so an AI agent can drive OmniRoute **by itself**. +> OmniRoute isn't just a server — it's a **full command-line cockpit** with **80+ commands**, plus open agent protocols so an AI agent can drive OmniRoute **by itself**. ### ⌨️ A real CLI (not just `start`) @@ -460,7 +536,7 @@ Expose OmniRoute over **MCP** or **A2A** and any capable agent gets the keys to | Protocol | Endpoint | Use it for | | ------------------ | ----------------------------------------------- | ------------------------------------------------------ | | 🧰 **MCP (stdio)** | `omniroute --mcp` | Plug into Claude Desktop, Cursor, any MCP client | -| 🌊 **MCP (HTTP)** | `http://localhost:20128/api/mcp/stream` | Remote MCP — **87 tools**, 30 scopes, full audit trail | +| 🌊 **MCP (HTTP)** | `http://localhost:20128/api/mcp/stream` | Remote MCP — **95 tools**, 30 scopes, full audit trail | | 📡 **MCP (SSE)** | `http://localhost:20128/api/mcp/sse` | Streaming MCP transport | | 🤝 **A2A** | `http://localhost:20128/.well-known/agent.json` | Agent-to-agent, **JSON-RPC 2.0** + SSE, 6 skills | @@ -479,9 +555,9 @@ claude mcp add-server omniroute --type http --url http://localhost:20128/api/mcp -> **Why use many tokens when few tokens do the trick?** Every request passes through OmniRoute's compression pipeline **transparently** — no client changes. It's now a **stack of 9 composable engines** that run in order and mix & match per routing combo — building on ideas from [RTK](https://github.com/rtk-ai/rtk), [Caveman](https://github.com/JuliusBrussee/caveman) (⭐ 51K+), [LLMLingua-2](https://github.com/microsoft/LLMLingua), and [Troglodita](https://github.com/leninejunior/troglodita) (PT-BR). +> **Why use many tokens when few tokens do the trick?** Every request passes through OmniRoute's compression pipeline **transparently** — no client changes. It's now a **stack of 10 composable engines** that run in order and mix & match per routing combo — building on ideas from [RTK](https://github.com/rtk-ai/rtk), [Caveman](https://github.com/JuliusBrussee/caveman) (⭐ 78K+), [LLMLingua-2](https://github.com/microsoft/LLMLingua), and [Troglodita](https://github.com/leninejunior/troglodita) (PT-BR). -### 🧱 The 9-engine stack +### 🧱 The 10-engine stack Engines run in pipeline order; each is independently toggleable and configurable per combo: @@ -491,11 +567,12 @@ Engines run in pipeline order; each is independently toggleable and configurable | 2 | **CCR** | Archives large blocks behind retrieve markers, fetched on demand | | 3 | **RTK** | Smart tool-result filtering, dedup & truncation (command-aware) | | 4 | **Headroom** | Lossless tabular compaction of homogeneous JSON arrays (~30%+) | -| 5 | **Caveman** | Rule-based prose compression (~65–75% on output) | -| 6 | **LLMLingua-2** | ML semantic pruning via MobileBERT ONNX — code-safe, async | -| 7 | **Lite** | Whitespace + image-URL trimming (latency-light baseline) | -| 8 | **Aggressive** | Summarization + progressive aging of old turns | -| 9 | **Ultra** | Heuristic token pruning with an optional small-model (SLM) tier | +| 5 | **Relevance** | Extractive sentence scoring against the last user query | +| 6 | **Caveman** | Rule-based prose compression (~65–75% on output) | +| 7 | **LLMLingua-2** | ML semantic pruning via MobileBERT ONNX — code-safe, async | +| 8 | **Lite** | Whitespace + image-URL trimming (latency-light baseline) | +| 9 | **Aggressive** | Summarization + progressive aging of old turns | +| 10 | **Ultra** | Heuristic token pruning with an optional small-model (SLM) tier | Code blocks, URLs and structured data are **always preserved** byte-perfect. **One-click presets** combine the engines: @@ -529,7 +606,7 @@ Code blocks, URLs and structured data are **always preserved** byte-perfect. **O ### 📖 How it works — pipeline, architecture & savings math ``` -Client (10,000 tok) ──▶ OmniRoute Compression (9 engines) ──▶ Provider (~1,080 tok, up to 95% saved) +Client (10,000 tok) ──▶ OmniRoute Compression (10 engines) ──▶ Provider (~1,080 tok, up to 95% saved) ``` Default stacked combo runs `RTK → Caveman`. When both act on the same tool/context payload, savings compound: @@ -544,7 +621,7 @@ Code blocks, URLs, JSON and structured data are **always protected** by the pres ### 🎚️ Beyond the engines — output styles, the adaptive dial & per-request control -The 9 engines above shrink what goes **in**. Three more layers shape **how**, **when**, and what comes **out**: +The 10 engines above shrink what goes **in**. Three more layers shape **how**, **when**, and what comes **out**: - **🪄 Output Styles** _(output-axis steering)_ — inject deterministic, cache-safe response-shaping instructions; combinable, each at `lite` / `full` / `ultra` intensity. Adding a style is a one-line registry entry: - **Terse prose** — drop filler / articles / hedging; keep technical substance exact. @@ -628,7 +705,7 @@ PORT=20128 npm run dev **📦 pnpm** ```bash -pnpm install -g omniroute && pnpm approve-builds -g && omniroute +pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core && omniroute ``` **🐧 Arch Linux (AUR)** @@ -666,6 +743,24 @@ podman compose --profile base up -d 📖 [Podman Guide](contrib/podman/README.md) — Quadlet setup, podman-compose, Quadlet. +**⚡ Faster / leaner install (skip the native build)** + +The native SQLite engine (`better-sqlite3`) is an **optional** dependency, so a global +install never blocks on compiling from source: it uses a prebuilt binary when one matches +your platform/Node, and otherwise falls back transparently to a pure-JS engine +(`node:sqlite` on Node 22+, else the bundled `sql.js` WASM) — no build tools required. + +To skip the post-install native warm-up entirely (CI, headless, or slow machines): + +```bash +OMNIROUTE_SKIP_POSTINSTALL=1 npm install -g omniroute # CI=1 also skips it +``` + +For the fastest installs prefer **pnpm** (content-addressed store + hard links — see above). +For a dashboard-free, headless runtime use the Docker `base` profile (above) or the +[Termux guide](docs/guides/TERMUX_GUIDE.md). The CLI and the web dashboard are served by the +same process on one port, so there is no separate CLI-only package today. +
@@ -720,16 +815,16 @@ podman compose --profile base up -d **The $0 Free Stack — combine into one unbreakable combo:** -| Provider | Prefix | Free models | Quota | -| ----------------- | ----------- | ----------------------------------------------- | ----------------- | -| **Kiro** | `kr/` | Claude Sonnet 4.5, Haiku 4.5, Opus 4.6 | 50 credits/mo | -| **Qoder** | `if/` | kimi-k2-thinking, qwen3-coder-plus, deepseek-r1 | ♾️ Unlimited | -| **Qwen** | `qw/` | qwen3-coder-plus/flash/next | ♾️ Unlimited | -| **Pollinations** | `pol/` | GPT-5, Claude, Gemini, DeepSeek, Llama 4 | No key needed | +| Provider | Prefix | Free models | Quota | +| ----------------- | ----------- | ----------------------------------------------- | ------------------ | +| **Kiro** | `kr/` | Claude Sonnet 4.5, Haiku 4.5, Opus 4.6 | 50 credits/mo | +| **Qoder** | `if/` | kimi-k2-thinking, qwen3-coder-plus, deepseek-r1 | ♾️ Unlimited | +| **Qwen** | `qw/` | qwen3-coder-plus/flash/next | ♾️ Unlimited | +| **Pollinations** | `pol/` | GPT-5, Claude, Gemini, DeepSeek, Llama 4 | No key needed | | **LongCat** | `lc/` | LongCat-2.0 | 10M one-time (KYC) | -| **Cloudflare AI** | `cf/` | 50+ models | 10K neurons/day | -| **NVIDIA NIM** | `nvidia/` | 129 models | ~40 RPM | -| **Cerebras** | `cerebras/` | Qwen3 235B, GPT-OSS 120B | 1M tok/day | +| **Cloudflare AI** | `cf/` | 50+ models | 10K neurons/day | +| **NVIDIA NIM** | `nvidia/` | 129 models | ~40 RPM | +| **Cerebras** | `cerebras/` | Qwen3 235B, GPT-OSS 120B | 1M tok/day | > 💡 The dashboard "cost" is a **savings tracker**, not a bill — OmniRoute never charges you. A "$290 total cost" using free models means **$290 saved**. @@ -778,9 +873,9 @@ Compression: aggressive (~50%) → double your free quota · Cost: $0/mo
-**Routing:** 15 strategies · task-aware smart routing · thinking budget controls · wildcard routing · system prompt injection. +**Routing:** 17 strategies · task-aware smart routing · thinking budget controls · wildcard routing · system prompt injection. **Compatibility:** OpenAI ↔ Claude ↔ Gemini ↔ Responses API · auto OAuth refresh (PKCE, 8 providers) · multi-account round-robin · Batch + Files API · live OpenAPI 3.0. -**Protocols:** MCP (87 tools, 3 transports, 30 scopes) · A2A (JSON-RPC 2.0, SSE, 6 skills) · ACP · cloud agents (Codex, Devin, Jules). +**Protocols:** MCP (95 tools, 3 transports, 30 scopes) · A2A (JSON-RPC 2.0, SSE, 6 skills) · ACP · cloud agents (Codex, Cursor, Devin, Jules). **Plugins:** custom plugin marketplace (system-configured registry URL with SSRF-guarded fetch) · install / enable / disable · Notion + Obsidian knowledge-base integrations (WebDAV file server, vault search, note CRUD). **Embedded services:** one-click install & lifecycle management of local sidecar services (CLIProxy, NineRouter). **Quality & Ops:** built-in **Evals** (golden-set: exact/contains/regex/custom) · guardrails (PII, injection, vision) · health dashboard · p50/p95/p99 telemetry · webhooks · compliance audit. @@ -804,7 +899,7 @@ Compression: aggressive (~50%) → double your free quota · Cost: $0/mo **Will I be charged by OmniRoute?** No — it's free, open-source software on your machine. You only pay paid providers directly. OmniRoute has no billing system. **Are FREE providers really unlimited?** Mostly — Qoder, Pollinations, LongCat, and Cloudflare are free with no per-account credit cap. Kiro is free too but capped at ~50 credits/month per account. Stack multiple free providers in a combo and auto-fallback keeps you serving for $0. **Will compression hurt quality?** No — it only compresses the **input**; code, URLs, JSON are always protected. -**Does it work where AI is blocked?** Yes — 3-level proxy + 1proxy marketplace reach all 236 providers. +**Does it work where AI is blocked?** Yes — 3-level proxy + 1proxy marketplace reach all 237 providers. 📖 [User Guide](docs/guides/USER_GUIDE.md) · [API Reference](docs/reference/API_REFERENCE.md) · [Environment Config](docs/reference/ENVIRONMENT.md) @@ -874,7 +969,7 @@ Compression: aggressive (~50%) → double your free quota · Cost: $0/mo - **Protocols**: MCP (stdio/HTTP) + A2A v0.3 (JSON-RPC 2.0 + SSE) - **Streaming**: Server-Sent Events (SSE) + WebSocket bridge (`/v1/ws`) - **Auth**: OAuth 2.0 (PKCE) + JWT + API Keys + MCP Scoped Authorization -- **Testing**: Node.js test runner + Vitest (**14,965 test cases** across 517 files — unit, integration, E2E, security, ecosystem) +- **Testing**: Node.js test runner + Vitest (**21,000+ test cases** across 2,586 files — unit, integration, E2E, security, ecosystem) - **Platforms**: Desktop (Electron), Android (Termux), PWA (any browser) - **CI/CD**: GitHub Actions (auto npm publish + Docker Hub on release) - **Website**: [omniroute.online](https://omniroute.online) @@ -937,7 +1032,7 @@ Compression: aggressive (~50%) → double your free quota · Cost: $0/mo | ------------------------------------------------- | --------------------------------------------------- | | [API Reference](docs/reference/API_REFERENCE.md) | All endpoints with examples | | [OpenAPI Spec](docs/openapi.yaml) | OpenAPI 3.0 specification | -| [MCP Server](open-sse/mcp-server/README.md) | 87 MCP tools, IDE configs, Python/TS/Go clients | +| [MCP Server](open-sse/mcp-server/README.md) | 95 MCP tools, IDE configs, Python/TS/Go clients | | [MCP Server Guide](docs/frameworks/MCP-SERVER.md) | MCP installation, transports, and tool reference | | [A2A Server](src/lib/a2a/README.md) | JSON-RPC 2.0 protocol, skills, streaming, task mgmt | | [A2A Server Guide](docs/frameworks/A2A-SERVER.md) | A2A agent card, tasks, skills, and streaming | @@ -951,7 +1046,7 @@ Compression: aggressive (~50%) → double your free quota · Cost: $0/mo | [Security Policy](SECURITY.md) | Vulnerability reporting and security practices | | [i18n Guide](docs/guides/I18N.md) | 40+ language support, translation workflow, RTL | | [Release Checklist](docs/ops/RELEASE_CHECKLIST.md) | Pre-release validation steps | -| [Coverage Plan](docs/ops/COVERAGE_PLAN.md) | Test coverage strategy and 14,965 test suite | +| [Coverage Plan](docs/ops/COVERAGE_PLAN.md) | Test coverage strategy and 21,000+ test suite |
@@ -965,23 +1060,23 @@ Compression: aggressive (~50%) → double your free quota · Cost: $0/mo - oyi77
+ oyi77
oyi77
- 🥇 190 commits • +72K lines
+ 🥇 189 commits • +155K lines
Analytics engine, SQL aggregations,
proxy marketplace, test coverage - Chris Staley
+ Chris Staley
Chris Staley
- 🥈 72 commits • +5.7K lines
+ 🥈 70 commits • +5.7K lines
SSE stream hardening, Responses API,
Gemini pagination, test regression fixes - zenobit
+ zenobit
zenobit
🥉 62 commits • +24K lines
@@ -989,20 +1084,28 @@ Compression: aggressive (~50%) → double your free quota · Cost: $0/mo - R.D. & Randi
+ R.D. & Randi
R.D. & Randi
- 🏅 107 commits • +28K lines
+ 🏅 108 commits • +30K lines
Endpoints page, tunnel integrations,
Docker workflows, A2A status, compression UI - benzntech
+ benzntech
benzntech
- 🏅 20 commits • +7.5K lines
+ 🏅 22 commits • +7.5K lines
Electron desktop app, auto-updater,
release build workflows, cross-platform CI + + + herjarsa
+ herjarsa +
+ 🏅 21 commits • +6K lines
+ Zero-latency combos, vision-bridge auto-routing,
catalog context-length, resilience 429 hints + @@ -1016,11 +1119,11 @@ Compression: aggressive (~50%) → double your free quota · Cost: $0/mo
-## 👥 Contributors +## 👥 280+ Contributors
-[![Contributors](https://contrib.rocks/image?repo=diegosouzapw/OmniRoute&max=100&columns=20&anon=1)](https://github.com/diegosouzapw/OmniRoute/graphs/contributors) +[![Contributors](https://contrib.rocks/image?repo=diegosouzapw/OmniRoute&max=200&columns=20&anon=1)](https://github.com/diegosouzapw/OmniRoute/graphs/contributors) ### How to Contribute @@ -1085,36 +1188,36 @@ OmniRoute stands on the shoulders of giants. It started as a fork of **[9router] | Project | ⭐ | How it inspired OmniRoute | | ------------------------------------------------------------------------------- | ----: | ------------------------------------------------------------------------------------------------------------------------------------- | -| **[9router](https://github.com/decolua/9router)** · decolua | 17.9k | The original project this fork is built on — extended here with multi-modal APIs and a full TypeScript rewrite. | -| **[CLIProxyAPI](https://github.com/router-for-me/CLIProxyAPI)** · router-for-me | 37.8k | The Go implementation that inspired this JavaScript / TypeScript port. | -| **[LiteLLM](https://github.com/BerriAI/litellm)** · BerriAI | 50.8k | The AI gateway whose public pricing dataset feeds our cost-tracking sync and whose provider-normalization model informed our routing. | +| **[9router](https://github.com/decolua/9router)** · decolua | 19.0k | The original project this fork is built on — extended here with multi-modal APIs and a full TypeScript rewrite. | +| **[CLIProxyAPI](https://github.com/router-for-me/CLIProxyAPI)** · router-for-me | 38.8k | The Go implementation that inspired this JavaScript / TypeScript port. | +| **[LiteLLM](https://github.com/BerriAI/litellm)** · BerriAI | 52.1k | The AI gateway whose public pricing dataset feeds our cost-tracking sync and whose provider-normalization model informed our routing. | ### 🗜️ Context & token compression — engines -| Project | ⭐ | How it inspired OmniRoute | -| ---------------------------------------------------------------------------- | ----: | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -| **[Caveman](https://github.com/JuliusBrussee/caveman)** · JuliusBrussee | 74.5k | The viral "why use many token when few token do trick" project — its caveman-speak philosophy powers our standard compression mode and 30+ filler/condensation rules. | -| **[RTK – Rust Token Killer](https://github.com/rtk-ai/rtk)** · rtk-ai | 63.6k | High-performance command-output compression — inspired our RTK engine, JSON filter DSL, raw-output recovery and the stacked RTK → Caveman pipeline. | -| **[headroom](https://github.com/chopratejas/headroom)** · chopratejas | 33.6k | Reversible context-compression (SmartCrusher) — inspired our `headroom` engine and the `ccr` retrieve-marker pattern. | -| **[LLMLingua](https://github.com/microsoft/LLMLingua)** · Microsoft | 6.3k | Prompt-compression research (LLMLingua / LLMLingua-2) — inspired our async, code-safe, fail-open `llmlingua` engine. | -| **[llmlingua-2-js](https://github.com/atjsh/llmlingua-2-js)** · atjsh | 27 | The JS/ONNX port (MobileBERT / XLM-RoBERTa) used as the worker-thread backend for our LLMLingua engine. | -| **[Troglodita](https://github.com/leninejunior/troglodita)** · Lenine Júnior | 15 | PT-BR token compression — powers our pt-BR language pack: pleonasm reduction and filler removal tuned for Brazilian-Portuguese grammar. | -| **[ponytail](https://github.com/DietrichGebert/ponytail)** · DietrichGebert | 51.4k | The viral "lazy senior dev" YAGNI-coder skill — inspired our **less-code** Output Style: smallest-working-change steering that cuts _generated_ code (the output-axis sibling to Caveman's terse prose). | +| Project | ⭐ | How it inspired OmniRoute | +| ----------------------------------------------------------------------------- | ----: | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | +| **[Caveman](https://github.com/JuliusBrussee/caveman)** · JuliusBrussee | 78.2k | The viral "why use many token when few token do trick" project — its caveman-speak philosophy powers our standard compression mode and 30+ filler/condensation rules. | +| **[RTK – Rust Token Killer](https://github.com/rtk-ai/rtk)** · rtk-ai | 67.3k | High-performance command-output compression — inspired our RTK engine, JSON filter DSL, raw-output recovery and the stacked RTK → Caveman pipeline. | +| **[headroom](https://github.com/headroomlabs-ai/headroom)** · headroomlabs-ai | 54.5k | Reversible context-compression (SmartCrusher) — inspired our `headroom` engine and the `ccr` retrieve-marker pattern. | +| **[LLMLingua](https://github.com/microsoft/LLMLingua)** · Microsoft | 6.4k | Prompt-compression research (LLMLingua / LLMLingua-2) — inspired our async, code-safe, fail-open `llmlingua` engine. | +| **[llmlingua-2-js](https://github.com/atjsh/llmlingua-2-js)** · atjsh | 28 | The JS/ONNX port (MobileBERT / XLM-RoBERTa) used as the worker-thread backend for our LLMLingua engine. | +| **[Troglodita](https://github.com/leninejunior/troglodita)** · Lenine Júnior | 16 | PT-BR token compression — powers our pt-BR language pack: pleonasm reduction and filler removal tuned for Brazilian-Portuguese grammar. | +| **[ponytail](https://github.com/DietrichGebert/ponytail)** · DietrichGebert | 68.8k | The viral "lazy senior dev" YAGNI-coder skill — inspired our **less-code** Output Style: smallest-working-change steering that cuts _generated_ code (the output-axis sibling to Caveman's terse prose). | ### 🧩 Compact formats, token research & code-aware tooling | Project | ⭐ | How it inspired OmniRoute | | ---------------------------------------------------------------------------------------------- | ----: | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -| **[TOON](https://github.com/toon-format/toon)** · toon-format | 24.6k | Token-Oriented Object Notation — its columnar, header-plus-rows model shaped our tabular compaction stage. | -| **[GCF – Graph Compact Format](https://github.com/blackwell-systems/gcf)** · Blackwell Systems | 11 | Schema-aware "JSON for LLMs" notation — co-inspired our lossless homogeneous-array compaction with `[N rows]` markers. | -| **[token-optimizer-mcp](https://github.com/ooples/token-optimizer-mcp)** · ooples | 409 | Brotli/SQLite cache + per-session context-delta — inspired our `session-dedup` engine. | -| **[token-savior](https://github.com/Mibayy/token-savior)** · Mibayy | 993 | Bash-output compaction + MCP profiles — inspired our compression bail-out discipline and MCP tool-manifest reduction. | -| **[token-saver](https://github.com/ppgranger/token-saver)** · ppgranger | 103 | Content-aware, per-file-type output compression with failure-aware bail-out — validated our per-type dispatch and minimum-gain skip. | -| **[token-optimizer](https://github.com/alexgreensh/token-optimizer)** · alexgreensh | 1.4k | "Find the ghost tokens" — its offload + recoverable-handle pattern informed our CCR offload thinking. | -| **[TokenMizer](https://github.com/Shweta-Mishra-ai/tokenmizer)** · Shweta-Mishra-ai | 1 | A session-graph + cross-turn line-dedup blueprint that informed our session-dedup design. | +| **[TOON](https://github.com/toon-format/toon)** · toon-format | 24.7k | Token-Oriented Object Notation — its columnar, header-plus-rows model shaped our tabular compaction stage. | +| **[GCF – Graph Compact Format](https://github.com/blackwell-systems/gcf)** · Blackwell Systems | 14 | Schema-aware "JSON for LLMs" notation — co-inspired our lossless homogeneous-array compaction with `[N rows]` markers. | +| **[token-optimizer-mcp](https://github.com/ooples/token-optimizer-mcp)** · ooples | 421 | Brotli/SQLite cache + per-session context-delta — inspired our `session-dedup` engine. | +| **[token-savior](https://github.com/Mibayy/token-savior)** · Mibayy | 1.0k | Bash-output compaction + MCP profiles — inspired our compression bail-out discipline and MCP tool-manifest reduction. | +| **[token-saver](https://github.com/ppgranger/token-saver)** · ppgranger | 110 | Content-aware, per-file-type output compression with failure-aware bail-out — validated our per-type dispatch and minimum-gain skip. | +| **[token-optimizer](https://github.com/alexgreensh/token-optimizer)** · alexgreensh | 1.5k | "Find the ghost tokens" — its offload + recoverable-handle pattern informed our CCR offload thinking. | +| **[TokenMizer](https://github.com/Shweta-Mishra-ai/tokenmizer)** · Shweta-Mishra-ai | 2 | A session-graph + cross-turn line-dedup blueprint that informed our session-dedup design. | | **[OmniCompress](https://github.com/jessefreitas/OmniCompress)** · jessefreitas | 2 | Rust columnar-JSON + content-addressed retrieve + cross-message dedup — validated our `headroom`/`ccr`/`session-dedup` engine design and the cache-stable "compressed form is position-independent" invariant. | -| **[mcp-compressor](https://github.com/atlassian-labs/mcp-compressor)** · Atlassian Labs | 80 | MCP tool-schema/description compression — informed our MCP tool-manifest cardinality reduction. | -| **[RepoMapper](https://github.com/pdavis68/RepoMapper)** · pdavis68 | 182 | Aider-style repo-map ranking — informed our repo-map / retrieval-ranking exploration. | +| **[mcp-compressor](https://github.com/atlassian-labs/mcp-compressor)** · Atlassian Labs | 89 | MCP tool-schema/description compression — informed our MCP tool-manifest cardinality reduction. | +| **[RepoMapper](https://github.com/pdavis68/RepoMapper)** · pdavis68 | 181 | Aider-style repo-map ranking — informed our repo-map / retrieval-ranking exploration. | | **[quiet-shell-mcp](https://github.com/mrsimpson/quiet-shell-mcp)** · mrsimpson | 4 | Declarative shell-output reduction over MCP — validated our declarative bash-output compaction. | | **[ts-morph](https://github.com/dsherret/ts-morph)** · David Sherret | 6.1k | TypeScript Compiler API toolkit — inspired our parser-based comment removal that preserves string, template and regex literals. | @@ -1122,27 +1225,27 @@ OmniRoute stands on the shoulders of giants. It started as a fork of **[9router] | Project | ⭐ | How it inspired OmniRoute | | ------------------------------------------------------------------ | ----: | ------------------------------------------------------------------------------------------------------------------- | -| **[Mem0](https://github.com/mem0ai/mem0)** · mem0ai | 58.9k | Universal memory layer — its proxy-as-write/read-boundary model shaped our memory architecture. | -| **[Letta (MemGPT)](https://github.com/letta-ai/letta)** · letta-ai | 23.4k | Stateful agents with tiered memory — inspired our Context Control & Recovery (CCR) tiered model. | +| **[Mem0](https://github.com/mem0ai/mem0)** · mem0ai | 59.8k | Universal memory layer — its proxy-as-write/read-boundary model shaped our memory architecture. | +| **[Letta (MemGPT)](https://github.com/letta-ai/letta)** · letta-ai | 23.6k | Stateful agents with tiered memory — inspired our Context Control & Recovery (CCR) tiered model. | | **[WFGY](https://github.com/onestardao/WFGY)** · onestardao | 1.8k | The ProblemMap taxonomy of 16 recurring RAG/LLM failure modes — the shared vocabulary in our troubleshooting guide. | ### 🛰️ Traffic inspection, MITM & transparent proxy | Project | ⭐ | How it inspired OmniRoute | | --------------------------------------------------------------------------------- | ---: | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -| **[llm-interceptor](https://github.com/chouzz/llm-interceptor)** · chouzz | 46 | MITM interception/analysis of coding-assistant ↔ LLM traffic — our Traffic Inspector ports its SSE merge, conversation normalization, host passthrough and secret masking (MIT). | -| **[ProxyBridge](https://github.com/InterceptSuite/ProxyBridge)** · InterceptSuite | 5.1k | Transparent per-process proxy routing — inspired our crash-safe MITM teardown, socket idle-timeouts, `/proc` process attribution and TPROXY capture. | +| **[llm-interceptor](https://github.com/chouzz/llm-interceptor)** · chouzz | 48 | MITM interception/analysis of coding-assistant ↔ LLM traffic — our Traffic Inspector ports its SSE merge, conversation normalization, host passthrough and secret masking (MIT). | +| **[ProxyBridge](https://github.com/InterceptSuite/ProxyBridge)** · InterceptSuite | 5.3k | Transparent per-process proxy routing — inspired our crash-safe MITM teardown, socket idle-timeouts, `/proc` process attribution and TPROXY capture. | ### 📚 Model data, observability & UI | Project | ⭐ | How it inspired OmniRoute | | -------------------------------------------------------------------------- | ----: | -------------------------------------------------------------------------------------------------------------------------- | -| **[models.dev](https://github.com/anomalyco/models.dev)** · SST / OpenCode | 5.1k | Open database of AI model specs, pricing and capabilities — synced natively into our model catalog. | -| **[React Flow / xyflow](https://github.com/xyflow/xyflow)** · xyflow | 37.1k | The node-based graph library powering our real-time Compression Studio and Combo/Routing Studio. | -| **[LangGraph](https://github.com/langchain-ai/langgraph)** · LangChain | 35.1k | LangGraph Studio's live workflow-graph visualization inspired our Studios' real-time cascade view. | -| **[Langfuse](https://github.com/langfuse/langfuse)** · Langfuse | 29.3k | Its trace → span → generation observability model shaped our Compression Studio waterfall. | +| **[models.dev](https://github.com/anomalyco/models.dev)** · SST / OpenCode | 5.6k | Open database of AI model specs, pricing and capabilities — synced natively into our model catalog. | +| **[React Flow / xyflow](https://github.com/xyflow/xyflow)** · xyflow | 37.4k | The node-based graph library powering our real-time Compression Studio and Combo/Routing Studio. | +| **[LangGraph](https://github.com/langchain-ai/langgraph)** · LangChain | 36.1k | LangGraph Studio's live workflow-graph visualization inspired our Studios' real-time cascade view. | +| **[Langfuse](https://github.com/langfuse/langfuse)** · Langfuse | 30.1k | Its trace → span → generation observability model shaped our Compression Studio waterfall. | | **[Kiali](https://github.com/kiali/kiali)** · Kiali | 3.6k | Istio service-mesh observability — inspired our circuit-breaker badges and error-edge visuals in the Routing/Combo Studio. | -| **[lobe-icons](https://github.com/lobehub/lobe-icons)** · LobeHub | 2.1k | AI/LLM brand logos that render the provider icons across our dashboard. | +| **[lobe-icons](https://github.com/lobehub/lobe-icons)** · LobeHub | 2.2k | AI/LLM brand logos that render the provider icons across our dashboard. | ### 🛡️ Security @@ -1168,7 +1271,7 @@ MIT License - see [LICENSE](LICENSE) for details. **[⬆ Back to top](#-omniroute)** · Built with ❤️ for the open-source AI community. -OmniRoute v3.8.24 · Node ≥22.0.0 · MIT License · omniroute.online +OmniRoute v3.8.43 · Node ≥22.0.0 · MIT License · omniroute.online
diff --git a/bin/cli/commands/doctor.mjs b/bin/cli/commands/doctor.mjs index 08f68ab84f..0a2b78db74 100644 --- a/bin/cli/commands/doctor.mjs +++ b/bin/cli/commands/doctor.mjs @@ -3,7 +3,7 @@ import net from "node:net"; import os from "node:os"; import path from "node:path"; import { createDecipheriv, scryptSync } from "node:crypto"; -import { pathToFileURL } from "node:url"; +import { fileURLToPath, pathToFileURL } from "node:url"; import { resolveDataDir, resolveStoragePath } from "../data-dir.mjs"; import { printHeading } from "../io.mjs"; import { t } from "../i18n.mjs"; @@ -397,7 +397,7 @@ async function checkServerLiveness(options = {}) { export async function collectDoctorChecks(context = {}, options = {}) { const rootDir = context.rootDir || - path.resolve(path.dirname(new URL(import.meta.url).pathname), "..", "..", ".."); + path.resolve(path.dirname(fileURLToPath(import.meta.url)), "..", "..", ".."); const dataDir = resolveDataDir(); const dbPath = resolveStoragePath(dataDir); diff --git a/bin/cli/commands/serve.mjs b/bin/cli/commands/serve.mjs index 2260127a38..9ba6579c3f 100644 --- a/bin/cli/commands/serve.mjs +++ b/bin/cli/commands/serve.mjs @@ -1,5 +1,5 @@ import { spawn } from "node:child_process"; -import { existsSync } from "node:fs"; +import { existsSync, readFileSync } from "node:fs"; import { join, dirname } from "node:path"; import { fileURLToPath } from "node:url"; import { platform, totalmem } from "node:os"; @@ -16,6 +16,7 @@ import { import { resolveTlsOptions } from "../../../scripts/dev/tls-options.mjs"; const __dirname = dirname(fileURLToPath(import.meta.url)); +const _pkg = JSON.parse(readFileSync(join(__dirname, "..", "..", "..", "package.json"), "utf8")); // URL scheme for the "OmniRoute is running" banner — flipped to https when // opt-in TLS (#5242) is active. Process-scoped: one `serve` run = one scheme. @@ -48,11 +49,13 @@ export function registerServe(program) { .option("--no-tray", t("serve.no_tray") || "Disable system tray icon") .option( "--tls-cert ", - t("serve.tls_cert") || "Path to a TLS certificate (PEM) to serve HTTPS (also OMNIROUTE_TLS_CERT)" + t("serve.tls_cert") || + "Path to a TLS certificate (PEM) to serve HTTPS (also OMNIROUTE_TLS_CERT)" ) .option( "--tls-key ", - t("serve.tls_key") || "Path to the TLS private key (PEM) to serve HTTPS (also OMNIROUTE_TLS_KEY)" + t("serve.tls_key") || + "Path to the TLS private key (PEM) to serve HTTPS (also OMNIROUTE_TLS_KEY)" ) .action(async (opts) => { await runServe(opts); @@ -60,6 +63,8 @@ export function registerServe(program) { } export async function runServe(opts = {}) { + const startedAt = performance.now(); + const { isNativeBinaryCompatible } = await import("../../../scripts/build/native-binary-compat.mjs"); const { getNodeRuntimeSupport, getNodeRuntimeWarning } = @@ -78,6 +83,7 @@ export async function runServe(opts = {}) { | |__| | | | | | | | | | | | \\ \\ (_) | |_| | || __/ \\____/|_| |_| |_|_| |_|_|_| \\_\\___/ \\__,_|\\__\\___| \x1b[0m`); + console.log(`\x1b[2m v${_pkg.version}\x1b[0m\n`); const nodeSupport = getNodeRuntimeSupport(); if (!nodeSupport.nodeCompatible) { @@ -187,7 +193,15 @@ export async function runServe(opts = {}) { } if (opts.noRecovery) { - return runWithoutRecovery(serverJs, env, memoryLimit, dashboardPort, apiPort, noOpen); + return runWithoutRecovery( + serverJs, + env, + memoryLimit, + dashboardPort, + apiPort, + noOpen, + startedAt + ); } return runWithSupervisor( @@ -199,6 +213,7 @@ export async function runServe(opts = {}) { noOpen, opts.log === true, opts.maxRestarts ?? 2, + startedAt, useTray ); } @@ -219,7 +234,7 @@ function runDaemon(serverJs, env, memoryLimit, dashboardPort, apiPort) { console.log(` \x1b[1mAPI Base:\x1b[0m ${urlScheme}://localhost:${apiPort}/v1`); } -function runWithoutRecovery(serverJs, env, memoryLimit, dashboardPort, apiPort, noOpen) { +function runWithoutRecovery(serverJs, env, memoryLimit, dashboardPort, apiPort, noOpen, startedAt) { // #5238: skip the explicit CLI --max-old-space-size when the user pinned the // heap via NODE_OPTIONS (a CLI arg would shadow/override their value). const server = spawn("node", [...buildNodeHeapArgs(process.env, memoryLimit), serverJs], { @@ -240,7 +255,7 @@ function runWithoutRecovery(serverJs, env, memoryLimit, dashboardPort, apiPort, (text.includes("Ready") || text.includes("started") || text.includes("listening")) ) { started = true; - onReady(dashboardPort, apiPort, noOpen); + onReady(dashboardPort, apiPort, noOpen, startedAt); } }); @@ -274,7 +289,7 @@ function runWithoutRecovery(serverJs, env, memoryLimit, dashboardPort, apiPort, setTimeout(() => { if (!started) { started = true; - onReady(dashboardPort, apiPort, noOpen); + onReady(dashboardPort, apiPort, noOpen, startedAt); } }, 15000); } @@ -288,6 +303,7 @@ async function runWithSupervisor( noOpen, showLog, maxRestarts, + startedAt, useTray = false ) { if (showLog) process.env.OMNIROUTE_SHOW_LOG = "1"; @@ -325,7 +341,7 @@ async function runWithSupervisor( waitForServer(dashboardPort, 60000).then(async (up) => { if (up) { if (useTray) await maybeStartTray(dashboardPort, apiPort, supervisor); - onReady(dashboardPort, apiPort, noOpen); + onReady(dashboardPort, apiPort, noOpen, startedAt); } }); } @@ -366,18 +382,20 @@ async function maybeStartTray(port, apiPort, supervisor) { } catch (err) { // tray is optional — do not fail the server, but surface why it failed so // "--tray shows nothing" is diagnosable instead of silent (#4605). - process.stderr.write( - `[omniroute][tray] failed to start: ${err?.message ?? String(err)}\n` - ); + process.stderr.write(`[omniroute][tray] failed to start: ${err?.message ?? String(err)}\n`); } } -async function onReady(dashboardPort, apiPort, noOpen) { +async function onReady(dashboardPort, apiPort, noOpen, startedAt) { const dashboardUrl = `${urlScheme}://localhost:${dashboardPort}`; const apiUrl = `${urlScheme}://localhost:${apiPort}`; + const elapsed = + typeof startedAt === "number" && Number.isFinite(startedAt) + ? ((performance.now() - startedAt) / 1000).toFixed(1) + : "0.0"; console.log(` - \x1b[32m✔ OmniRoute is running!\x1b[0m + \x1b[32m✔ OmniRoute is running!\x1b[0m \x1b[2m(started in ${elapsed}s)\x1b[0m \x1b[1m Dashboard:\x1b[0m ${dashboardUrl} \x1b[1m API Base:\x1b[0m ${apiUrl}/v1 diff --git a/bin/cli/commands/setup-claude.mjs b/bin/cli/commands/setup-claude.mjs index 5161e7b9ee..2bdedfa64b 100644 --- a/bin/cli/commands/setup-claude.mjs +++ b/bin/cli/commands/setup-claude.mjs @@ -50,18 +50,78 @@ export function buildProfileSettings(modelId, baseUrl, cfg) { return JSON.stringify(settings, null, 2) + "\n"; } +/** + * Generate Claude Code profile files for a live model catalog. Shared by the + * `setup-claude` CLI command and the post-model-sync auto-sync so both stay + * behaviorally identical. Writes `/profiles//settings.json` + * (directory-per-profile); never touches the active/default Claude config. + * @param {Array} models + * @param {{claudeHome?:string, baseUrl:string, dryRun?:boolean, only?:string}} opts + * @returns {Promise<{written:number, skipped:number, profiles:Array<{name:string, model:string, filePath:string}>}>} + */ +export async function syncClaudeProfilesFromModels(models, opts = {}) { + const claudeHome = opts.claudeHome || join(os.homedir(), ".claude"); + const profilesRoot = join(claudeHome, "profiles"); + const baseUrl = opts.baseUrl; + const dryRun = Boolean(opts.dryRun); + const onlyFilter = opts.only ? opts.only.split(",").map((s) => s.trim()) : null; + + if (!dryRun && !existsSync(profilesRoot)) { + mkdirSync(profilesRoot, { recursive: true }); + } + + let written = 0; + let skipped = 0; + const profiles = []; + + for (const m of models) { + const id = typeof m === "string" ? m : (m.id ?? ""); + if (!id) { + skipped++; + continue; + } + if (onlyFilter && !onlyFilter.some((f) => id.includes(f))) { + skipped++; + continue; + } + + const cfg = categoriseModel(id); + if (!cfg) { + skipped++; + continue; + } + + const dir = join(profilesRoot, cfg.name); + const filePath = join(dir, "settings.json"); + const content = buildProfileSettings(id, baseUrl, cfg); + + if (dryRun) { + console.log(`\n── [dry-run] ${filePath} ──`); + console.log(content); + } else { + mkdirSync(dir, { recursive: true }); + writeFileSync(filePath, content, "utf8"); + } + profiles.push({ name: cfg.name, model: id, filePath }); + written++; + } + + return { written, skipped, profiles }; +} + /** * @param {{remote?:string, port?:string, apiKey?:string, claudeHome?:string, dryRun?:boolean, only?:string}} opts * @returns {Promise} */ export async function runSetupClaudeCommand(opts = {}) { const port = Number(opts.port ?? process.env.PORT ?? 20128) || 20128; - const baseUrl = (opts.remote ?? `http://localhost:${port}`).replace(/\/+$/, "").replace(/\/v1$/, ""); + const baseUrl = (opts.remote ?? `http://localhost:${port}`) + .replace(/\/+$/, "") + .replace(/\/v1$/, ""); const apiKey = opts.apiKey ?? opts["api-key"] ?? process.env.OMNIROUTE_API_KEY ?? ""; const claudeHome = opts.claudeHome ?? opts["claude-home"] ?? join(os.homedir(), ".claude"); const profilesRoot = join(claudeHome, "profiles"); const dryRun = Boolean(opts.dryRun ?? opts["dry-run"]); - const onlyFilter = opts.only ? opts.only.split(",").map((s) => s.trim()) : null; printHeading("OmniRoute → Claude Code profile generator"); printInfo(`Connecting to ${baseUrl} …`); @@ -89,42 +149,25 @@ export async function runSetupClaudeCommand(opts = {}) { printInfo(`Received ${models.length} models from ${baseUrl}`); - if (!dryRun && !existsSync(profilesRoot)) { - mkdirSync(profilesRoot, { recursive: true }); - } + const { written, skipped, profiles } = await syncClaudeProfilesFromModels(models, { + claudeHome, + baseUrl, + dryRun, + only: opts.only, + }); - let written = 0; - for (const m of models) { - const id = typeof m === "string" ? m : m.id ?? ""; - if (!id) continue; - if (onlyFilter && !onlyFilter.some((f) => id.includes(f))) continue; - - const cfg = categoriseModel(id); - if (!cfg) continue; - - const dir = join(profilesRoot, cfg.name); - const filePath = join(dir, "settings.json"); - const content = buildProfileSettings(id, baseUrl, cfg); - - if (dryRun) { - console.log(`\n── [dry-run] ${filePath} ──`); - console.log(content); - } else { - mkdirSync(dir, { recursive: true }); - writeFileSync(filePath, content, "utf8"); - printSuccess(` ✓ profiles/${cfg.name}/settings.json (${id})`); - } - written++; - } - - const skipped = models.length - written; if (!dryRun) { + for (const profile of profiles) { + printSuccess(` ✓ profiles/${profile.name}/settings.json (${profile.model})`); + } console.log(""); printSuccess(`${written} Claude Code profiles written to ${profilesRoot}`); if (skipped > 0) printInfo(`${skipped} models skipped (no matching profile pattern)`); console.log("\nTo use a profile:"); console.log(" omniroute launch --profile # e.g. omniroute launch --profile glm52"); - console.log(" # or: CLAUDE_CONFIG_DIR=~/.claude/profiles/ claude (export ANTHROPIC_AUTH_TOKEN first)"); + console.log( + " # or: CLAUDE_CONFIG_DIR=~/.claude/profiles/ claude (export ANTHROPIC_AUTH_TOKEN first)" + ); } else { console.log(`\n[dry-run] ${written} profiles would be written (${skipped} skipped)`); } @@ -143,7 +186,10 @@ export function registerSetupClaude(program) { .option("--remote ", "Remote OmniRoute URL, e.g. http://192.168.0.15:20128") .option("--api-key ", "OmniRoute API key (defaults to OMNIROUTE_API_KEY env var)") .option("--claude-home ", "Claude home dir (default: ~/.claude)") - .option("--only ", "Comma-separated substrings — only matching model IDs (e.g. glm,kimi)") + .option( + "--only ", + "Comma-separated substrings — only matching model IDs (e.g. glm,kimi)" + ) .option("--dry-run", "Print what would be written without touching the filesystem") .action(async (opts) => { const exitCode = await runSetupClaudeCommand(opts); diff --git a/bin/cli/commands/setup-codex.mjs b/bin/cli/commands/setup-codex.mjs index 76aee2fb61..feb0451a42 100644 --- a/bin/cli/commands/setup-codex.mjs +++ b/bin/cli/commands/setup-codex.mjs @@ -39,29 +39,83 @@ export function categoriseModel(modelId) { { re: /kmc\/kimi-k2\.6/, name: "kimi-k26", ctx: 131072, compact: 112000, toolLimit: 32768 }, { re: /glm\/glm-5\.2-max/, name: "glm52max", ctx: 131072, compact: 112000, toolLimit: 32768 }, { re: /glm\/glm-5\.2$/, name: "glm52", ctx: 131072, compact: 112000, toolLimit: 32768 }, - { re: /opencode-go\/mimo-v2\.5-pro/, name: "mimo-pro", ctx: 131072, compact: 112000, toolLimit: 32768 }, - { re: /opencode-go\/qwen3\.7-plus/, name: "qwen37plus", ctx: 32768, compact: 28000, toolLimit: 16384 }, + { + re: /opencode-go\/mimo-v2\.5-pro/, + name: "mimo-pro", + ctx: 131072, + compact: 112000, + toolLimit: 32768, + }, + { + re: /opencode-go\/qwen3\.7-plus/, + name: "qwen37plus", + ctx: 32768, + compact: 28000, + toolLimit: 16384, + }, ]; // ── Good models (high effort) ───────────────────────────────────────────── const goodPatterns = [ - { re: /ollamacloud\/deepseek-v4-pro/, name: "deepseek-pro", ctx: 131072, compact: 112000, toolLimit: 32768 }, - { re: /opencode-go\/mimo-v2\.5$/, name: "mimo", ctx: 131072, compact: 112000, toolLimit: 32768 }, + { + re: /ollamacloud\/deepseek-v4-pro/, + name: "deepseek-pro", + ctx: 131072, + compact: 112000, + toolLimit: 32768, + }, + { + re: /opencode-go\/mimo-v2\.5$/, + name: "mimo", + ctx: 131072, + compact: 112000, + toolLimit: 32768, + }, ]; // ── Simple models (no effort) ───────────────────────────────────────────── const simplePatterns = [ { re: /ollamacloud\/gemma4:31b/, name: "gemma4", ctx: 32768, compact: 28000, toolLimit: 16384 }, - { re: /ollamacloud\/nemotron-3-super/, name: "nemotron", ctx: 32768, compact: 28000, toolLimit: 16384 }, - { re: /ollamacloud\/gpt-oss:20b/, name: "gptoss", ctx: 32768, compact: 28000, toolLimit: 16384 }, + { + re: /ollamacloud\/nemotron-3-super/, + name: "nemotron", + ctx: 32768, + compact: 28000, + toolLimit: 16384, + }, + { + re: /ollamacloud\/gpt-oss:20b/, + name: "gptoss", + ctx: 32768, + compact: 28000, + toolLimit: 16384, + }, ]; // ── Fast models (low effort) ────────────────────────────────────────────── const fastPatterns = [ - { re: /ollamacloud\/deepseek-v4-flash/, name: "deepseek-flash", ctx: 65536, compact: 56000, toolLimit: 16384 }, - { re: /ollamacloud\/gemini-3-flash/, name: "gemini-flash", ctx: 1000000, compact: 850000, toolLimit: 32768 }, + { + re: /ollamacloud\/deepseek-v4-flash/, + name: "deepseek-flash", + ctx: 65536, + compact: 56000, + toolLimit: 16384, + }, + { + re: /ollamacloud\/gemini-3-flash/, + name: "gemini-flash", + ctx: 1000000, + compact: 850000, + toolLimit: 32768, + }, { re: /glm\/glm-5-turbo/, name: "glm5turbo", ctx: 131072, compact: 112000, toolLimit: 16384 }, - { re: /glm\/glm-4\.7-flash/, name: "glm47flash", ctx: 131072, compact: 112000, toolLimit: 16384 }, + { + re: /glm\/glm-4\.7-flash/, + name: "glm47flash", + ctx: 131072, + compact: 112000, + toolLimit: 16384, + }, ]; for (const p of thinkingPatterns) { @@ -80,6 +134,92 @@ export function categoriseModel(modelId) { return null; } +function firstPositiveNumber(...values) { + for (const value of values) { + if (typeof value === "number" && Number.isFinite(value) && value > 0) { + return value; + } + } + return null; +} + +function shortHash(value) { + let hash = 5381; + for (let i = 0; i < value.length; i++) { + hash = ((hash << 5) + hash) ^ value.charCodeAt(i); + } + return (hash >>> 0).toString(36); +} + +function profileNameFromModelId(modelId) { + const normalized = String(modelId) + .toLowerCase() + .replace(/[^a-z0-9]+/g, "-") + .replace(/^-+|-+$/g, ""); + const base = normalized || "model"; + if (base.length <= 96) return base; + return `${base.slice(0, 84).replace(/-+$/g, "")}-${shortHash(base)}`; +} + +function hasAnyValue(values, patterns) { + return values.some((value) => patterns.some((pattern) => pattern.test(value))); +} + +export function isCodexCompatibleTextModel(model) { + if (typeof model === "string") return true; + + const id = String(model?.id ?? "").toLowerCase(); + const type = String(model?.type ?? "").toLowerCase(); + const outputModalities = Array.isArray(model?.output_modalities) + ? model.output_modalities.map((value) => String(value).toLowerCase()) + : []; + + if (type && !["chat", "text", "language", "llm", "model"].includes(type)) { + return false; + } + + const unsupportedPatterns = [ + /(^|[/_-])(image|img|video|veo|seedance|audio|speech|voice|tts|stt|whisper)([/_-]|$)/, + /(^|[/_-])(embedding|embeddings|embed|rerank|moderation|transcription)([/_-]|$)/, + ]; + if (hasAnyValue([id, type], unsupportedPatterns)) return false; + + const nonTextModalities = [/^(image|video|audio)$/]; + if (hasAnyValue(outputModalities, nonTextModalities)) return false; + + if (outputModalities.length > 0 && !outputModalities.includes("text")) { + return false; + } + + return true; +} + +export function fallbackCodexProfile(modelId, model) { + if (!isCodexCompatibleTextModel(model)) return null; + + const ctx = + typeof model === "string" + ? 128000 + : (firstPositiveNumber( + model.context_length, + model.max_context_window_tokens, + model.max_input_tokens + ) ?? 128000); + const maxOutput = + typeof model === "string" + ? null + : firstPositiveNumber(model.max_output_tokens, model.output_token_limit); + const toolLimit = Math.min(Math.max(maxOutput ?? 16384, 8192), 32768); + + return { + name: profileNameFromModelId(modelId), + ctx, + compact: Math.floor(ctx * 0.85), + summary: false, + toolLimit, + }; +} + /** Build the TOML content for a single profile. */ function buildProfileToml(modelId, cfg) { const lines = [ @@ -105,6 +245,52 @@ function buildProfileToml(modelId, cfg) { return lines.join("\n") + "\n"; } +export async function syncCodexProfilesFromModels(models, opts = {}) { + const codexHome = opts.codexHome || join(os.homedir(), ".codex"); + const dryRun = Boolean(opts.dryRun); + const onlyFilter = opts.only ? opts.only.split(",").map((s) => s.trim()) : null; + + if (!dryRun && !existsSync(codexHome)) { + mkdirSync(codexHome, { recursive: true }); + } + + let written = 0; + let skipped = 0; + const profiles = []; + + for (const m of models) { + const id = typeof m === "string" ? m : (m.id ?? ""); + if (!id) { + skipped++; + continue; + } + if (onlyFilter && !onlyFilter.some((f) => id.includes(f))) { + skipped++; + continue; + } + + const cfg = categoriseModel(id) ?? fallbackCodexProfile(id, m); + if (!cfg) { + skipped++; + continue; + } + + const filePath = join(codexHome, `${cfg.name}.config.toml`); + const content = buildProfileToml(id, cfg); + + if (dryRun) { + console.log(`\n── [dry-run] ${filePath} ──`); + console.log(content); + } else { + writeFileSync(filePath, content, "utf8"); + } + profiles.push({ name: cfg.name, model: id, filePath }); + written++; + } + + return { written, skipped, profiles }; +} + // ── Command ─────────────────────────────────────────────────────────────────── /** @@ -146,39 +332,17 @@ export async function runSetupCodexCommand(opts = {}) { printInfo(`Received ${models.length} models from ${baseUrl}`); - // ── Ensure codex home exists ────────────────────────────────────────────── - if (!dryRun && !existsSync(codexHome)) { - mkdirSync(codexHome, { recursive: true }); - } - // ── Generate profiles ───────────────────────────────────────────────────── - let written = 0; - let skipped = 0; - - for (const m of models) { - const id = typeof m === "string" ? m : (m.id ?? ""); - if (!id) continue; - if (onlyFilter && !onlyFilter.some((f) => id.includes(f))) continue; - - const cfg = categoriseModel(id); - if (!cfg) continue; - - const filePath = join(codexHome, `${cfg.name}.config.toml`); - const content = buildProfileToml(id, cfg); - - if (dryRun) { - console.log(`\n── [dry-run] ${filePath} ──`); - console.log(content); - } else { - writeFileSync(filePath, content, "utf8"); - printSuccess(` ✓ ${cfg.name}.config.toml (${id})`); - } - written++; - } - - skipped = models.length - written; + const { written, skipped, profiles } = await syncCodexProfilesFromModels(models, { + codexHome, + dryRun, + only: opts.only, + }); if (!dryRun) { + for (const profile of profiles) { + printSuccess(` ✓ ${profile.name}.config.toml (${profile.model})`); + } console.log(""); printSuccess(`${written} profiles written to ${codexHome}`); if (skipped > 0) { @@ -210,10 +374,7 @@ export function registerSetupCodex(program) { "--api-key ", "OmniRoute API key for the remote instance (defaults to OMNIROUTE_API_KEY env var)" ) - .option( - "--codex-home ", - "Directory where profile files are written (default: ~/.codex)" - ) + .option("--codex-home ", "Directory where profile files are written (default: ~/.codex)") .option( "--only ", "Comma-separated substrings — only generate profiles for matching model IDs (e.g. glm,kimi)" diff --git a/config/quality/complexity-baseline.json b/config/quality/complexity-baseline.json index d429f43987..4114fd995a 100644 --- a/config/quality/complexity-baseline.json +++ b/config/quality/complexity-baseline.json @@ -1,6 +1,8 @@ { "_comment": "Catraca de complexidade (check-complexity.mjs, ESLint core rules complexity>=15 e max-lines-per-function>80 sobre src+open-sse+electron+bin via eslint.complexity.config.mjs). Conta total de violacoes; so pode cair. --update ratcheta.", - "count": 1981, + "count": 1995, + "_rebaseline_2026_07_02_5798_release_green": "1982->1995 (+13). Inherited v3.8.43 cycle drift surfaced by the release-green unblock #5798 / PR #5896: check:complexity measures 1995 on BOTH the pristine release tip (0d3875a98) and this PR's HEAD — identical, so all +13 came from the 2026-07-01/02 merge burst (providers/usage/dashboard fixes merged via --admin while the fast-gates queue was base-red). This PR touches only gate scripts, docs, baselines and test files — 0 production logic. Tighten via --update next cycle.", + "_rebaseline_2026_07_01_v3843_release": "1981->1982 (+1). v3.8.43 cycle drift, surfaced after check:mutation-test-coverage was fixed (it was masked behind that earlier step in the Fast Quality Gates chain). 1982 = the value measured by check:complexity on BOTH fce85136c (release tip) and 6d7060e21 (release + the 5 CI fixes) — identical, so all +1 is inherited cycle drift; the fixes touch only test files + linkify.ts safeHttpHref (cyclomatic ~4, well under the >=15 threshold, 0 new violations) + config JSON. Tighten via --update next cycle.", "_rebaseline_2026_06_28_v3840_5237_reconcile": "1980->1981 (+1). Inherited release/v3.8.40 drift surfaced while merging PR #5237 (impersonation-UA refresh) — the +1 is present on the pristine release tip (d8a392a47) WITHOUT #5237's changes, so it is #5222 (antigravity fallback-LRU retry) / #5221 (command-code) growth that merged via --admin without ratcheting complexity (the PR->release fast-gates do not run check:complexity). #5237 itself is complexity-net-zero: its only edits are a single UA constant, a regenerated golden snapshot, and baseline JSONs. Structural reduction tracked in #3501.", "_rebaseline_2026_06_27_v3838_release": "1978->1980 (+2). v3.8.38 cycle-close drift surfaced by the release-green pre-flight (check:complexity does NOT run on PR->release fast-gates). +2 from late-cycle feature/fix merges (compression fidelity-gate steps #5143, SSE hardening). Release-finalize working tree touches ONLY CHANGELOG.md + i18n mirrors + the 2 baseline JSONs — 0 production-code change. Structural reduction tracked in #3501.", "_rebaseline_2026_06_26_v3838_ownerprs_batch": "1972->1978 (+6). Drift do lote de merges de PRs do dono + contribuidores em release/v3.8.38 (sessao /review-prs): #4845 (antigravity convertGeminiToOpenAI), #5105 (executor zenmux-free), #5020 (executor grok-cli), #4940 (usage dedupe guard), #5093 (resilience: quota cutoff/gemini mime/model-lockout cooldown), #5015 (quota hydration + auto-combo scoping). Cada um e crescimento de feature/fix legitimo recem-TDD'd, nao regressao; o gate check:complexity NAO roda no fast-path PR->release, entao o ramo acumula sem rebaselinar (mesma familia dos rebaselines anteriores). #5121 cookie-dedup foi mantido complexity-NEUTRO via extracao do helper findExistingCookieConnection. Reducao estrutural fica como debt (#3501).", diff --git a/config/quality/dependency-allowlist.json b/config/quality/dependency-allowlist.json index 24985ff314..c5c9518414 100644 --- a/config/quality/dependency-allowlist.json +++ b/config/quality/dependency-allowlist.json @@ -38,6 +38,7 @@ "bcryptjs", "better-sqlite3", "bottleneck", + "bun", "c8", "clsx", "commander", diff --git a/config/quality/file-size-baseline.json b/config/quality/file-size-baseline.json index e957dfe278..15a134b796 100644 --- a/config/quality/file-size-baseline.json +++ b/config/quality/file-size-baseline.json @@ -1,5 +1,9 @@ { "_comment": "Catraca de tamanho (check-file-size.mjs). frozen so pode encolher; arquivos novos <= cap. --update ratcheta.", + "_rebaseline_2026_07_02_5798_release_green": "Release-green unblock #5798 / PR #5896 (2026-07-02): the quality.yml fast-gates job was base-red for EVERY PR->release (whole queue failing), from growth inherited via already-merged PRs — no offending PR branch left to fix. Prod frozen raised: AddApiKeyModal.tsx 869->905, providerPageHelpers.ts 996->1021, RequestLoggerV2.tsx 1316->1553, src/sse/services/auth.ts 2403->2405, antigravity.ts 1806->1813, base.ts 1502->1536 (1533 inherited + 3 lines from this PR's own typecheck:core fix in resolveBaseUrl), advancedTools.ts 1118->1120, accountFallback.ts 1783->1790, openai-to-kiro.ts 842->853, openai-responses.ts 1035->1092, stream.ts 2710->2727; new-above-cap frozen: webProvidersA.ts 805, tokenHealthCheck.ts 830. Test frozen raised: cc-compatible-provider 1179->1217, translator-openai-to-kiro 999->1088, web-cookie-providers-new 827->845; new-above-cap: response-sanitizer.test.ts 906. These files remain frozen (cannot grow further); the release captain's rebaseline-at-release supersedes this note.", + "_rebaseline_2026_06_30_5552_flat_rate_cost": "Issue #5552 own growth: src/app/api/usage/analytics/route.ts 941->942 (+1 = the `flatRateAsZero: true` cost option at the existing computeUsageRowCost chokepoint, so subscription/cookie-web providers show $0 instead of an inflated per-token estimate in analytics). The flat-rate classifier (isFlatRateProvider + the provider-id set) lives in a new leaf src/lib/usage/flatRateProviders.ts (61 LOC, 1502 (+2 = import + the single `requestCredentials = withForcedResponsesUpstream(...)` const threaded through buildUrl/buildHeaders/applyConfiguredUserAgent/ccRequestDefaults/transformRequest at the existing fetch-loop chokepoint), open-sse/executors/default.ts 876->877 (+1 = the `_omnirouteForceResponsesUpstream` short-circuit in the buildUrl `/responses` vs `/chat/completions` decision), tests/unit/executor-default-base.test.ts 1477->1523 (+46 = the new regression test that asserts a Responses-shaped MCP request routes to /responses for openai-compatible providers). The detection helpers (shouldForceResponsesUpstream/withForcedResponsesUpstream/isRecord, ~50 LOC) were EXTRACTED out of base.ts into a new leaf open-sse/executors/forceResponsesUpstream.ts (60 LOC, 851 (+5 = import + the `strategy === \"fusion\"` conditional that mounts the new FusionDefaultsFields component). The bulk of the fusion-defaults UI (judgeModel + fusionTuning inputs) was EXTRACTED to a new src/app/(dashboard)/dashboard/settings/components/FusionDefaultsFields.tsx (92 LOC, 997 + chat 1632->1635 + chatHelpers 811->842 (#5351 opencode visible-rotation-logs/ProxyEgress/SQL-vars), base.ts 1475->1497 + openai-to-claude 805->823 (#5352 thinking hydrate/redacted-replay + #5342 empty-messages guard), accountFallback 1777->1783 (#5346 402 auto-disable depleted key) + account-fallback-service.test 1569->1572 (#5346 test), provider-validation-specialty.test 2801->2843 (#5358 grok cf_clearance + #5337 gemini catalog). Cycle-close drift from merged campaign PRs absorbed at release close per generate-release Phase 0 (the PR->release fast-gates do NOT run check:file-size); all irreducible chokepoint/feature growth next to existing branches, covered by per-PR tests. Structural shrink tracked under decomposition roadmap #3501.", "_rebaseline_2026_06_26_v3838_ownerprs_batch": "Lote /review-prs v3.8.38 (PRs do dono + contribuidores): src/lib/db/providers.ts 1093->1107 (+14 = #5121 cookie-dedup branch extraido para o helper findExistingCookieConnection — a extracao reduz a complexidade ciclomatica de createProviderConnection mas adiciona ~14 linhas ao arquivo; trade-off consciente complexity-vs-file-size) e src/lib/usage/usageHistory.ts 934->983 (+49 = #4940 dedup guard SELECT-before-INSERT + endpoint backfill + scheduleStatsEvent debounce). Crescimento de feature/fix legitimo recem-TDD'd; o fast-path PR->release nao roda check:file-size, entao o ramo acumula sem rebaselinar. Reducao estrutural fica como debt (#3501).", "_rebaseline_2026_06_26_relgreen_db_test": "Release-green follow-up: tests/unit/db-core-init.test.ts 867->877 (+10 = the invalid-DATA_DIR test now captures the rejection and asserts both Error type AND message — restores net-neutral assert count after #5117's consolidation, satisfying check:test-masking — instead of a single assert.rejects).", @@ -129,8 +133,8 @@ "frozen": { "_rebaseline_pr1043_minimax_tts": "Upstream port decolua/9router#1043 (toanalien) own growth: audioSpeech.ts 965->1061 (+96). Adds MiniMax T2A v2 TTS dispatch (handleMinimaxSpeech + hexToBytes helper) — provider entry was already in audioRegistry (format: minimax-tts) but no handler existed, falling through to the OpenAI-compatible default that fails (T2A has custom shape + hex-encoded audio + base_resp envelope). New branch sits next to the other inline provider branches (xiaomi-mimo, coqui, tortoise, aws-polly) — extracting would just create indirection. Covered by tests/unit/minimax-tts-1043.test.ts (3 tests, GREEN: success, base_resp error, invalid-hex).", "open-sse/config/providerRegistry.ts": 4731, - "open-sse/executors/antigravity.ts": 1806, - "open-sse/executors/base.ts": 1500, + "open-sse/executors/antigravity.ts": 1813, + "open-sse/executors/base.ts": 1540, "open-sse/executors/chatgpt-web.ts": 3206, "open-sse/executors/claude-web.ts": 1057, "open-sse/executors/codex.ts": 1541, @@ -152,17 +156,17 @@ "open-sse/handlers/videoGeneration.ts": 1265, "open-sse/mcp-server/schemas/tools.ts": 1497, "open-sse/mcp-server/server.ts": 1555, - "open-sse/mcp-server/tools/advancedTools.ts": 1118, + "open-sse/mcp-server/tools/advancedTools.ts": 1120, "_rebaseline_2026_06_27_5193_antigravity_basered": "Base-red (pre-existing release drift, fast-gate PR->release skips check:file-size): accountFallback.ts 1773->1777 and src/app/api/providers/[id]/test/route.ts 924->940 were already over their frozen caps on release/v3.8.39 independent of any antigravity change. Owner chose to rebaseline (keep the documented issue-reference comments #1846/#1449/#347 etc.) rather than accept the contributor comment-stripping in #5200/#5198. Reverted #5200 to restore the comments; bumped these two frozen caps to the actual base sizes. No logic change.", - "open-sse/services/accountFallback.ts": 1783, - "open-sse/services/batchProcessor.ts": 828, + "open-sse/services/accountFallback.ts": 1790, + "open-sse/services/batchProcessor.ts": 915, "open-sse/services/browserBackedChat.ts": 850, "open-sse/services/claudeCodeCompatible.ts": 1202, "_rebaseline_pr4592_exclude_exhausted_auto": "Reconcile #4592 already-merged growth: combo.ts 2991->3036 (+45, terminal-status quota-cutoff exclusion in buildAutoCandidates + opt-in gate). Fast-gate PR->release does not run check:file-size.", "_rebaseline_2026_06_24_headroom_strategy": "Headroom-aware connection selection (dario technique): combo.ts 3168->3180 (+12 = a new `else if (strategy === \"headroom\")` dispatch branch in handleComboChat that delegates to orderTargetsByHeadroom + its log line, plus the import). The actual logic lives OUT of the god-file: the pure ranker rankByHeadroom/computeHeadroom is the new leaf open-sse/services/combo/headroomRanking.ts (91 LOC, 3190 (+10 = one new `else if (strategy === \"quota-share\")` dispatch branch in handleComboChat that delegates 100% to selectQuotaShareTarget + its log line, plus the import). All the new logic lives OUT of the god-file in two new leaves under open-sse/services/combo/: quotaShareInflight.ts (in-flight counter with TTL/lease, ~150 LOC 3225 (+35) = one new `else if (strategy === \"task-aware\")` dispatch branch delegating 100% to selectTaskAwareTarget + its imports/log lines. All scoring/classification logic lives OUT of the god-file in the new leaf open-sse/services/taskAwareRouting.ts (553 LOC 854, -35), but the StackOptions.fidelityGate field, the `const fidelityGate` reads at the two stacked-loop dispatch chokepoints, and the import of FidelityGateConfig are irreducible wiring that cannot leave strategySelector without an architectural refactor of the pre-existing stacked pipeline. Net: 889->854 (+6 vs the pre-Milestone-B frozen 848). Covered by tests/unit/compression/*.test.ts (940 pass).", "_rebaseline_2026_06_28_5243_risk_gate_prepass": "PR #5243 (compression risk-gate pre-pass) own growth: open-sse/services/compression/strategySelector.ts 854->899 (+45). The three exported entry points (applyCompression/applyStackedCompression/applyStackedCompressionAsync) become thin wrappers over pure-extracted private bodies (runCompression/runStackedCompression/runStackedCompressionAsync) so the risk-gate mask->run->restore wrapper sits strictly OUTSIDE the per-step loop — a single universal integration point. The wrapper logic itself (resolveRiskGate/withRiskGate) lives in the new riskGate/strategyWrap.ts (960 (+61 = the opt-in result-memoization branches in applyCompression/applyCompressionAsync — principal+determinism gate, makeMemoKey lookup/store with model+supportsVision folded into the key, recompute-with-memo-off). Default off (memoizeCompressionResults), so zero behavior change. The memo helpers live in the leaf resultMemo.ts (969 (gate units). #5193 (+~4: remote paste instruction shown for all remote incl. Google + its rationale comment) and #5203 (+~5: handleManualSubmit credential-blob branch + button guard; submit logic extracted to oauthBlobSubmit.ts to minimize). Frozen set to the SUM so either merge order passes. Cohesive at the existing manual-submit chokepoint.", "src/shared/components/OAuthModal.tsx": 969, - "src/shared/components/RequestLoggerV2.tsx": 1316, + "src/shared/components/RequestLoggerV2.tsx": 1553, "src/shared/components/analytics/charts.tsx": 1558, "src/shared/constants/cliTools.ts": 875, "src/shared/constants/pricing.ts": 1662, @@ -246,15 +250,19 @@ "src/shared/services/cliRuntime.ts": 1090, "src/shared/validation/schemas.ts": 2523, "_rebaseline_2026_06_28_5275_correlation_id_extract": "Extraction of the safe CorrelationId subset of #5275 (hartmark) — request correlation id stored in call_logs (migration 109) and returned via the X-Correlation-Id response header, WITHOUT the combo/resilience or build/lazy-loading changes (those stay in #5275). Own growth: callLogs.ts 975->985 (correlation_id column on CallLogSummaryRow + read/map), usageHistory.ts 983->988 (correlationId metadata normalize), chat.ts 1575->1632 (withCorrelationId response wiring + combo-failure log carrying correlationId), chatHelpers.ts new 811 (withCorrelationId helper + reqId threading; was 7913017, combos/page 4594->4608, AddApiKeyModal 868->869, providerPageHelpers 974->996, chat.ts 1635->1647, auth.ts 2401->2403, batchProcessor 828->915, combo.ts 3368->3387) + 2 novos acima do cap (huggingchat.ts 813, tests web-cookie-providers-new 827) + 4 test files cresceram. Modularizacao deferida (blast-radius mid-release); congelado no estado atual p/ o proximo ciclo ratchetar daqui.", + "src/lib/providers/validation/webProvidersA.ts": 805, + "src/lib/tokenHealthCheck.ts": 830 }, "testCap": 800, "testFrozen": { @@ -263,11 +271,11 @@ "tests/integration/skills-pipeline.test.ts": 918, "tests/unit/account-fallback-service.test.ts": 1572, "tests/unit/arena-elo-sync.test.ts": 830, - "tests/unit/batch_api.test.ts": 1303, - "tests/unit/cc-compatible-provider.test.ts": 1179, - "tests/unit/chatcore-sanitization.test.ts": 829, + "tests/unit/batch_api.test.ts": 1324, + "tests/unit/cc-compatible-provider.test.ts": 1217, + "tests/unit/chatcore-sanitization.test.ts": 831, "tests/unit/chatcore-translation-paths.test.ts": 2810, - "tests/unit/chatgpt-web.test.ts": 3159, + "tests/unit/chatgpt-web.test.ts": 3170, "tests/unit/combo-routing-engine.test.ts": 3213, "tests/unit/combo-strategy-fallbacks.test.ts": 880, "tests/unit/db-core-init.test.ts": 877, @@ -276,7 +284,7 @@ "tests/unit/deepseek-web.test.ts": 1081, "tests/unit/executor-antigravity.test.ts": 942, "tests/unit/executor-codex.test.ts": 1347, - "tests/unit/executor-default-base.test.ts": 1477, + "tests/unit/executor-default-base.test.ts": 1523, "tests/unit/grok-web.test.ts": 2437, "tests/unit/image-generation-handler.test.ts": 2019, "tests/unit/model-sync-route.test.ts": 1016, @@ -286,25 +294,27 @@ "_rebaseline_2026_06_27_5193_antigravity_test": "#5193 own test growth: oauth-providers-config.test.ts 870->873 (+3: antigravity projectId assertion + 50ms tick for the now fire-and-forget onboarding, matching the no-PKCE/no-openid flow).", "tests/unit/oauth-providers-config.test.ts": 873, "tests/unit/perplexity-web.test.ts": 959, - "tests/unit/provider-models-route.test.ts": 1618, - "tests/unit/provider-validation-specialty.test.ts": 2843, + "tests/unit/provider-models-route.test.ts": 1628, + "tests/unit/provider-validation-specialty.test.ts": 2874, "_rebaseline_pr4613_compatible_provider_groups": "Reconcile #4613 already-merged growth: providers-page-utils.test.ts 1004->1052 (+48, buildCompatibleProviderGroups partition unit test). Fast-gate PR->release does not run check:file-size, so this surfaced post-merge.", "tests/unit/providers-page-utils.test.ts": 1052, "tests/unit/reasoning-cache.test.ts": 980, "tests/unit/route-edge-coverage.test.ts": 1234, "tests/unit/search-handler-extended.test.ts": 1124, "tests/unit/sse-auth.test.ts": 1553, - "tests/unit/stream-utils.test.ts": 2435, + "tests/unit/stream-utils.test.ts": 2445, "tests/unit/token-refresh-service.test.ts": 1353, "tests/unit/translator-friendly-test-bench.test.tsx": 848, "tests/unit/translator-helper-branches.test.ts": 870, "tests/unit/translator-openai-responses-req.test.ts": 1097, "tests/unit/translator-openai-to-gemini.test.ts": 1579, - "tests/unit/translator-openai-to-kiro.test.ts": 999, + "tests/unit/translator-openai-to-kiro.test.ts": 1088, "tests/unit/translator-resp-gemini-to-openai.test.ts": 1234, "tests/unit/usage-service-hardening.test.ts": 1633, "tests/unit/vscode-token-routes.test.ts": 1212, - "tests/unit/combo-config.test.ts": 881 + "tests/unit/combo-config.test.ts": 881, + "tests/unit/web-cookie-providers-new.test.ts": 845, + "tests/unit/response-sanitizer.test.ts": 906 }, "_rebaseline_2026_06_09": "Re-baseline consciente pre-release v3.8.19: 9 arquivos cresceram durante o ciclo (features mergeadas: RequestLoggerV2 +281 request-logger rework, stream +101, combo +73, chatCore +45, catalog +32 fable-5/catalog-flag, callLogs +4, accountFallback +2, usageHistory novo 840) + core.ts +7 (fix resetAllDbModuleState, PR 3536). A catraca segue valendo destes valores — proximo crescimento falha. Decisao: encolher (esp. RequestLoggerV2/chatCore) e a issue #3501 ficam para o ciclo seguinte.", "_rebaseline_2026_06_11_phase1f": "Phase 1f (#3501): ProviderDetailPageClient.tsx 4948→4062 (-886 LOC); 3 novos hooks extraídos. useProviderConnections.ts=954 acima do cap=800 — justificado: extração direta do god-component (zero lógica nova), própria redução do cliente supera o custo. useProviderSettings.ts=263 e useProviderModels.ts=154 já abaixo do cap.", diff --git a/config/quality/quality-baseline.json b/config/quality/quality-baseline.json index 9b4c150fa7..e85d042676 100644 --- a/config/quality/quality-baseline.json +++ b/config/quality/quality-baseline.json @@ -2,8 +2,10 @@ "_comment": "Catraca de qualidade. 'down' = nao pode aumentar; 'up' = nao pode cair. Atualize via 'npm run quality:ratchet -- --update' (somente quando melhora). Cada valor e um numero REAL medido, nunca um chute. Cobertura entra na Fase 4 a partir de um run de cobertura mergeada no CI.", "metrics": { "eslintWarnings": { - "value": 4121, + "value": 4199, + "_rebaseline_2026_07_02_v3843_release_close": "4158->4199 (+41). v3.8.43 release-close drift measured by the release-green pre-flight (the Quality Ratchet does NOT run on PR->release fast-gates, so warnings accrued unmeasured across the ~120 commits merged after the mid-cycle 4158 rebaseline — the compression T02/T05/T06/T07/T08/T10 engine families, memory typed decay, provider adds Ollama/SenseNova, ~55 SSE/translator/kiro/oauth/dashboard fixes, and the god-file decomposition wave). Trust-but-verify: measured 4199 via `npm run lint` on the release-finalize working tree INCLUDING my changes (CHANGELOG/i18n/README docs + kiro pricing data entry + the 3 base-red CODE fixes: opencode fabrication removal, resolveEffectiveKey type-widen, openai-to-claude claudeFinishEmitted flag + 4 test-alignment files + golden snapshot regen) — the code fixes NET-REMOVE lines and add no `any`/unused, and lint reported 4199 both before and after them, so all +41 is inherited cycle drift (`any` warn-allowed in open-sse/ + tests/). Tighten via --require-tighten next cycle.", "direction": "down", + "_rebaseline_2026_07_01_v3843_release": "4121->4158 (+37). v3.8.43 cycle drift surfaced by the release-green pre-flight; the Quality Ratchet does NOT run on PR->release fast-gates, so warnings accrued unmeasured across this cycle. 4158 = the value measured by the CI Quality Ratchet on the release tip fce85136c (release PR #5609). Trust-but-verify: the fix/release-v3843-ci-reds branch touches only test files (rtk-mcp-tools de-flake, compression-studio e2e anchor, oauth-error-linkify hardening test) + src/shared/utils/linkify.ts (eslint-clean, 0 warnings) + stryker.conf.json + this baseline -> 0 new warnings, so all +37 is inherited cycle drift (any warn-allowed in open-sse/ + tests/). Tighten via --require-tighten next cycle.", "_rebaseline_2026_06_30_v3842_release": "4116->4121 (+5). v3.8.42 cycle drift surfaced by the release-green pre-flight (the Quality Ratchet does NOT run on PR->release fast-gates, so warnings accrued unmeasured across this cycle's 90 commits — chatgpt-web PoW sha3-512 BoringSSL fix #5540, provider baseUrl/i18n umbrella #5511, proxy union proxyUrlMap+acct.proxy #5521, dead-code + duplication waves #5468-#5495, tls-options packaging #5503, release-freeze + .npmrc fetch-retries #5506, dast-smoke spawn-prefix client-safe extraction #5546, plus ~30 SSE/translator/combo/dashboard fixes). Trust-but-verify: measured 4121 via `npm run check:release-green` on the working tree INCLUDING my reconciliation (CHANGELOG/i18n/golden snapshot + file-size baseline) — those touch only config JSON + a provider snapshot (eslint-ignored) and contribute 0 warnings; all +5 is inherited cycle drift (`any` warn-allowed in open-sse/ + tests/). Tighten via --require-tighten next cycle.", "_rebaseline_2026_06_29_v3841_release": "4103->4116 (+13). v3.8.41 cycle drift surfaced by the release-green collect (the Quality Ratchet does NOT run on PR->release fast-gates, so warnings accrued unmeasured across this cycle's 52 commits — relay backend #5315, gemini catalog #5337, services dashboard #5299, empty-Claude-messages guard #5342, thinking-budget/redacted-replay + marker opt-out #5312/#5352/#5367, opencode proxy-pool + observability #5217/#5370/#5351, cors + HTTPS-serve #5242/#5360/#5361, grok cf_clearance #5350/#5358, oauth/chatgpt-web/routing/cli/dashboard/rerank #5326/#5240/#5239/#5238/#5264/#5332, partially offset by the dead-code sweep #5321-#5371). Trust-but-verify: measured 4116 via `npm run quality:collect` on the working tree INCLUDING my reconciliation (CHANGELOG/i18n/README/env docs + baselines) AND the lint-fix in useServiceLogs.ts — that fix REMOVES a setState-in-effect ERROR (eslintErrors stays 0) and adds an `open` listener with no `any`/unused, contributing 0 warnings; all +13 is inherited cycle drift (`any` warn-allowed in open-sse/ + tests/). Tighten via --require-tighten next cycle.", "_rebaseline_2026_06_29_v3840_release": "4090->4103 (+13). v3.8.40 cycle drift surfaced by the release-green pre-flight + the release PR Quality Ratchet (the ratchet does NOT run on PR->release fast-gates, so warnings accrued unmeasured across this cycle's ~57 commits — compression roadmap relevance/hard-budget/memoization/transparency/saliency/splitter/tool_search/RTK/QuantumLock #5289/#5288/#5286/#5284/#5285/#5283/#5269/#5268/#5260, ~20 SSE/translator/combo fixes #5248/#5250/#5254/#5261/#5255/#5273/#5258, M365 Copilot provider #5302, public-origin centralization #5278). Trust-but-verify: measured 4103 locally via `npm run quality:collect` on the release tip INCLUDING my reconciliation commits (CHANGELOG + main merge + the 2 regression test fixes 165c823f5) — the test fixes add 0 `any`/warnings (health-autopilot added a NextRequest import + asserts; chat-pipeline changed one Accept string + a comment), so all +13 is inherited cycle drift (`any` warn-allowed in open-sse/ + tests/). Tighten via --require-tighten next cycle.", @@ -103,16 +105,19 @@ "_rebaseline_2026_06_28_v3839_release": "78.4 -> 77.5 (-0.9, beyond the 0.5 eps). v3.8.39 cycle drift surfaced ONLY on the release PR (i18n-ui-coverage does NOT run on PR->release fast-gates). The cycle added new UI strings (compression studio TOON A/B table, antigravity remote-login dashboard field, amber warning icon) to the en denominator faster than the 37 non-en locales were translated; those locales need `npm run i18n:run` with OMNIROUTE_TRANSLATION_API_KEY (unavailable locally) — same precedent as _rebaseline_2026_06_18_v3828_cycle_close + _quality_rebaseline_2026_06_20_ci_ratchet. Measured by CI collect-metrics (run 28317145160) = 77.5. My release-finalize tree changes no src/i18n/messages/*.json. Tightening is tracked as follow-up (run i18n:run with creds)." }, "deadExports": { - "value": 225, + "value": 227, "direction": "down", + "_rebaseline_2026_07_01_v3843_release": "225->227 (+2). v3.8.43 cycle drift, surfaced in the Quality Ratchet job after eslintWarnings was rebaselined (check:dead-code runs there). 227 = measured by check:dead-code (knip) on the release tip 4635076eb. The 5 CI fixes add 0 dead exports: safeHttpHref in linkify.ts is module-local AND used (called by linkifyText); no new exports; test files are not scanned. Tighten via --update next cycle.", "dedicatedGate": true, "_rebaseline_2026_06_30_v3842_deadcode_wave": "310 -> 225. Measured by `node scripts/check/check-dead-code.mjs` on the v3.8.42 tip after the JxnLexn dead-code (#5463/#5464/#5466) + duplication (#5471..#5500) wave landed: DEAD_EXPORTS=133 + DEAD_FILES=92 = 225. The stale 310 was the v3.8.38 release snapshot never ratcheted on PR->release fast-gates (check:dead-code runs only on ci.yml PR->main, not quality.yml). Tightening to the true measured value; release-time captain rebaselines up if parallel cycle merges add dead exports.", "_rebaseline_2026_06_27_v3838_release": "345->346 (+1). v3.8.38 cycle drift surfaced by the release-green pre-flight (Quality Ratchet does NOT run on PR->release fast-gates). Net +1 inherited from this cycle's feature/fix merges (new executors/providers, compression fidelity-gate module) minus #5138's removal of dead legacy store modules. Release-finalize working tree touches ONLY CHANGELOG.md + i18n mirrors + README + baselines — 0 production-code change. Structural cleanup tracked as debt.", "_rebaseline_2026_06_26_v3837_release": "343->345. v3.8.37 cycle drift surfaced by the release-green pre-flight (the Quality Ratchet does NOT run on PR->release fast-gates, so warnings/complexity accrued unmeasured across this cycle's 76 commits — provider adds DGrid/Pioneer/xAI, headroom proxy lifecycle #4649, ~50 SSE/translator fixes, Engine Combos #5062). Trust-but-verify: this release-finalize working tree touches ONLY CHANGELOG.md, docs/i18n/*/CHANGELOG.md mirrors, and these baselines — 0 production-code change, so all drift is inherited cycle drift (`any` warn-allowed in open-sse/ + tests/). Tighten via --require-tighten next cycle." }, "cognitiveComplexity": { - "value": 842, + "value": 856, + "_rebaseline_2026_07_02_5798_release_green": "845->856 (+11). Inherited v3.8.43 cycle drift surfaced by the release-green unblock #5798 / PR #5896: check:cognitive-complexity measures 856 on BOTH the pristine release tip (0d3875a98) and this PR's HEAD — identical, so the PR is cognitive-net-zero (it touches only gate scripts, docs, baselines and test files). Drift from the 2026-07-01/02 merge burst. Tighten via --update next cycle.", "direction": "down", + "_rebaseline_2026_07_01_v3843_release": "842->845 (+3). v3.8.43 cycle drift, surfaced after check:mutation-test-coverage was fixed (masked behind it in the Fast Quality Gates chain). 845 = measured by check:cognitive-complexity on BOTH fce85136c and 6d7060e21 (identical) — all +3 is inherited cycle drift; the 5 CI fixes add 0 (safeHttpHref cognitive ~2, under the 15 threshold). Tighten via --update next cycle.", "dedicatedGate": true, "_rebaseline_2026_06_29_v3841_release": "841->842 (+1). v3.8.41 cycle drift surfaced by the release-PR Quality Ratchet (cognitive-complexity does NOT run on PR->release fast-gates). The Phase-0 pre-flight measured 840 on the pre-campaign tip; the campaign's +34 later commits (thinking-budget/marker #5312/#5352/#5367, opencode proxy-pool/observability #5217/#5370/#5351, cors/HTTPS #5242/#5360/#5361, grok #5350/#5358, oauth/routing/cli #5326/#5239/#5238) added +2 net. My release-finalize changes are docs/baselines + a lint-fix (useServiceLogs open-listener), a test alignment (encryption.spec) and a pack-allowlist entry — all complexity-neutral. Structural shrink tracked in #3501.", "_rebaseline_2026_06_27_v3838_release": "833->841 (+8). v3.8.38 cycle drift surfaced by the release-green pre-flight (cognitive-complexity does NOT run on PR->release fast-gates). Inherited drift from this cycle's ~78 feature/fix merges (compression fidelity-gate #5143, SSE/streaming hardening #5124/#5108/#5085, resilience #5093, quota keepalive #5102, contributor provider/translator branches). god-file decomposition #3501 is complexity-neutral. Release-finalize working tree touches ONLY CHANGELOG.md + i18n mirrors + README + baselines — 0 production-code change. Structural shrink tracked in #3501.", diff --git a/config/quality/test-masking-allowlist.json b/config/quality/test-masking-allowlist.json index 6e49b03066..ea8c1f3d08 100644 --- a/config/quality/test-masking-allowlist.json +++ b/config/quality/test-masking-allowlist.json @@ -7,5 +7,9 @@ "tests/unit/compression/session-dedup.test.ts": "v3.8.29 #4226: the vestigial SessionDedup round-trip helper was removed from source; its 2 asserts were removed accordingly (32→30). Helper no longer exists. Verified legitimate, not masking. Prune after v3.8.29 merges to main.", "tests/unit/compression/ultra.test.ts": "v3.8.29 #4253: the vestigial SLM seam + dead deprecated alias were removed from the ultra compression engine; 6 asserts covering the removed seam were removed accordingly (49→43). Verified legitimate, not masking. Prune after v3.8.29 merges to main.", "tests/unit/db-backup-extended.test.ts": "v3.8.29 #4132: db-backup de-flake — 1 timing-sensitive assertion on fire-and-forget backup completion was removed in favor of awaiting actual completion (44→43). Verified legitimate, not masking. Prune after v3.8.29 merges to main.", - "@omniroute/opencode-plugin/tests/combos.test.ts": "v3.8.31 #4384: the plugin now prefixes every catalog key with the `omniroute` provider id and drops the legacy `combo/` namespace; the test asserting raw-deletion + a `combo/` key (a namespace that no longer exists) was removed and the remaining asserts switched to `omniroute/` keys (82→81). Asserts updated to the new key contract, not weakened. Verified legitimate. Prune after v3.8.31 merges to main." + "@omniroute/opencode-plugin/tests/combos.test.ts": "v3.8.31 #4384: the plugin now prefixes every catalog key with the `omniroute` provider id and drops the legacy `combo/` namespace; the test asserting raw-deletion + a `combo/` key (a namespace that no longer exists) was removed and the remaining asserts switched to `omniroute/` keys (82→81). Asserts updated to the new key contract, not weakened. Verified legitimate. Prune after v3.8.31 merges to main.", + "tests/unit/chatgpt-web.test.ts": "v3.8.43 #5549: fix(chatgpt-web) restore dot-form Pro model ids — dois assert.equal separados (base Pro slug + pass-through slug) consolidados num único assert.equal(body.model, expectedSlugById[omniId], ...) orientado por tabela de mapeamento (281→280). Asserts consolidados, não enfraquecidos. Verificado legítimo. Prune após v3.8.43 mergear para main.", + "tests/unit/chatcore-sanitization.test.ts": "v3.8.43 #5805: fix(translator) strip orphaned tool results — orphaned tool_result blocks (no matching tool_use) are now removed by stripOrphanedToolResults BEFORE content normalization, so the 3 positive `[Tool Result: …]`-text asserts were replaced by removal asserts (no tool_result block, no text), net 65→64. Behavior aligned to the merged #5805 contract; the sibling 'preserves Claude passthrough tool_result' assert (matching tool_use) is untouched. Verified legitimate, not masking. Prune after v3.8.43 merges to main.", + "src/app/(dashboard)/dashboard/providers/[id]/__tests__/useModelVisibilityHandlers.test.tsx": "v3.8.43 #5856: fix(dashboard) unify CSRF origin fallback — the model-visibility handler no longer issues a separate `/api/auth/csrf` fetch, so the two asserts pinning the 2nd fetch + its CSRF header were removed and the fetch-count assert updated 2→1 (7→5). Asserts follow the reduced fetch behavior, not weakened. Verified legitimate. Prune after v3.8.43 merges to main.", + "tests/unit/provider-validation-specialty.test.ts": "v3.8.43 #5855: fix(qwen-web) unblock validator (retired endpoint) — the old `chat.qwen.ai/api/v2/user` probe asserts (exact URL / Authorization / source / Cookie / WAF-error) no longer apply after the endpoint migration and were replaced by new chathub-path behavior asserts (valid/error/warning), net 406→400. Asserts migrated to the new API surface (#5855/#5432), not weakened. Verified legitimate. Prune after v3.8.43 merges to main." } diff --git a/docs/architecture/REPOSITORY_MAP.md b/docs/architecture/REPOSITORY_MAP.md index 133e9d0548..05b4488e14 100644 --- a/docs/architecture/REPOSITORY_MAP.md +++ b/docs/architecture/REPOSITORY_MAP.md @@ -98,7 +98,6 @@ OmniRoute/ | **knip.json** | Knip config — unused files/exports/deps (feeds the dead-code gate) | | **stryker.conf.json** | Stryker mutation-testing config | | **.size-limit.json** | size-limit bundle budget config | -| **semcheck.yaml** | semcheck (spec↔code drift) config | | **promptfooconfig.yaml** | promptfoo eval config | | **.gitleaks.toml** | gitleaks secret-scan ruleset | | **.zizmor.yml** | zizmor GitHub-Actions security-lint config | diff --git a/docs/architecture/RESILIENCE_GUIDE.md b/docs/architecture/RESILIENCE_GUIDE.md index 6abe76222a..d321b6236f 100644 --- a/docs/architecture/RESILIENCE_GUIDE.md +++ b/docs/architecture/RESILIENCE_GUIDE.md @@ -81,7 +81,7 @@ OmniRoute has three distinct but related resilience mechanisms. Each has a diffe **Terminal states (NOT cooldowns):** -- `banned` +- `banned` — set by banned-keyword / account-ban detection (see [BAN_DETECTION](../security/BAN_DETECTION.md)) - `expired` - `credits_exhausted` diff --git a/docs/architecture/ROUTER_BACKENDS.md b/docs/architecture/ROUTER_BACKENDS.md new file mode 100644 index 0000000000..e99e1f8e30 --- /dev/null +++ b/docs/architecture/ROUTER_BACKENDS.md @@ -0,0 +1,149 @@ +--- +title: "Router Backends & Embedded Services (ADR)" +version: 3.8.43 +lastUpdated: 2026-07-02 +--- + +# Router Backends & Embedded Services — architecture contract (ADR) + +> **Status:** Accepted · **Context:** [#5670](https://github.com/diegosouzapw/OmniRoute/issues/5670), +> [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) · **Contract:** `domain/routing/routerBackends.ts` +> (typed registry — code lands with [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) + +This ADR pins down how `ts` (native), `bifrost`, `cliproxy`, `9router`, and +VibeProxy-compatible engines relate to each other, so contributors stop +conflating two things that are architecturally distinct. It documents the typed +registry introduced by the router-backend-registry work as the single source of +truth for that model. + +## The core distinction — two orthogonal axes + +An engine's role is described by **two independent axes**, encoded together in the +registry's `RouterBackendDefinition`: + +1. **Lifecycle** (`RouterBackendLifecycle`) — _how the engine runs_: + - `in-process` — runs inside the OmniRoute Node process (the native TS pipeline). + - `supervised` — a local child process OmniRoute installs/starts/stops/health-checks + via `ServiceSupervisor`, then consumes as a provider connection. + - `external` — an HTTP endpoint OmniRoute dispatches to but does **not** manage + (configured by an env base URL). + - `disabled` — registered but not selectable. +2. **Selection axis** (relay routing backend) — _whether the relay dispatches to it_: + `RelayRoutingBackend = "ts" | "bifrost" | "auto"` in + `src/app/api/v1/relay/chat/completions/routingBackend.ts`. + +The mistake to avoid: treating "embedded service" and "routing backend" as one +list. They are not. A `supervised` engine (9router/cliproxy) is a **provider +connection consumed by the native pipeline**, not an alternate relay dispatch +backend. `bifrost` is the reverse — a relay dispatch backend that (historically) +was `external`-only. + +## The registry — single source of truth + +The `domain/routing/routerBackends.ts` contract (code lands with +[#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) declares every engine once, with its +lifecycle, capabilities, service identity, default port, health config, and +telemetry support. Consumers look engines up via `getRouterBackend(id)`, +`listRouterBackends()`, and `listRouterBackendsByCapability(cap)` instead of +special-casing each sidecar. + +| Backend | Lifecycle | Service (axis A) | Relay backend (axis B) | Health | Default port | +| ----------- | ------------ | ---------------- | ---------------------- | ------------- | ------------ | +| `ts` | `in-process` | — | `ts` (native) | — | — | +| `bifrost` | `external`¹ | —¹ | `bifrost` / `auto` | `/health` | — | +| `cliproxy` | `supervised` | `cliproxy` | — (provider) | `/v1/models` | 8317 | +| `9router` | `supervised` | `9router` | — (provider) | `/api/health` | 20130 | +| `vibeproxy` | `external` | — | — (provider adapter) | `/v1/models` | — | + +¹ Bifrost's promotion to a `supervised` embedded service (installable/startable +from `/api/services/bifrost/`) is tracked in +[#5817](https://github.com/diegosouzapw/OmniRoute/pull/5817); until it merges, +Bifrost is `external`-only (reachable solely via `BIFROST_BASE_URL`). + +`capabilities` (`chat`, `responses`, `streaming`, `tools`, `vision`, +`oauth-backed`, `dashboard-embed`, `model-sync`, `native-hot-path`) let callers +filter by what an engine can actually do rather than hard-coding per-id branches. + +## Axis A — embedded services (supervised process side) + +- **Registry of supervised processes:** `src/lib/services/bootstrap.ts` `SERVICES[]` + (today: `9router`, `cliproxy`). +- **Lifecycle owner:** `src/lib/services/ServiceSupervisor.ts` — `start()` spawns the + child, gates on `waitForHealthy()`, taps stdout/stderr into a ring buffer; + `stop()` SIGTERM→SIGKILL; all serialized under a lock. +- **State union** (`src/lib/services/types.ts`): + `not_installed | stopped | starting | running | stopping | error`, plus an + orthogonal `HealthState = healthy | unhealthy | unknown`. +- **Why a separate process (not an in-proc SDK)?** Process isolation is what makes + install/start/stop/health/logs independently controllable per sidecar and lets the + loopback spawn-guard apply. Modeling an in-proc adapter is future work — the + `native-hot-path` capability flag is where that would be expressed. + +### Lifecycle route contract (`/api/services//…`) + +Status codes are **state/verb/path-specific by design** — this is the contract, not +inconsistency: + +| Call | Condition | Status | +| ---------------------------- | ------------------------------- | ------------------------------------ | +| `POST .../start` | service `not_installed` | **409** (precondition) | +| `POST .../stop` | already stopped | **200** (idempotent no-op) | +| `GET .../status` | OK | **200** (`live ?? row ?? "unknown"`) | +| `POST .../start` | spawn failure | **503** (transient) | +| `GET .../status`, `.../stop` | uncaught error | **500** | +| `GET /api/services//logs` | unknown tool `` | **404** `Service '' not found` | +| `GET .../status?reveal=key` | missing `X-Reveal-Confirm: yes` | **403** (9router only) | +| **any** `/api/services/*` | caller not loopback/private-LAN | **403 LOCAL_ONLY** | + +All error bodies are shaped by `createErrorResponse()` → +`{ error: { message, type }, requestId }`, where `type` is derived from the status +(`500→server_error`, `404→not_found`, `409→conflict`, else `invalid_request`) and is +the machine-actionable discriminator. Messages are pre-sanitized +(`sanitizeErrorMessage()`, Hard Rule #12). + +**The loopback guard** is the most common source of a `403`: `/api/services/` is in +`LOCAL_ONLY_API_PREFIXES` (`src/server/authz/routeGuard.ts`) and +`src/server/authz/policies/management.ts` rejects any non-loopback / non-private-LAN +caller **before auth**, because these routes spawn child processes (Hard Rules 15 +and 17). Reaching them through a public tunnel is `403` by design. + +## Axis B — relay routing backend (dispatch side) + +Only the relay proxy path `/api/v1/relay/chat/completions` selects a dispatch +backend; the main `/api/v1/chat/completions` surface never consults +`routingBackend.ts`. + +- **Selection** (`resolveRelayRoutingBackend`): a single global env toggle — + `OMNIROUTE_RELAY_BACKEND` / `RELAY_ROUTING_BACKEND` ∈ {`ts`, `bifrost`, `auto`}. + If unset, `auto` when Bifrost is configured+enabled, else `ts`. +- **Behavior:** + - `bifrost` (forced): Bifrost failure → hard `502`, no fallback. + - `auto`: try Bifrost, on failure/cooldown silently fall through to native. + - `ts` / post-fallback: the native `open-sse` translator/executor pipeline. +- **Cooldown:** per-`baseUrl` failure cooldown in `bifrostCooldown.ts`. + +Selection is **all-or-nothing at the relay level today** — there is no per-provider +or per-request engine swap on `release/v3.8.43`. The per-request gate is being added +by the sidecar-manifest work +([#5869](https://github.com/diegosouzapw/OmniRoute/pull/5869) manifest + +[#5870](https://github.com/diegosouzapw/OmniRoute/pull/5870) `shouldTryBifrostForRequest`), +which lets `auto` route only manifest-eligible providers through Bifrost. + +## Dashboard integration + +The services dashboard polls `GET /api/services//status` every 5s via +`src/app/(dashboard)/dashboard/providers/services/hooks/useServiceStatus.ts`, +returning `{ tool, state, pid, port, health, installedVersion, latestVersion, +updateAvailable, autoStart, … }`. There is no shared availability-context provider — +each component calls the hook per tool. On `!res.ok` the hook currently surfaces a +bare `HTTP `; mapping the `error.type` field to a human explanation is a +tracked UX improvement, not a contract change. + +## Consequences + +- New engines register once in `ROUTER_BACKENDS`; consumers gain them via capability + queries without new per-id branches. +- "Is this a service or a routing backend?" is answered by the `lifecycle` field, not + by which list an id happens to appear in. +- The Bifrost supervision (#5817) and native hot-path migration (#5670) build on this + shared contract instead of special-casing each sidecar. diff --git a/docs/frameworks/MEMORY.md b/docs/frameworks/MEMORY.md index 306eac9ca0..14f35f68a6 100644 --- a/docs/frameworks/MEMORY.md +++ b/docs/frameworks/MEMORY.md @@ -223,7 +223,13 @@ chronological order if the FTS table is missing or the FTS query throws. ### Optional: Qdrant (vector store tier 2) `src/lib/memory/qdrant.ts` implements an optional Qdrant integration as tier 2 -vector store. Enabled via `qdrantEnabled` in settings / toggle in Engine tab. +vector store. Retrieval only routes to Qdrant when the engine selector +`memoryVectorStore === "qdrant"` — the default `"auto"` (and `"sqlite-vec"`) +**never** select Qdrant. The Engine-tab toggle sets **both** `qdrantEnabled` and +`memoryVectorStore` together: enabling makes Qdrant the primary store, disabling +resets to `"auto"` (#5597 — before that fix, enabling was inert because nothing +wrote the engine selector). If Qdrant is unreachable or returns nothing, retrieval +falls back to sqlite-vec → FTS5. - `upsertSemanticMemoryPoint()` — embed `key + content` with the configured embedding model, ensure the collection exists (creates cosine-distance @@ -251,6 +257,26 @@ routes under `src/app/api/settings/qdrant/` are all wired as of v3.8.6: | `/api/settings/qdrant/cleanup` | `POST` | Remove expired / old points | | `/api/settings/qdrant/embedding-models` | `GET` | List available embedding models | +**Behavior notes (what to expect):** + +- **Engine selection** — enabling Qdrant in the Engine tab makes it the primary + store (sets `memoryVectorStore="qdrant"`); disabling resets to `"auto"` (#5597). +- **No back-fill** — only memories created/updated **after** Qdrant is enabled are + written to it (fire-and-forget dual-write). Pre-existing SQLite memories are **not** + migrated; "Reindex Now" rebuilds the sqlite-vec index only, not Qdrant. +- **Vector dimension is auto-detected** from the actual embedding on first use — there + is no dimension field to fill in. Changing the embedding model after a collection + exists is **not** auto-handled: the existing collection is left untouched, dimension- + mismatched writes/searches fail and fall back to sqlite-vec. Recreate the collection + (new name, or delete it in Qdrant) to switch embedders. +- **Distance metric** — always **Cosine** (hardcoded on collection creation; not + configurable). +- **Auth** — API key only (sent as the `api-key` header; optional for unauthenticated + local Docker). JWT/RBAC are not used. +- **Config fields** — the UI exposes `host`, `port`, `collection`, `embeddingModel`, + `apiKey`. `vectorSize` / `hnswEfConstruct` are env/DB only and `vectorSize` is not + used for collection creation (dimension comes from the embedding). + ### Vector quantization (int8 — opt-in, both backends) Both vector backends support **opt-in int8 quantization** to cut the memory diff --git a/docs/guides/CLAUDE-CODE-CONFIGURATION.md b/docs/guides/CLAUDE-CODE-CONFIGURATION.md index 96ab1f918d..a34ee464fd 100644 --- a/docs/guides/CLAUDE-CODE-CONFIGURATION.md +++ b/docs/guides/CLAUDE-CODE-CONFIGURATION.md @@ -80,6 +80,14 @@ profile per model at `~/.claude/profiles//settings.json`, reusing the > active context), or export `ANTHROPIC_AUTH_TOKEN` yourself and run > `CLAUDE_CONFIG_DIR=~/.claude/profiles/ claude`. +**Auto-sync after model discovery (opt-in).** OmniRoute can regenerate these same +`~/.claude/profiles//settings.json` files automatically whenever a provider model +sync changes the live catalog — so new/renamed models get profiles without re-running the +command. It is **off by default**: toggle it from the **CLI Code dashboard** ("CLI profile +auto-sync" → Claude Code), or set `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES=true` (it also honors +`CLI_ALLOW_CONFIG_WRITES`, on by default). When enabled it only writes profile files; it never +changes your active/default Claude config, auth, or the `~/.claude/settings.json`. + ### Generating + using profiles ```bash diff --git a/docs/guides/CLI-INTEGRATIONS.md b/docs/guides/CLI-INTEGRATIONS.md index d9b04f301f..ca4d90a363 100644 --- a/docs/guides/CLI-INTEGRATIONS.md +++ b/docs/guides/CLI-INTEGRATIONS.md @@ -37,7 +37,7 @@ server and writes the config locally. | Command | Tool | What it writes | Key flags | Local vs remote | |---------|------|----------------|-----------|-----------------| -| `omniroute setup-codex` | OpenAI Codex CLI | `~/.codex/.config.toml` — one profile per matched model (`codex --profile `) | `--remote` `--api-key` `--only` `--dry-run` `--port` `--codex-home` | Both | +| `omniroute setup-codex` | OpenAI Codex CLI | `~/.codex/.config.toml` — one profile per compatible text model (`codex --profile `) | `--remote` `--api-key` `--only` `--dry-run` `--port` `--codex-home` | Both | | `omniroute setup-claude` | Claude Code | `~/.claude/profiles//settings.json` — one profile per matched model (`CLAUDE_CONFIG_DIR`) | `--remote` `--api-key` `--only` `--dry-run` `--port` `--claude-home` | Both | | `omniroute setup-opencode` | OpenCode (openai-compatible) | `~/.config/opencode/opencode.json` — `omniroute` provider with every catalog model (`opencode -m omniroute/`) | `--remote` `--api-key` `--only` `--model` `--dry-run` `--port` | Both | | `omniroute setup-cline` | Cline | `~/.cline/data/{globalState,secrets}.json` (CLI mode) + prints VS Code extension settings | `--remote` `--api-key` `--model` `--yes` `--dry-run` `--port` `--cline-dir` | Both | diff --git a/docs/guides/CODEX-CLI-CONFIGURATION.md b/docs/guides/CODEX-CLI-CONFIGURATION.md index 873fba3dec..efc1c5f0b7 100644 --- a/docs/guides/CODEX-CLI-CONFIGURATION.md +++ b/docs/guides/CODEX-CLI-CONFIGURATION.md @@ -241,7 +241,9 @@ omniroute setup-codex --only glm,kimi omniroute setup-codex --codex-home /path/to/.codex ``` -The command fetches `/v1/models`, categorises each model (thinking / good / simple / fast) and writes `~/.codex/.config.toml` for each. Idempotent — safe to re-run. +The command fetches `/v1/models`, uses tuned profiles for known models, falls back to catalog metadata for other compatible text models, and writes `~/.codex/.config.toml` for each. Idempotent — safe to re-run. + +OmniRoute can also **auto-sync** these same profile files after a successful provider model discovery/import changes the live catalog. This is **opt-in and off by default**: toggle it from the **CLI Code dashboard** ("CLI profile auto-sync" → Codex), or set `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES=true` (it also honors `CLI_ALLOW_CONFIG_WRITES`, on by default). When enabled it only writes separate `~/.codex/*.config.toml` profile files; it never changes the active/default `~/.codex/config.toml`, Codex-lb settings, auth, or provider selection. --- diff --git a/docs/guides/DOCKER_GUIDE.md b/docs/guides/DOCKER_GUIDE.md index bfbb6b1829..4e7fb6875a 100644 --- a/docs/guides/DOCKER_GUIDE.md +++ b/docs/guides/DOCKER_GUIDE.md @@ -190,7 +190,7 @@ services: - omniroute-data:/app/data environment: - PORT=20128 - # Browser-facing origin for OAuth callbacks, dashboard links, and same-origin checks. + # Browser-facing origin for OAuth callbacks, dashboard links, and generated public URLs. - NEXT_PUBLIC_BASE_URL=https://your-domain.com # Internal server-to-server URL for scheduled jobs / self-fetches. - BASE_URL=http://omniroute:20128 @@ -210,9 +210,11 @@ volumes: ``` Caddy sets the standard forwarding headers for the upstream container. OmniRoute uses -`NEXT_PUBLIC_BASE_URL` as the canonical public origin; only enable `OMNIROUTE_TRUST_PROXY` for -advanced deployments where you intentionally want OmniRoute to derive the public origin from -trusted forwarded headers instead of explicit configuration. +`NEXT_PUBLIC_BASE_URL` as the canonical public origin for OAuth callbacks and generated public +links; authenticated dashboard writes use same-origin requests plus session-bound CSRF +protection. Only enable `OMNIROUTE_TRUST_PROXY` for advanced deployments where you intentionally +want OmniRoute to derive the public origin from trusted forwarded headers instead of explicit +configuration. ## Cloudflare Quick Tunnel diff --git a/docs/guides/SETUP_GUIDE.md b/docs/guides/SETUP_GUIDE.md index 2f9a500cf8..0e6564311c 100644 --- a/docs/guides/SETUP_GUIDE.md +++ b/docs/guides/SETUP_GUIDE.md @@ -34,12 +34,11 @@ Dashboard opens at `http://localhost:20128` and API base URL is `http://localhos ### pnpm ```bash -pnpm install -g omniroute -pnpm approve-builds -g # Select all packages → approve +pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core omniroute ``` -> **pnpm users:** `pnpm approve-builds -g` is required to enable native build scripts for `better-sqlite3` and `@swc/core`. +> **pnpm users:** the `--allow-build` flag is required to enable native build scripts for `better-sqlite3` and `@swc/core`. The `pnpm approve-builds -g` command is not supported for global installs on pnpm v11. ### Arch Linux (AUR) diff --git a/docs/i18n/ar/CHANGELOG.md b/docs/i18n/ar/CHANGELOG.md index b92249b937..595fc5acd6 100644 --- a/docs/i18n/ar/CHANGELOG.md +++ b/docs/i18n/ar/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/ar/README.md b/docs/i18n/ar/README.md index f5cd3a5cbf..47c0ae199d 100644 --- a/docs/i18n/ar/README.md +++ b/docs/i18n/ar/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/az/CHANGELOG.md b/docs/i18n/az/CHANGELOG.md index d1f39b7f0c..31584a5e8a 100644 --- a/docs/i18n/az/CHANGELOG.md +++ b/docs/i18n/az/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/az/README.md b/docs/i18n/az/README.md index 454354df10..f16efdc76c 100644 --- a/docs/i18n/az/README.md +++ b/docs/i18n/az/README.md @@ -757,11 +757,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/bg/CHANGELOG.md b/docs/i18n/bg/CHANGELOG.md index d1f39b7f0c..31584a5e8a 100644 --- a/docs/i18n/bg/CHANGELOG.md +++ b/docs/i18n/bg/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/bg/README.md b/docs/i18n/bg/README.md index 77a70239f7..fc10386760 100644 --- a/docs/i18n/bg/README.md +++ b/docs/i18n/bg/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/bn/CHANGELOG.md b/docs/i18n/bn/CHANGELOG.md index 9f69b33761..ecb7c8e47d 100644 --- a/docs/i18n/bn/CHANGELOG.md +++ b/docs/i18n/bn/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/bn/README.md b/docs/i18n/bn/README.md index 16de9b3d59..9f62edcc5f 100644 --- a/docs/i18n/bn/README.md +++ b/docs/i18n/bn/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/cs/CHANGELOG.md b/docs/i18n/cs/CHANGELOG.md index 527b6939f9..b33dcd9dcb 100644 --- a/docs/i18n/cs/CHANGELOG.md +++ b/docs/i18n/cs/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/cs/README.md b/docs/i18n/cs/README.md index 68ee3620e0..21f3c6ef0e 100644 --- a/docs/i18n/cs/README.md +++ b/docs/i18n/cs/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/da/CHANGELOG.md b/docs/i18n/da/CHANGELOG.md index d0adabf3da..0525ff7242 100644 --- a/docs/i18n/da/CHANGELOG.md +++ b/docs/i18n/da/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/da/README.md b/docs/i18n/da/README.md index a3f4db41c2..729ef0985c 100644 --- a/docs/i18n/da/README.md +++ b/docs/i18n/da/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/de/CHANGELOG.md b/docs/i18n/de/CHANGELOG.md index 2560adec57..b5b2b63797 100644 --- a/docs/i18n/de/CHANGELOG.md +++ b/docs/i18n/de/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/de/README.md b/docs/i18n/de/README.md index d13ada0de0..9e18a1e463 100644 --- a/docs/i18n/de/README.md +++ b/docs/i18n/de/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/es/CHANGELOG.md b/docs/i18n/es/CHANGELOG.md index aa4daceaf2..7bd26eab9b 100644 --- a/docs/i18n/es/CHANGELOG.md +++ b/docs/i18n/es/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/es/README.md b/docs/i18n/es/README.md index 90ef0046a2..87db76afd8 100644 --- a/docs/i18n/es/README.md +++ b/docs/i18n/es/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/fa/CHANGELOG.md b/docs/i18n/fa/CHANGELOG.md index 3cfc0b5a6b..de9a2f73cd 100644 --- a/docs/i18n/fa/CHANGELOG.md +++ b/docs/i18n/fa/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/fa/README.md b/docs/i18n/fa/README.md index 0c6b008e24..dd3e697ddc 100644 --- a/docs/i18n/fa/README.md +++ b/docs/i18n/fa/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/fi/CHANGELOG.md b/docs/i18n/fi/CHANGELOG.md index cd8a8e353f..4d44cfd43a 100644 --- a/docs/i18n/fi/CHANGELOG.md +++ b/docs/i18n/fi/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/fi/README.md b/docs/i18n/fi/README.md index b309c8cc86..0bfeeb9f89 100644 --- a/docs/i18n/fi/README.md +++ b/docs/i18n/fi/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/fr/CHANGELOG.md b/docs/i18n/fr/CHANGELOG.md index 35362bff09..6d251e12b6 100644 --- a/docs/i18n/fr/CHANGELOG.md +++ b/docs/i18n/fr/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/fr/README.md b/docs/i18n/fr/README.md index 7dd2961804..1707dfb54f 100644 --- a/docs/i18n/fr/README.md +++ b/docs/i18n/fr/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/gu/CHANGELOG.md b/docs/i18n/gu/CHANGELOG.md index 51c1cc5991..b456a14e03 100644 --- a/docs/i18n/gu/CHANGELOG.md +++ b/docs/i18n/gu/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/gu/README.md b/docs/i18n/gu/README.md index 297d2d8800..4bfc39c456 100644 --- a/docs/i18n/gu/README.md +++ b/docs/i18n/gu/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/he/CHANGELOG.md b/docs/i18n/he/CHANGELOG.md index 8235aa1cfa..89d09bc81d 100644 --- a/docs/i18n/he/CHANGELOG.md +++ b/docs/i18n/he/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/he/README.md b/docs/i18n/he/README.md index e7171cd1f5..ff97439ea4 100644 --- a/docs/i18n/he/README.md +++ b/docs/i18n/he/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/hi/CHANGELOG.md b/docs/i18n/hi/CHANGELOG.md index 8195dafa29..022cc03f83 100644 --- a/docs/i18n/hi/CHANGELOG.md +++ b/docs/i18n/hi/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/hi/README.md b/docs/i18n/hi/README.md index e3e1137c1b..a1a61f0d9e 100644 --- a/docs/i18n/hi/README.md +++ b/docs/i18n/hi/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/hu/CHANGELOG.md b/docs/i18n/hu/CHANGELOG.md index 2accdab046..7e1094da11 100644 --- a/docs/i18n/hu/CHANGELOG.md +++ b/docs/i18n/hu/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/hu/README.md b/docs/i18n/hu/README.md index 1800c66be2..2a86990e2e 100644 --- a/docs/i18n/hu/README.md +++ b/docs/i18n/hu/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/id/CHANGELOG.md b/docs/i18n/id/CHANGELOG.md index 304e3c180f..2813e7ac16 100644 --- a/docs/i18n/id/CHANGELOG.md +++ b/docs/i18n/id/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/id/README.md b/docs/i18n/id/README.md index b0e108ea77..c0a4b64de9 100644 --- a/docs/i18n/id/README.md +++ b/docs/i18n/id/README.md @@ -754,11 +754,10 @@ npm install -g omniroute omniroute ``` -> **pengguna pnpm:** Jalankan `pnpm approve-builds -g` setelah instalasi untuk mengaktifkan skrip build asli yang diperlukan oleh `better-sqlite3` dan `@swc/core`: +> **pengguna pnpm:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Pilih semua paket → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/in/CHANGELOG.md b/docs/i18n/in/CHANGELOG.md index f3f84fa09c..bcfeaade5e 100644 --- a/docs/i18n/in/CHANGELOG.md +++ b/docs/i18n/in/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/in/README.md b/docs/i18n/in/README.md index 211fe26061..23ecd7d8b3 100644 --- a/docs/i18n/in/README.md +++ b/docs/i18n/in/README.md @@ -755,11 +755,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/it/CHANGELOG.md b/docs/i18n/it/CHANGELOG.md index 40f79e3604..800b252e70 100644 --- a/docs/i18n/it/CHANGELOG.md +++ b/docs/i18n/it/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/it/README.md b/docs/i18n/it/README.md index 034b3c2fed..b946ba324c 100644 --- a/docs/i18n/it/README.md +++ b/docs/i18n/it/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/ja/CHANGELOG.md b/docs/i18n/ja/CHANGELOG.md index 90dce1964c..be80401d52 100644 --- a/docs/i18n/ja/CHANGELOG.md +++ b/docs/i18n/ja/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/ja/README.md b/docs/i18n/ja/README.md index f02f0bb835..79ced09058 100644 --- a/docs/i18n/ja/README.md +++ b/docs/i18n/ja/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/ko/CHANGELOG.md b/docs/i18n/ko/CHANGELOG.md index 9790e00a1e..35281cc458 100644 --- a/docs/i18n/ko/CHANGELOG.md +++ b/docs/i18n/ko/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/ko/README.md b/docs/i18n/ko/README.md index ca55ef57c5..202648b907 100644 --- a/docs/i18n/ko/README.md +++ b/docs/i18n/ko/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/mr/CHANGELOG.md b/docs/i18n/mr/CHANGELOG.md index d0dff86c88..edde061766 100644 --- a/docs/i18n/mr/CHANGELOG.md +++ b/docs/i18n/mr/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/mr/README.md b/docs/i18n/mr/README.md index cb54e3a758..9f85717833 100644 --- a/docs/i18n/mr/README.md +++ b/docs/i18n/mr/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/ms/CHANGELOG.md b/docs/i18n/ms/CHANGELOG.md index b296700c27..c20b9f5284 100644 --- a/docs/i18n/ms/CHANGELOG.md +++ b/docs/i18n/ms/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/ms/README.md b/docs/i18n/ms/README.md index 6319ff2ee0..fe202f1078 100644 --- a/docs/i18n/ms/README.md +++ b/docs/i18n/ms/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/nl/CHANGELOG.md b/docs/i18n/nl/CHANGELOG.md index e5c182a312..3a49f71458 100644 --- a/docs/i18n/nl/CHANGELOG.md +++ b/docs/i18n/nl/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/nl/README.md b/docs/i18n/nl/README.md index 9cfb24c8ad..4c27729c75 100644 --- a/docs/i18n/nl/README.md +++ b/docs/i18n/nl/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/no/CHANGELOG.md b/docs/i18n/no/CHANGELOG.md index f9ba0c569c..aa3ecfc80f 100644 --- a/docs/i18n/no/CHANGELOG.md +++ b/docs/i18n/no/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/no/README.md b/docs/i18n/no/README.md index 86cd0157b1..681ed3f7bf 100644 --- a/docs/i18n/no/README.md +++ b/docs/i18n/no/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/phi/CHANGELOG.md b/docs/i18n/phi/CHANGELOG.md index f1cd2a45ec..3fed9c0458 100644 --- a/docs/i18n/phi/CHANGELOG.md +++ b/docs/i18n/phi/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/phi/README.md b/docs/i18n/phi/README.md index cdee7e0713..2ad45eaf9f 100644 --- a/docs/i18n/phi/README.md +++ b/docs/i18n/phi/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/pl/CHANGELOG.md b/docs/i18n/pl/CHANGELOG.md index 1df54da6d4..b2b275f538 100644 --- a/docs/i18n/pl/CHANGELOG.md +++ b/docs/i18n/pl/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/pl/README.md b/docs/i18n/pl/README.md index 9a6864d878..a329ac2494 100644 --- a/docs/i18n/pl/README.md +++ b/docs/i18n/pl/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/pt-BR/CHANGELOG.md b/docs/i18n/pt-BR/CHANGELOG.md index 37354af189..2603b1352c 100644 --- a/docs/i18n/pt-BR/CHANGELOG.md +++ b/docs/i18n/pt-BR/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/pt-BR/README.md b/docs/i18n/pt-BR/README.md index cc6147766e..844577c979 100644 --- a/docs/i18n/pt-BR/README.md +++ b/docs/i18n/pt-BR/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/pt/CHANGELOG.md b/docs/i18n/pt/CHANGELOG.md index dd96273225..794882a1bc 100644 --- a/docs/i18n/pt/CHANGELOG.md +++ b/docs/i18n/pt/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/pt/README.md b/docs/i18n/pt/README.md index 8a6c3bd5dc..b44cd0dc2f 100644 --- a/docs/i18n/pt/README.md +++ b/docs/i18n/pt/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/ro/CHANGELOG.md b/docs/i18n/ro/CHANGELOG.md index 76c0d3b7ae..442c9577ad 100644 --- a/docs/i18n/ro/CHANGELOG.md +++ b/docs/i18n/ro/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/ro/README.md b/docs/i18n/ro/README.md index 4d70312403..660237476f 100644 --- a/docs/i18n/ro/README.md +++ b/docs/i18n/ro/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/ru/CHANGELOG.md b/docs/i18n/ru/CHANGELOG.md index 28ddb1c3ee..f2bcc29907 100644 --- a/docs/i18n/ru/CHANGELOG.md +++ b/docs/i18n/ru/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/ru/README.md b/docs/i18n/ru/README.md index 9ce9cb2eeb..a26a4c9ca6 100644 --- a/docs/i18n/ru/README.md +++ b/docs/i18n/ru/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/sk/CHANGELOG.md b/docs/i18n/sk/CHANGELOG.md index 8612764e0d..b691acf9bb 100644 --- a/docs/i18n/sk/CHANGELOG.md +++ b/docs/i18n/sk/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/sk/README.md b/docs/i18n/sk/README.md index c73823d3c5..c5fcaadf5a 100644 --- a/docs/i18n/sk/README.md +++ b/docs/i18n/sk/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/sv/CHANGELOG.md b/docs/i18n/sv/CHANGELOG.md index 0038bff5d1..8cd03360d7 100644 --- a/docs/i18n/sv/CHANGELOG.md +++ b/docs/i18n/sv/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/sv/README.md b/docs/i18n/sv/README.md index 18ae0fb25b..ed3b81555b 100644 --- a/docs/i18n/sv/README.md +++ b/docs/i18n/sv/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/sw/CHANGELOG.md b/docs/i18n/sw/CHANGELOG.md index a76d1df253..f96f4493fe 100644 --- a/docs/i18n/sw/CHANGELOG.md +++ b/docs/i18n/sw/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/sw/README.md b/docs/i18n/sw/README.md index 75e7569985..d977641a34 100644 --- a/docs/i18n/sw/README.md +++ b/docs/i18n/sw/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/ta/CHANGELOG.md b/docs/i18n/ta/CHANGELOG.md index 96d9c0f353..87e0bd9ccb 100644 --- a/docs/i18n/ta/CHANGELOG.md +++ b/docs/i18n/ta/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/ta/README.md b/docs/i18n/ta/README.md index f68a8a5ebb..ae475f09e3 100644 --- a/docs/i18n/ta/README.md +++ b/docs/i18n/ta/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/te/CHANGELOG.md b/docs/i18n/te/CHANGELOG.md index 38b63d0937..6ae74a28db 100644 --- a/docs/i18n/te/CHANGELOG.md +++ b/docs/i18n/te/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/te/README.md b/docs/i18n/te/README.md index 070b6c2a30..b74ca3e14d 100644 --- a/docs/i18n/te/README.md +++ b/docs/i18n/te/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/th/CHANGELOG.md b/docs/i18n/th/CHANGELOG.md index e858cdb112..893556f19e 100644 --- a/docs/i18n/th/CHANGELOG.md +++ b/docs/i18n/th/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/th/README.md b/docs/i18n/th/README.md index 1bb83c3853..7b0f48e844 100644 --- a/docs/i18n/th/README.md +++ b/docs/i18n/th/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/tr/CHANGELOG.md b/docs/i18n/tr/CHANGELOG.md index e78b7b2dfd..e39fb48283 100644 --- a/docs/i18n/tr/CHANGELOG.md +++ b/docs/i18n/tr/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/tr/README.md b/docs/i18n/tr/README.md index bd23e93cf2..4268f9ffab 100644 --- a/docs/i18n/tr/README.md +++ b/docs/i18n/tr/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/uk-UA/CHANGELOG.md b/docs/i18n/uk-UA/CHANGELOG.md index bdd8ac052b..9a17657000 100644 --- a/docs/i18n/uk-UA/CHANGELOG.md +++ b/docs/i18n/uk-UA/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/uk-UA/README.md b/docs/i18n/uk-UA/README.md index 6c7a70891a..6a99371d5f 100644 --- a/docs/i18n/uk-UA/README.md +++ b/docs/i18n/uk-UA/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/ur/CHANGELOG.md b/docs/i18n/ur/CHANGELOG.md index 87e2f8792c..e405d76c37 100644 --- a/docs/i18n/ur/CHANGELOG.md +++ b/docs/i18n/ur/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/ur/README.md b/docs/i18n/ur/README.md index f1efaf5e6d..4351ad5a48 100644 --- a/docs/i18n/ur/README.md +++ b/docs/i18n/ur/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/vi/CHANGELOG.md b/docs/i18n/vi/CHANGELOG.md index 878000619a..ae3daff0c5 100644 --- a/docs/i18n/vi/CHANGELOG.md +++ b/docs/i18n/vi/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/vi/README.md b/docs/i18n/vi/README.md index 3736093bd9..4d15324bbd 100644 --- a/docs/i18n/vi/README.md +++ b/docs/i18n/vi/README.md @@ -756,11 +756,10 @@ npm install -g omniroute omniroute ``` -> **pnpm users:** Run `pnpm approve-builds -g` after install to enable native build scripts required by `better-sqlite3` and `@swc/core`: +> **pnpm users:** Pass `--allow-build` at install time to enable native build scripts required by `better-sqlite3` and `@swc/core` (the `approve-builds -g` command is not supported for global installs on pnpm v11): > > ```bash -> pnpm install -g omniroute -> pnpm approve-builds -g # Select all packages → approve +> pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core > omniroute > ``` diff --git a/docs/i18n/zh-CN/CHANGELOG.md b/docs/i18n/zh-CN/CHANGELOG.md index c7bae62a6f..30f8a24b68 100644 --- a/docs/i18n/zh-CN/CHANGELOG.md +++ b/docs/i18n/zh-CN/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/zh-CN/README.md b/docs/i18n/zh-CN/README.md index deb166045f..855e2bbb4e 100644 --- a/docs/i18n/zh-CN/README.md +++ b/docs/i18n/zh-CN/README.md @@ -632,7 +632,7 @@ PORT=20128 npm run dev **📦 pnpm** ```bash -pnpm install -g omniroute && pnpm approve-builds -g && omniroute +pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core && omniroute ``` **🐧 Arch Linux（AUR）** diff --git a/docs/i18n/zh-TW/CHANGELOG.md b/docs/i18n/zh-TW/CHANGELOG.md index 540e28da43..d8dcc93c70 100644 --- a/docs/i18n/zh-TW/CHANGELOG.md +++ b/docs/i18n/zh-TW/CHANGELOG.md @@ -6,6 +6,338 @@ ## [3.8.31] — 2026-06-20 +## [3.8.43] — 2026-07-02 + +### ✨ New Features + +- **usage (quota percentages + provider USD drilldown):** `@@om-usage` and the HTTP usage endpoint now report personal API-key quotas as **remaining percentages** (USD amounts stay out of the command output), provider quota remaining is scaled by the configured quota cutoff so the protected reserve reads as 0% left, and the quota dashboard regains a **provider USD cost drilldown** (`/api/usage/provider-window-costs` + `ProviderUsdCostModal`, management-auth gated). Also honors **observed provider quota resets**: a same-`resetAt` reset (usage dropping back to the reset floor) is detected and preferred over stale recorded weekly events for provider USD windows and API-key USD quotas. New `src/lib/usage/providerWindowCosts.ts`. Regression guards: `tests/unit/provider-window-costs.test.ts`, `tests/unit/internal-usage-command.test.ts`, `tests/unit/api-key-usage-limits.test.ts`, `tests/unit/lib/quota-reset-events.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **dashboard (live WS behind reverse proxy):** the live dashboard WebSocket can now be fronted by a reverse proxy or Cloudflare Tunnel via `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` (e.g. `wss://ws.my-ai.com/live-ws`). The URL is honored both at build time (env inlined into the bundle) and at **runtime** for prebuilt Docker/npm images: the `/api/v1/ws?handshake=1` handshake now echoes a lazily-read `live.publicUrl` (only `ws://`/`wss://` values are accepted; anything else is rejected to `null`), and `useLiveDashboard` resolves the URL from that handshake before connecting, falling back to the previous `ws(s)://hostname:20129` default. Also documents `LIVE_WS_ALLOWED_HOSTS` and aligns the GitLab Duo OAuth scopes line in `.env.example` with the live config (`ai_features read_user`). Regression guard: `tests/unit/live-ws-public-url.test.ts` (5). ([#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877) by [@ianriizky](https://github.com/ianriizky)) + +- **providers (CLI profile auto-sync):** opt-in toggles to auto-regenerate CLI tool profiles after a provider model sync. When enabled, a model-catalog change (re)writes that tool's profile files from the live catalog — Codex (`~/.codex/*.config.toml`) and now **Claude Code** (`~/.claude/profiles//settings.json`, via an extracted `syncClaudeProfilesFromModels` + a new `claudeProfileAutoSync.ts` mirroring the Codex path). Both are **off by default** and never touch the active/default CLI config; they are backed by the `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` / `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` feature flags (DB/dashboard override > env > default "false") and additionally gated behind the existing `CLI_ALLOW_CONFIG_WRITES` write-guard. A **"CLI profile auto-sync"** card on the CLI Code dashboard toggles each (moved from the providers dashboard in [#5778](https://github.com/diegosouzapw/OmniRoute/pull/5778) — thanks [@rdself](https://github.com/rdself)). Regression guards: `tests/unit/claude-profile-auto-sync-gate.test.ts`, `tests/unit/codex-profile-auto-sync-gate.test.ts`, `tests/unit/cli/setup-claude.test.ts` (follow-up to #5737). + +- **cli (startup banner):** the `serve` startup banner now prints the running OmniRoute version (`v3.8.x`) beneath the ASCII logo, so the active version is visible at a glance without a separate `--version` call. Regression guard: `tests/unit/cli-serve-version-banner.test.ts`. Thanks [@chirag127](https://github.com/chirag127) ([#5752](https://github.com/diegosouzapw/OmniRoute/pull/5752)). + +- **analytics (subscription cost):** flat-rate providers now show **$0** in cost analytics instead of an inflated per-token estimate. Subscription / coding-plan providers (every cookie-web provider — ChatGPT Web, grok-web, … — plus the dedicated **Minimax Coding**, **Kimi Coding**, **GLM Coding**, **Alibaba Coding Plan**, and **Xiaomi MiMo** plans) bill a flat fee, not per token, yet still carry per-token pricing rows used for estimates — so the analytics dashboard over-reported their cost. A new flat-rate classifier (`src/lib/usage/flatRateProviders.ts`) is consulted by the analytics surfaces (analytics route, usage stats, usage analytics) via an opt-in `flatRateAsZero` cost option, so those providers read $0 while **budget / quota / routing keep estimating unchanged**. Deliberately NOT zeroed: `codex`/`cx` (OmniRoute actively tracks Codex token cost — Fast-tier multipliers, GPT-5.x pricing — and Codex can be a metered account), `byteplus` (metered ModelArk), `minimax-cn` (metered China API). Regression guard: `tests/unit/flat-rate-cost-5552.test.ts`. ([#5552](https://github.com/diegosouzapw/OmniRoute/issues/5552)) + +- **mcp (RTK):** expose the RTK tool-output **learn/discover** workflow as two new MCP tools so an agent can grow the RTK filter catalog without leaving the protocol. `omniroute_rtk_discover` analyzes recently captured raw tool output (`discoverRepeatedNoise` / `suggestFilter`) and returns candidate noise patterns plus a suggested filter; `omniroute_rtk_learn` lists the captured command samples (`listRtkCommandSamples`) and resolves a command to its RTK filter id (`commandToId`). Both are read-only (scope `read:compression`), wrap the existing RTK discovery primitives (no new logic in the engine), and log to the MCP audit trail. Regression guard: `tests/unit/compression/rtk-mcp-tools.test.ts` (4). gaps v3.8.42 — T07. + +- **compression (LLM tier):** add an **opt-in, default-off LLM-tier compression engine** (`llm`) that condenses the prose of non-system messages via a pluggable chat-completion backend. It mirrors the `llmlingua` engine's contract but is **safe by construction**: the default backend is a **no-op pass-through** (the engine never mutates the payload until an operator both enables it _and_ wires a real backend via `setLlmCompressorBackend()`), it is **not** part of the default stacked pipeline, `enabled` defaults to `false`, fenced code blocks and `system` messages are never sent to the model, and every backend error **fails open** (the original segment/body is kept, never thrown). A `minTokens` floor skips small prompts. The real production backend is intentionally a VPS-validated follow-up (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. New `open-sse/services/compression/engines/llm/index.ts`. Regression guard: `tests/unit/compression/llm-compressor-engine.test.ts` (8). gaps v3.8.42 — T05/C3. + +- **memory (typed decay):** add **opt-in typed memory decay** (TV6) so the conversational memory store stops accumulating stale `episodic` noise. Each injected memory now tracks an `access_count` + `last_accessed_at` (always-on, non-destructive telemetry; migration `111_memory_typed_decay`), and an **opt-in, default-off** sweep (`MEMORY_TYPED_DECAY_ENABLED`, default `false`) deletes memories that are past a **per-type TTL** and not immune. Only `episodic` decays by default (30d, env-tunable); `factual`/`procedural`/`semantic` are immune, and any memory accessed `>= 3` times earns access immunity (mirroring "guardrail/convention/decision never decay"). The decay clock re-bases on the last access, so used memories survive. Deletions reuse `deleteMemory` (SQLite + sqlite-vec + Qdrant stay in sync) and fail open; an optional periodic sweep is doubly opt-in (also needs `MEMORY_TYPED_DECAY_SWEEP_INTERVAL>0`). With the flag off nothing is ever deleted (Rule #20 spirit). New `src/lib/memory/typedDecay.ts`. Regression guard: `tests/unit/memory/typed-decay.test.ts` (15). gaps v3.8.42 — T10/TV6. + +- **dashboard (combos):** the named-combos editor now lets you **drag to reorder** the stacked-compression pipeline instead of only editing fixed-position steps. A new pure model (`src/shared/components/compression/compressionPipelineModel.ts`) owns add/remove/move/update with the engine→intensity invariant and a never-empty guarantee, and a `@dnd-kit/sortable` editor (`CompressionPipelineEditor.tsx`, matching the sidebar reorder pattern) replaces the inline list in `CompressionCombosPageClient`. Order persists through the existing combos endpoint. Regression guards: `tests/unit/compression-pipeline-model.test.ts` (11) + `tests/unit/ui/compression-pipeline-editor.test.tsx` (4). A dedicated `tests/e2e/compression-studio.spec.ts` (Tela A render + tab switch) closes the studios e2e gap the combo-live spec did not cover. gaps v3.8.42 — T06 + T03. + +- **compression (pipeline):** add an **opt-in, default-off per-engine circuit-breaker** to the stacked compression pipeline (T02). When an engine throws repeatedly **across requests**, its breaker opens and the stacked loops skip that engine (keeping the body verbatim for that step — fail-open) for a cooldown, then probe once (lazy half-open); success closes it, a failed probe re-opens it. This is distinct from the provider circuit-breaker (`src/shared/utils/circuitBreaker.ts`, provider-scoped + DB-persisted) — the new `pipelineEngineBreaker.ts` is engine-scoped, process-local, and adds zero DB/IO on the hot path. It composes with the existing per-request TV1 bail-out (which skips within a single request); the breaker adds cross-request memory. **Default off** (`COMPRESSION_PIPELINE_BREAKER_ENABLED=false`) → byte-identical to the pre-breaker pipeline (a throwing engine still propagates unless TV1 is separately enabled). Configurable per-call, per-`CompressionConfig`, or via env (`_THRESHOLD`/`_COOLDOWN_MS`). Regression guard: `tests/unit/compression/pipeline-circuit-breaker.test.ts` (9, incl. a throwing-engine integration); existing strategySelector/bail-out suites stay green. gaps v3.8.42 — T02 (2.2). + +- **compression (CCR):** the CCR retrieval-feedback (H8) is now **graduated** instead of a binary cliff. Previously a block retrieved `>= 3` times was flagged do-not-compress and everything below that stayed fully compressible. Now each prior retrieval raises a block's **effective `minChars` linearly** (`effectiveMinChars`), so frequently-retrieved content is compressed progressively less; the `>= 3` exclusion is preserved (as `Infinity`). The ramp is controlled by a `retrievalRampFactor` (default `2`, per-combo config or `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`); `1` reproduces the exact legacy binary behavior. Per-`(principal, hash)` isolation is unchanged. Regression guard: `tests/unit/compression/ccr-retrieval-ramp.test.ts` (12); existing CCR suites (51) stay green. gaps v3.8.42 — T08/H8. + +- **compression (cache-aware):** add an **opt-in, default-off usage-observed prefix freeze** (H5). The cache-aware guard previously preserved the system prompt only for providers a _static_ heuristic recognized as caching. It now also **learns** which system prompts actually recur: once a system prompt has been observed `>=` a threshold across requests, it is treated as a stable cacheable prefix and preserved from compression **even for providers the static check misses** — recovering prompt-cache hits that a prefix-compressing mode would otherwise bust. Content-addressed by a hash of the system prompt (OpenAI / Claude / Gemini shapes), in-memory + bounded, zero DB/IO; a "freeze" only _preserves_ the prefix, so it never mutates a payload. Default OFF (`COMPRESSION_PREFIX_FREEZE_ENABLED`, threshold `_THRESHOLD`); respects the `never` preserve-mode (never freezes). New `open-sse/services/compression/prefixFreeze.ts`, wired into `resolveCacheAwareConfig`. Regression guard: `tests/unit/compression/prefix-freeze.test.ts` (10); 44 existing cache-aware / preserve-mode tests stay green. gaps v3.8.42 — T08/H5. + +- **compression (read-lifecycle):** add a new **opt-in, default-off `read-lifecycle` engine** (H7) that collapses **stale/superseded file-Read tool results**. In agentic conversations the same file is Read repeatedly; an earlier Read becomes stale once the same path is **re-read** (superseded by a newer view) or **modified** by a later Write/Edit. The engine replaces those earlier Read results with a short stub — keeping only the current (last, un-superseded) Read intact — recovering the tokens the model no longer needs. Unlike `session-dedup` (identical-content) or `ccr` (reversible markers), this is semantic + **lossy**, so it is opt-in (`enabled` defaults `false`). Conservative by construction: matches only well-known Read/Write tool names, compares exact paths, collapses a Read only when a strictly-later invocation touches the same path, and **fail-opens** on any unexpected shape. Supports both the **Anthropic** (`tool_use`/`tool_result`) and **OpenAI** (`tool_calls` + `role:"tool"`) shapes. New `open-sse/services/compression/engines/readLifecycle/index.ts`. Regression guard: `tests/unit/compression/read-lifecycle.test.ts` (10). gaps v3.8.42 — T08/H7. + +- **observability (correlation IDs):** requests now carry a correlation id threaded through logs so a single request can be traced end-to-end across the pipeline. ([#5834](https://github.com/diegosouzapw/OmniRoute/pull/5834) — thanks @hartmark) + +- **cli (startup banner — boot time):** the `serve` ready banner now shows how long startup took, so slow-boot conditions are visible at a glance. ([#5799](https://github.com/diegosouzapw/OmniRoute/pull/5799) — thanks @ishatiwari21) + +- **api (quota-policy bypass scope):** add an **opt-in** API-key provider quota-policy bypass scope, so a designated key can be exempted from provider quota enforcement without disabling quotas globally. ([#5731](https://github.com/diegosouzapw/OmniRoute/pull/5731) — thanks @Witroch4) + +- **providers (Ollama local):** add a first-class **Ollama** local-provider card to the providers dashboard so the local LLM runtime can be configured like any other provider. ([#5712](https://github.com/diegosouzapw/OmniRoute/pull/5712) — thanks @diegosouzapw) + +- **codex (fallback profiles):** generate fallback CLI profiles for Codex-compatible models so compatible models get a usable profile automatically. ([#5701](https://github.com/diegosouzapw/OmniRoute/pull/5701) — thanks @skyzea1) + +- **api (response-body validation + failover):** add a **configurable response-body validation** step that can fail a target over to the next candidate when the upstream returns a structurally-invalid body (routing/#4985). ([#5684](https://github.com/diegosouzapw/OmniRoute/pull/5684) — thanks @diegosouzapw) + +- **providers (SenseNova):** complete the SenseNova free **Token Plan** — chat completions plus **Text-to-Image** (ported from 9router#2233). ([#5679](https://github.com/diegosouzapw/OmniRoute/pull/5679) — thanks @diegosouzapw) + +- **db (self-correcting context windows):** add **self-correcting model context-window overrides** so a model whose advertised context length is wrong is corrected automatically (models/#5004). ([#5667](https://github.com/diegosouzapw/OmniRoute/pull/5667) — thanks @diegosouzapw) + +- **routing (latency strategy):** optimize the latency routing strategy using observed per-target performance metrics for better candidate selection. ([#5629](https://github.com/diegosouzapw/OmniRoute/pull/5629) — thanks @KooshaPari) + +- **compression (preserveSystemPrompt mode):** add a `preserveSystemPrompt` mode enum (`always` | `whenNoCache` | `never`) with legacy back-compat, giving operators explicit control over when the system prompt is protected from compression (T05/C5). ([#5653](https://github.com/diegosouzapw/OmniRoute/pull/5653) — thanks @diegosouzapw) + +- **commandCode (vision):** add multimodal **image** support for Command Code vision models. ([#5557](https://github.com/diegosouzapw/OmniRoute/pull/5557) — thanks @Stazyu) + +- **compression (read-lifecycle engine):** T08/H7 (2.5) — an opt-in read-lifecycle engine that collapses superseded file reads so stale earlier reads of the same file are pruned from the context. ([#5754](https://github.com/diegosouzapw/OmniRoute/pull/5754) — thanks @diegosouzapw) + +- **compression (usage-observed prefix freeze):** T08/H5 (2.4) — opt-in prefix freeze driven by observed usage, keeping a stable cached prefix from being rewritten by downstream engines. ([#5744](https://github.com/diegosouzapw/OmniRoute/pull/5744) — thanks @diegosouzapw) + +- **compression (CCR retrieval-feedback ramp):** T08/H8 (2.3) — a graduated Context-Compression-Ratio retrieval-feedback ramp that tunes compression aggressiveness from retrieval signals. ([#5739](https://github.com/diegosouzapw/OmniRoute/pull/5739) — thanks @diegosouzapw) + +- **compression (per-engine circuit breaker):** T02 — an opt-in per-engine pipeline circuit-breaker that disables a misbehaving compression engine without failing the whole pipeline. ([#5735](https://github.com/diegosouzapw/OmniRoute/pull/5735) — thanks @diegosouzapw) + +- **compression (LLM-tier engine):** T05/C3 — an opt-in LLM-tier compression engine that uses a model pass for higher-ratio semantic compression. ([#5702](https://github.com/diegosouzapw/OmniRoute/pull/5702) — thanks @diegosouzapw) + +- **dashboard (compression pipeline editor):** T06/T03 — a drag-to-reorder compression pipeline editor plus a compression-studio e2e flow. ([#5727](https://github.com/diegosouzapw/OmniRoute/pull/5727) — thanks @diegosouzapw) + +- **memory (typed decay):** T10/TV6 — opt-in typed memory decay so aged, low-value memories fade on a per-type schedule. ([#5723](https://github.com/diegosouzapw/OmniRoute/pull/5723) — thanks @diegosouzapw) + +- **mcp (RTK tools):** T07 — expose the RTK learn/discover capabilities as first-class MCP tools. ([#5691](https://github.com/diegosouzapw/OmniRoute/pull/5691) — thanks @diegosouzapw) + +- **providers (CLI profile auto-sync):** opt-in CLI profile auto-sync toggles, including Claude Code auto-sync, so generated CLI profiles can track provider changes automatically. ([#5755](https://github.com/diegosouzapw/OmniRoute/pull/5755) — thanks @diegosouzapw) + +### 🔧 Bug Fixes + +- **fix(opencode):** stop fabricating `User-Agent: opencode/local` and `x-opencode-client: cli` headers when the client sends none — the executor-dedup refactor ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720)) accidentally re-introduced header fabrication, violating the forward-only contract (inventing opencode-internal values risks upstream rejection). Restored to forward-only: those headers are emitted only when a real client source is present. Regression guard: `tests/unit/opencode-executor.test.ts`. (thanks @diegosouzapw) + +- **fix(executors):** `resolveEffectiveKey` returns `undefined` (not `""`) when no API key is present — a type-coercion cleanup ([#5798](https://github.com/diegosouzapw/OmniRoute/pull/5798)) changed `apiKey ?? ""` to satisfy the typechecker, silently mutating auth-key resolution semantics. Widened the return type to `string | undefined` and reverted the coercion so OAuth-only credentials resolve correctly. Regression guard: `tests/unit/refactor-buildHeaders-preamble.test.ts`. (thanks @diegosouzapw) + +- **fix(translator):** restore the terminal `message_delta` + `message_stop` on Responses→Claude streams — the doubled-tool-args dedup ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828)) guarded the finish handler on the shared `state.finishReason`, which the openai-responses→openai leg sets first in the hub path, so the openai→claude leg dropped its terminal events and the stream ended after `content_block_delta`. The dedup now uses a dedicated `state.claudeFinishEmitted` flag. Regression guard: `tests/unit/claude-code-rendering-fixes.test.ts`. (thanks @diegosouzapw) + +- **fix(pricing):** add the Kiro `claude-sonnet-5` pricing row so the newly-catalogued model ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796)) no longer reports `$0.00` usage. Regression guard: `tests/unit/catalog-updates-v3x.test.ts`. (thanks @diegosouzapw) + +- **fix(github):** keep Copilot access-token sessions active. GitHub Copilot device-flow accounts can hold a GitHub access token plus a short-lived Copilot token **without** a refresh token; the proactive health check treated that as terminal `no_refresh_token` and marked the connection expired minutes after login. The health check now keeps those sessions active, clears stale `no_refresh_token` state, and refreshes the Copilot sub-token when needed. Regression guard: `tests/unit/token-health-no-refresh-token-expired-5326.test.ts`. Extracted from [#5863](https://github.com/diegosouzapw/OmniRoute/pull/5863) by [@Witroch4](https://github.com/Witroch4). + +- **fix(kiro):** bound the Claude model-id dash→dot normalization to a 1–2 digit minor so date-suffixed ids (e.g. claude-opus-4-20250514) are no longer corrupted. (thanks @voravitl) + +- **fix(usage):** preserve (bounded) tool definitions in request logs even when the request body is truncated, so the request-details view can still show available tools. (thanks @noir017) + +- **fix(providers):** route OpenAI responses-only models to `/v1/responses` instead of 404ing on `/v1/chat/completions`. The curated `gpt-5.5-pro` / `gpt-5.4-pro` entries never worked (OpenAI only serves `*-pro` reasoning models via the Responses API), and "Test all models" surfaced the same 404s. The registry entries now carry `targetFormat: "openai-responses"` (reusing the existing per-model translation plumbing shared with `gh`/`codex`), `DefaultExecutor.buildUrl` swaps the `openai` endpoint to `/responses` in lockstep (honoring custom base URLs), and a `-pro` suffix heuristic covers dynamically-synced ids such as `o1-pro` / `gpt-5.2-pro` (same spirit as the gh executor's `/codex/i` routing, 9router#102). Legacy completions-only ids (e.g. `gpt-3.5-turbo-instruct`) are out of scope — they are not in the catalog and OmniRoute has no legacy `/v1/completions` upstream. Regression guard: `tests/unit/openai-responses-only-models-5842.test.ts` (8). Thanks [@maikokan](https://github.com/maikokan). ([#5842](https://github.com/diegosouzapw/OmniRoute/issues/5842)) + +- **fix(image):** keep bare codex image aliases (e.g. `gpt-5.5`) resolving to the codex image pipeline even when a combo shares the same name. A chat combo named `gpt-5.5` used to shadow the bare image alias in `resolveImageRouteModel`, hijacking `/v1/images/*` requests to a chat target (regression path adjacent to [#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)); codex bare models are now reserved before bare-combo resolution, while non-codex aliases (e.g. `gpt-image-2`) remain user-shadowable (#3214/#3215 behavior preserved). Regression guard: `tests/unit/image-routes-combo-edits-3214-3215.test.ts` (9). ([#5902](https://github.com/diegosouzapw/OmniRoute/pull/5902) by [@KooshaPari](https://github.com/KooshaPari)) + +- **fix(ci):** re-green the `release/v3.8.43` fast-gates queue — every PR→release was inheriting base-reds ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)). Five distinct blockers cleared: (1) stale `modelContextOverrides` entry in the `check:db-rules` intentionally-internal allowlist ([#5827](https://github.com/diegosouzapw/OmniRoute/pull/5827) allowlisted it while the [#5609](https://github.com/diegosouzapw/OmniRoute/issues/5609) fix re-exported it from `localDb.ts`; the re-export stays, the obsolete entry goes, classification guard re-pinned to 33); (2) `LIVE_WS_ALLOWED_HOSTS` / `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` documented in `docs/reference/ENVIRONMENT.md` (env/docs contract, from [#5877](https://github.com/diegosouzapw/OmniRoute/pull/5877)); (3) the Router Backends ADR's references to the not-yet-merged registry ([#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868)) marked as landing-with-PR so `check:fabricated-docs --strict` passes; (4) `antigravity-429-quota-tdd` + `middleware-header-strip-5849` added to stryker `tap.testFiles` (`check:mutation-test-coverage`); (5) file-size / complexity / cognitive-complexity ratchets rebaselined with justification notes — all drift measured identical on the pristine tip and this PR (net-zero). Regression guard: `tests/unit/check-db-rules-classification.test.ts`. ([#5798](https://github.com/diegosouzapw/OmniRoute/issues/5798)) + +- **providers (codex image auto-routing regression):** an unprefixed `gpt-5.5` request from a codex-only setup (no OpenAI connection) now correctly infers the `codex` provider again — the OpenAI static-catalog short-circuit in `resolveModelByProviderInference` was preempting the codex-preference block, so `gpt-5.5` (added to the OpenAI catalog) stopped auto-routing to Codex image generation. Users with an active OpenAI connection are unaffected (OpenAI stays default). Regression guard: `tests/unit/codex-gpt55-routing-5887.test.ts`. ([#5887](https://github.com/diegosouzapw/OmniRoute/issues/5887)) + +- **api (proxy header hygiene):** upstream `x-middleware-*` control headers (emitted by providers hosted behind Next.js, e.g. synthetic.new) are now stripped from proxied responses instead of forwarded verbatim — forwarding `x-middleware-rewrite` made Next 16 throw `NextResponse.rewrite() was used in a app route handler` and return 500 despite a successful upstream call. Applies to both streaming and JSON paths. Regression guard: `tests/unit/middleware-header-strip-5849.test.ts`. ([#5849](https://github.com/diegosouzapw/OmniRoute/issues/5849)) + +- **docs (pnpm global install):** replaced the unsupported `pnpm approve-builds -g` step with the install-time `pnpm add -g omniroute@latest --allow-build=better-sqlite3` flag across README + Setup Guide (and i18n mirrors), fixing native-build approval for pnpm v11 global installs. ([#5554](https://github.com/diegosouzapw/OmniRoute/issues/5554)) + +- **dashboard (token badge):** the red "Token Expired" connection badge no longer flashes for OAuth refresh-capable providers (Antigravity/Gemini) whose access token merely lapsed but is auto-refreshed — it now shows only when the connection is terminally expired (`testStatus === "expired"`). Continuation of #5326. Regression guard: `tests/unit/ui/connection-row-token-badge-5836.test.tsx`. ([#5836](https://github.com/diegosouzapw/OmniRoute/issues/5836)) + +- **db (auto backup toggle):** full pre-write SQLite backups now honor the persisted `backup.autoBackupEnabled` dashboard setting — previously only the `DISABLE_SQLITE_AUTO_BACKUP` env var was checked, so disabling auto-backup in the UI had no effect and ~70MB pre-write snapshots kept firing. Manual and pre-restore backups still always run. Regression guard: `tests/unit/db-backup-autobackup-setting-5871.test.ts`. ([#5871](https://github.com/diegosouzapw/OmniRoute/issues/5871)) + +- **providers (auto/ routing for custom providers):** custom OpenAI-/Anthropic-compatible providers (dynamic `*-compatible-*` connection IDs) are no longer excluded from `auto/` routing — the Auto-Combo virtual factory previously skipped any connection whose provider was absent from the static registry. It now falls back to the connection's `defaultModel`. Regression guard: `tests/unit/auto-custom-provider-5873.test.ts`. ([#5873](https://github.com/diegosouzapw/OmniRoute/issues/5873)) + +- **middleware (hook sandbox):** operator-authored pre-request hook code now runs inside a hardened Node `vm` sandbox (minimal context, no ambient globals/`process.env`, execution timeout, no `require`) instead of `new Function()` in the main process — closing the Hard Rule #3 / SonarCloud S1523 exposure. Regression guard: `tests/unit/middleware-hook-sandbox-5872.test.ts`. ([#5872](https://github.com/diegosouzapw/OmniRoute/issues/5872)) + +- **mcp-server (auth forwarding):** the per-caller MCP identity forwarded via `withMcpHttpAuthContext` now wins over the static `OMNIROUTE_API_KEY` env fallback in the internal-fetch helpers (`apiFetch`, `omniRouteFetch`) — previously the env key was spread after the forwarded headers and clobbered the caller's `Authorization`. Regression guard: `open-sse/mcp-server/__tests__/httpAuthContext.test.ts`. ([#5819](https://github.com/diegosouzapw/OmniRoute/issues/5819)) + +- **dashboard (Modal provider — two-field auth):** the Modal provider connection form now exposes **two fields — Token ID + Token Secret —** instead of a single API-key input, since Modal authenticates with `Authorization: Bearer :`. The dashboard combines the two fields into the `id:secret` credential before saving (`combineModalCredential`, trims both parts), while a value pasted in the legacy single-field format keeps working verbatim (empty secret → passthrough), so existing saved connections need no migration; the key-help link points at Modal's token settings. Regression guard: `tests/unit/modal-credential-combine.test.ts` (5). ([#5881](https://github.com/diegosouzapw/OmniRoute/pull/5881), closes [#5446](https://github.com/diegosouzapw/OmniRoute/issues/5446)) Follow-up: the **Validation Model Id** field is now pre-filled for Modal with the same model the server-side validator probes (`Qwen/Qwen3-4B-Thinking-2507-FP8`, shared via `MODAL_DEFAULT_VALIDATION_MODEL_ID` in `src/shared/constants/modal.ts`), closing the last checklist item of #5446. Regression guard: `tests/unit/modal-validation-model-prefill.test.ts`. + +- **api (chat completions — early SSE keepalive gate):** the `/v1/chat/completions` route wrapped the response in the early-stream keepalive whenever `stream` was not explicitly `false`, so a client that omitted `stream` and asked for JSON (`Accept: application/json`) could receive premature SSE framing. The keepalive wrapper is now gated on an explicit `stream: true` in the body or an Accept header that forces SSE (`acceptHeaderForcesStream`); the parsed body is passed to the chat handler untouched, so the actual stream/JSON framing stays decided by `chatCore`/`resolveStreamFlag` — preserving OmniRoute's legacy streaming default when `stream` is omitted and the per-key `streamDefaultMode: "json"` opt-in. Regression guard: `tests/unit/chat-combo-live-test.test.ts` ("returns JSON without early SSE framing when stream is omitted and Accept is application/json"). ([#5866](https://github.com/diegosouzapw/OmniRoute/pull/5866) by [@rdself](https://github.com/rdself)) + +- **fix(github):** drop a trailing assistant prefill before dispatching to GitHub Copilot chat to avoid 400 errors. (thanks @baslr) + +- **fix(oauth):** prevent cross-IdP account overwrites by disambiguating OAuth connections on `username` when present, not email alone. (thanks @KunN-21) + +- **fix(mitm):** best-effort revert privileged /etc/hosts entries on exit when a sudo password is cached, instead of always leaving orphaned state. (thanks @manhdzzz) + +- **providers (Kiro — Claude Sonnet 5):** the Kiro provider's model catalog was missing `claude-sonnet-5`, so the model could not be selected or routed even on accounts that already had access to it ("claude-sonnet-5 is not supported"). Added the model to the Kiro registry (`open-sse/config/providers/registry/kiro/index.ts`) as a 1M-context / 128K-output Claude model, mirroring the existing Claude entries; the registry `models[]` feeds both the model selector and the live CodeWhisperer `ListAvailableModels` fallback, so the model is now selectable and routable. Regression guard: `tests/unit/kiro-claude-sonnet-5-2267.test.ts`. (thanks [@openbioinfo](https://github.com/openbioinfo)) + +- **settings (model aliases — self-heal after restart):** the Settings → Routing page showed "No exact-match aliases configured" after a server restart even though the aliases were persisted in the DB. Aliases are held in a module-local `_customAliases` map in `modelDeprecation.ts` that the boot path hydrates, but Next.js compiles the app-route module graph separately from the startup graph (the same webpack chunk-splitting class as #5312), so the `GET /api/settings/model-aliases` handler read a different, un-hydrated copy. The handler now self-heals: when its in-memory alias map is empty it reads `settings.modelAliases` from the DB (via the existing `getSettings()` db module — no raw SQL in the route) and repopulates the map, so the UI reflects the persisted aliases on the first GET after a restart. Follow-up: the root cause is now also fixed — the `_customAliases` store in `modelDeprecation.ts` is backed by `globalThis` (key `__omniroute_customAliases__`), so the startup and app-route module graphs share **one** store and the route reads the boot-hydrated aliases directly (the DB self-heal remains as a harmless fallback), mirroring the same `globalThis` singleton pattern already applied to `thinkingBudget.ts`/`backgroundTaskDetector.ts` (#5312). Regression guards: `tests/unit/model-aliases-settings-route-selfheal.test.ts` + `tests/unit/model-aliases-globalthis-5777.test.ts`. ([#5777](https://github.com/diegosouzapw/OmniRoute/pull/5777) — thanks [@jleonar2](https://github.com/jleonar2)) + +- **providers (grok-cli token auto-refresh):** grok-cli OAuth tokens were never proactively refreshed before their real expiry. `mapTokens` hardcoded `expiresIn: 21600` (6 h) regardless of the token's actual lifetime, so the persisted `expiresAt` was always "now + 6 h" and the proactive `tokenHealthCheck` sweep (refresh when `expiresAt - now < 5 min`) fired 6 h after import instead of shortly before the token really expired. `mapTokens` now computes `expiresIn` from the authoritative `expires_at` field in `~/.grok/auth.json` (ISO → epoch-seconds) with a fallback to the JWT `exp` claim (payload-only decode, no signature trust); the hardcoded `21600` is kept only when neither is present. An already-expired token (real `expires_at`/`exp` in the past) is now clamped to a positive `expiresIn` via `Math.max(1, …)`, so the import route stores a near-future `expiresAt` and AutoCombo refreshes the connection instead of reading a past date and excluding it outright. Regression guards: 5 cases in `tests/unit/grok-cli-oauth.test.ts` (JWT `exp`, JSON `expires_at`, the `21600` fallback, and the two expired-token clamps). ([#5775](https://github.com/diegosouzapw/OmniRoute/pull/5775) — thanks [@Chewji9875](https://github.com/Chewji9875)) + +- **compression (CCR retrieve via MCP HTTP):** the `omniroute_ccr_retrieve` MCP tool returned `"CCR block not found"` for blocks stored earlier in the **same** session when called over the MCP HTTP transports (SSE / Streamable HTTP), e.g. from OpenCode in a Docker deployment. Compression stores each block keyed by the API-key principal (`String(apiKeyInfo.id)`), but the tool resolved the caller via `extra.authInfo.clientId` — which the MCP SDK never populates for API-key auth — so it fell back to `"anonymous"` and the compound store-key never matched. The retrieve tool now resolves the caller's API-key id from the MCP HTTP auth context (`httpAuthContext`) using the **same** `getApiKeyMetadata` lookup used at storage time, so retrieval matches storage. Cross-tenant IDOR isolation is preserved: a different key resolves to a different id → miss; no key → the anonymous bucket only. Regression guard: `tests/unit/compression/ccr-mcp-principal-5649.test.ts` (extraction, distinct-principal isolation, fail-closed, end-to-end store→retrieve). ([#5649](https://github.com/diegosouzapw/OmniRoute/issues/5649)) + +- **compression (context-editing telemetry):** streaming responses now record Context Editing savings. Anthropic surfaces `context_management.applied_edits[]` on the final `message_delta` snapshot of an SSE stream, but the streaming reconstruction (`buildStreamSummaryFromEvents` → Claude branch) dropped `context_management` entirely **and** no telemetry hook was wired into the streaming finalizer — so the delegated server-side context-clear savings (`cleared_input_tokens` / `cleared_tool_uses`) surfaced under engine `context-editing` in compression analytics **only for non-streaming responses**. The collector now preserves `context_management` from the final snapshot (last-writer-wins), and `onStreamComplete` mirrors the non-streaming `recordContextEditingTelemetryHook` (best-effort, Claude-only, HTTP 200 only). Purely additive telemetry — no payload mutation, no new env flag, no behavior change when the stream carries no `context_management`. Regression guard: `tests/unit/context-editing-streaming-telemetry.test.ts` (3). gaps v3.8.42 — T01 (5.1). + +- **proxy (relay test diagnostics):** the Proxy Pool "Test" button showed a bare "failed" with **nothing in the server logs** when a **relay** (Vercel / Deno / Cloudflare) _responded_ with a non-200 — e.g. a `401` from an auth-token mismatch after a `STORAGE_ENCRYPTION_KEY` rotation. The relay success-path response set `success: false` but carried no `error` field, so the dashboard had no reason to show and the server logged nothing. The test now returns an actionable `error` (the HTTP status, plus an auth/encryption-key hint on `401`/`403`) and logs the failure server-side; the SOCKS5/HTTP proxy path now logs its failures too. Shaping extracted to `buildRelayTestResult` with a regression guard (`tests/unit/proxy-relay-test-error-5716.test.ts`). Note: this surfaces _why_ a relay fails — it does not repair a genuinely broken/misconfigured relay. ([#5716](https://github.com/diegosouzapw/OmniRoute/issues/5716)) + +- **fix(dashboard):** add error boundaries for the Combos and MITM Proxy pages so a render error shows a recoverable fallback instead of a blank page. (thanks @wahyuzero) + +- **providers (onboarding wizard — unsupported validation):** adding a provider whose credentials have **no live validator** (LMArena, PiAPI, …) failed silently in the Add-Provider wizard. The `/api/providers/validate` endpoint returns `HTTP 400 + { unsupported: true }` for these (#5565/#5567), but the wizard's `validateOnboardingApiKey` ran it through `expectOk`, which threw on the non-200 — so the flow jumped to the error step and the connection was **never created**. The wizard now treats `unsupported: true` as a non-blocking "can't verify" and proceeds to save, mirroring `AddApiKeyModal`. Regression guard added to `tests/unit/provider-onboarding-wizard.test.ts`. (related to [#5692](https://github.com/diegosouzapw/OmniRoute/issues/5692)) + +- **dashboard (Quick Start step 1):** the Quick Start "Create API key" step told users to "Go to **Endpoint** → Registered Keys" and linked to `/dashboard/endpoint`, but API keys are created on the **API Manager** page (`/dashboard/api-manager`, sidebar "API Keys") — the Endpoint page has no "Registered Keys" section, so users followed the link and could not find where to create a key. Step 1 now reads "Go to **API Keys**" and links to `/dashboard/api-manager`. Regression guard: `tests/unit/ui/quick-start-api-keys-link-5695.test.ts`. ([#5695](https://github.com/diegosouzapw/OmniRoute/issues/5695)) + +- **providers (DashScope/Alibaba setup link):** the "Get API key" link for the **Alibaba** and **Alibaba (China)** providers pointed at the bare API host (`dashscope-intl.aliyuncs.com` / `dashscope.aliyuncs.com`), which returns **404** in a browser — API hostnames have no homepage. Repointed to the consoles where keys are actually issued: `bailian.console.alibabacloud.com` (international) and `dashscope.console.aliyun.com` (China). Same class as #5572/#5574/#5576; regression guard added to `tests/unit/provider-setup-links-5572.test.ts`. ([#5665](https://github.com/diegosouzapw/OmniRoute/issues/5665)) + +- **thinking / runtime-config (module-graph fix):** operator-configured proxy settings that are hydrated at **boot** but read **per-request** were silently ignored in production. Next.js compiles `instrumentation.ts` (boot hydration via `applyRuntimeSettings` / restore hooks) as a **separate webpack module graph** from the app-route / open-sse executors, so a module-local `let _config` singleton is **duplicated** — the boot copy is hydrated but the request path reads a different, un-hydrated copy. Live VPS validation proved the Thinking-Budget hydration ran to completion at boot yet `base.ts` still saw the `passthrough` default (this is why #5312 fix A stayed broken even after the boot-wiring fix). Fixed by backing the singletons with `globalThis` (the pattern `systemPrompt.ts` already uses for the Global System Prompt, #2470), so all module-graph copies share one instance: **`thinkingBudget.ts`** (the dashboard Thinking-Budget mode now reaches the executor), **`backgroundTaskDetector.ts`** (the opt-in background-model degradation now actually fires on requests), and **`systemTransforms.ts`** (operator pipeline overrides now reach the request path). `payloadRules.ts` was already safe (it lazily self-loads from the DB per request, #2986). Regression guards: `tests/unit/thinking-budget-globalthis-5312.test.ts` + `tests/unit/runtime-config-globalthis-5312.test.ts` (assert globalThis-backed sharing; a module-local `let` fails them). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **thinking (Claude OAuth):** restore the proxy-level **Thinking-Budget** config on startup. The dashboard mode (`auto`/`custom`/`adaptive`) is persisted under `settings.thinkingBudget`, but the boot-time hydration (`hydrateThinkingBudgetConfig`) was only wired into `src/server-init.ts` — an **unused module that never runs in production** — so the operator's choice silently reverted to the `passthrough` default on every restart (#5312 fix A was non-functional, even though its direct unit test passed). The hydration now runs in the real boot path (`src/instrumentation-node.ts`), alongside the Global System Prompt restore. Surfaced by live Anthropic-OAuth validation on the VPS. Regression guard: `tests/unit/thinking-budget-boot-wiring-5312.test.ts` (asserts the production boot module calls the hydration, not just the function in isolation). ([#5312](https://github.com/diegosouzapw/OmniRoute/issues/5312)) + +- **translator/chatcore (hardening):** re-apply two defensive review-fixes that were dropped in a branch rebuild before #5661 / #5662 landed. (1) `mergeConsecutiveSameRoleContents` (OpenAI→Gemini) now shallow-copies each entry and its `parts` array instead of pushing the input reference, so the consecutive-same-role merge never mutates the caller's objects. (2) `defaultClaudeToolType` (Claude tool defaults) now passes any non-object array entry (`null` / primitive) through unchanged instead of spreading it into a fabricated `{ type: "custom", … }` tool. No behavior change on real payloads (Gemini contents are freshly built; Claude tools are always objects); both properties are now locked by regression tests in `tests/unit/translator-gemini-consecutive-role-2191.test.ts` and `tests/unit/claude-tool-type-default-2195.test.ts`. + +- **providers (grok-cli):** truncate the tool list when it exceeds a provider's hard limit, so grok-cli (`cli-chat-proxy.grok.com`, max 200 tools) no longer rejects requests with `Maximum tools limit reached`. Adds a proactive `PROVIDER_TOOL_LIMITS` map (`grok-cli: 200`, consulted before the reactive cache), a corrected limit-parsing regex that captures the stated maximum (`200`) instead of the supplied count (`427`), and removes the broken `< MAX_TOOLS_LIMIT` truncation gate so truncation now fires whenever `tools.length` exceeds the effective limit. Regression guard: `tests/unit/tool-limit-detector.test.ts`. ([#5563](https://github.com/diegosouzapw/OmniRoute/pull/5563) — thanks @Chewji9875) + +- **resilience (antigravity):** record model lockout for Antigravity `429 rate_limit_exceeded` errors. Antigravity's `"Resource has been exhausted (e.g. check quota)."` text was matched by overly broad `QUOTA_PATTERNS` and misclassified as `QUOTA_EXHAUSTED`, so the combo retry path was skipped (`providerExhausted`) and the model was never cooled down. Classification now prefers the structured error code — `classifyErrorText(structuredError?.code || errorText)` — so a `rate_limit_exceeded` code is treated as a transient rate-limit (not quota), and the two broad patterns (`/resource.*exhaust/i`, `/check.*quota/i`) were replaced with Antigravity-specific ones (`individual quota reached`, `enable overages`). ([#5579](https://github.com/diegosouzapw/OmniRoute/pull/5579) — thanks @Chewji9875) + +- **providers (OpenAI-compatible):** Codex MCP / `tool_search` deferred discovery (and `apply_patch`) now works through a **Custom OpenAI-compatible provider**. When such a provider received a Responses-API-shaped request that carried MCP / `tool_search` tools, OmniRoute downgraded it to `/chat/completions`, which drops the deferred tool-discovery mechanism — so the MCP namespaces never surfaced to the model and `apply_patch` was mis-handled as a JSON tool. The executor now detects a Responses-shaped request (`input` / `previous_response_id` / `max_output_tokens` / `reasoning`) that carries `namespace` / `tool_search*` tools and routes it to the upstream `/responses` endpoint natively instead of downgrading (it can also be forced via `providerSpecificData._omnirouteForceResponsesUpstream`). This is a distinct code path from the official Codex OAuth backend (#3033 / #4539, which the earlier fix never touched). Regression guard: `tests/unit/executor-default-base.test.ts`. Thanks to @KooshaPari for the fix. ([#5483](https://github.com/diegosouzapw/OmniRoute/issues/5483)) + +- **dashboard (routing):** selecting the **fusion** strategy on the Global Routing defaults tab now reveals fusion-specific config instead of only the generic resilience fields. Fusion's engine knobs — `judgeModel` (the model that synthesizes the panel answers) and `fusionTuning` (`minPanel` / `stragglerGraceMs` / `panelHardTimeoutMs`) — already existed in the schema and the per-combo editor, but the Global Routing tab never surfaced them, so picking "fusion" there was effectively a no-op. The fields are now shown (extracted into a new `FusionDefaultsFields` component). Voting / aggregation-mode / per-provider-weight are intentionally not shown — those don't exist in the fusion engine. Regression guard: `tests/unit/ui/combo-defaults-fusion-5598.test.tsx`. ([#5598](https://github.com/diegosouzapw/OmniRoute/issues/5598)) + +- **dashboard (free proxy pool):** the free proxy pool "Sync All" no longer fails silently with `Total: 0`. Three fixes: (1) the **IPLocate** source fetched `…/protocols/.json` and parsed it as JSON, but the upstream list is plain text (`.txt`, one `ip:port` per line) — every protocol 404'd / failed to parse; it now fetches `.txt` and parses the line list. (2) The sync route **isolates each source** in its own try/catch, so one provider throwing (e.g. a TLS handshake failure) no longer aborts the whole sync — the working sources still populate the pool. (3) The UI now **surfaces the per-source errors** the route already returns, instead of discarding the response, so a partial/empty sync explains itself. Regression guards: `tests/unit/free-proxy-providers.test.ts`, `tests/unit/proxy-pool-sync-4878.test.ts`, `tests/unit/free-pool-tab.test.tsx`. ([#5595](https://github.com/diegosouzapw/OmniRoute/issues/5595)) + +- **dashboard (memory engine):** the memory engine status page no longer mixes English and Portuguese. The embedding / vector-store / rerank **status detail strings** were hardcoded in Portuguese in the backend (`resolveEmbeddingSource`, `engineStatus`), e.g. `auto: nenhuma fonte de embedding disponível` and `sqlite-vec ativo, dim=…`, while the surrounding UI labels render from the English i18n bundle — so an English user saw a half-translated page. The backend detail strings are now English (`auto: no embedding source available`, `sqlite-vec active, dim=…`, etc.), matching the rest of the page. Regression guard: `tests/unit/memory-engine-status.test.ts`. ([#5596](https://github.com/diegosouzapw/OmniRoute/issues/5596)) + +- **providers (cline):** stop falsely mapping valid **Cline** (OAuth) responses to `502 empty_choices` + account cooldown. `detectMalformedNonStream` only recognized `choices[].message.content` as a **string**, but some OpenAI-compatible upstreams — Cline via OAuth among them — return `content` as an **array of Anthropic-style text blocks** inside an OpenAI envelope. A non-empty response (recvBytes > 0) was therefore classified as `empty_choices` and turned into a 502 that also cooled the account down. The malformed-response detector now also treats a content array carrying at least one non-empty `text` block as real output. Regression guard: `tests/unit/diagnostics.test.ts`. ([#5559](https://github.com/diegosouzapw/OmniRoute/issues/5559)) + +- **embedded services (Windows):** fix **CLIProxyAPI install failing instantly with `spawn unzip ENOENT`** on Windows. The binary extractor spawned `unzip`, which is not a Windows system command — it only ships inside Git for Windows' `usr/bin`, a directory Node's `spawn` PATH never sees, so even users with Git installed hit the error. On Windows the extractor now uses PowerShell's built-in `Expand-Archive` (via `execFileAsync`, no shell — paths pass as a single non-interpreted arg, with `''`-escaping + `-LiteralPath` as defense in depth); other platforms keep using `unzip`. This is distinct from #5379 (that was `npm.cmd` needing `shell: true`). Regression guard: `tests/unit/binary-manager-extract-zip-5590.test.ts`. ([#5590](https://github.com/diegosouzapw/OmniRoute/issues/5590)) + +- **storage (daemon):** fix a Node.js **out-of-memory crash on startup** when `storage.sqlite` grows large (~170 MB+). The boot-time call-log cleanup (`cleanupExpiredLogs` → `rotateCallLogs`) ran two **unbounded `SELECT … FROM call_logs … .all()`** queries — `listReferencedArtifacts` (every artifact path) and `deleteCallLogsBefore` (every id before the retention cutoff). `node:sqlite`'s `StatementSync.all()` materializes the entire result set as JS objects at once, so on a large table the V8 heap blew up and the process crashed before binding (`FATAL ERROR: … heap out of memory`, native frame `node::sqlite::StatementSync::All`). Both queries now page through `call_logs` in bounded 5 000-row chunks (new `src/lib/usage/callLogsBoundedQueries.ts`), keeping peak memory flat regardless of table size — no more manual `--max-old-space-size` bump required. Regression guard: `tests/unit/call-log-oom-unbounded-5618.test.ts`. ([#5618](https://github.com/diegosouzapw/OmniRoute/issues/5618)) + +- **dashboard (provider setup):** fix three provider setup links that pointed at 404 pages. **Ollama Cloud** / **ollama-search** linked to `ollama.com/settings/api-keys` → corrected to `ollama.com/settings/keys` (the page moved; Ollama Cloud is a real keyed service, so the field stays). **SearchAPI** linked to the bare `searchapi.io/docs` (404) → `searchapi.io/docs/google`. **You.com** linked to `you.com/docs/search/overview` (404) → `you.com/business/api/` (the developer portal). All three replacements were verified live. Regression guard: `tests/unit/provider-setup-links-5572.test.ts`. ([#5572](https://github.com/diegosouzapw/OmniRoute/issues/5572), [#5574](https://github.com/diegosouzapw/OmniRoute/issues/5574), [#5576](https://github.com/diegosouzapw/OmniRoute/issues/5576)) + +- **providers (AI/ML API):** the model-import step now loads the **live** AI/ML API catalog (400+ models) instead of falling back to a stale 6-model seed. The registry had no `modelsUrl`, so the route silently used the bundled catalog with an "API unavailable — using local catalog" warning even when the key was valid. AI/ML API exposes its full catalog at the public, **auth-free** `https://api.aimlapi.com/models` endpoint (a bare array of `{ id, type, info }`, distinct from the OpenAI-compat `/v1/models`); it's now wired into the models route's discovery config, with the bundled catalog kept as the offline fallback. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5570](https://github.com/diegosouzapw/OmniRoute/issues/5570)) + +- **providers (CablyAI):** mark **CablyAI** deprecated — `cablyai.com` no longer resolves (DNS `NXDOMAIN`, verified 2026-06-30); the domain is gone. The provider is removed from the models-route discovery config so the import step returns a clean error instead of an **unhandled 500 crash** (the dead-domain fetch threw with no local-catalog fallback), and the registry entry now carries `deprecated: true` / `riskNoticeVariant: "deprecated"` so the dashboard flags existing connections (same treatment as the shut-down `glhf`/`kluster.ai` gateways). Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5568](https://github.com/diegosouzapw/OmniRoute/issues/5568)) + +- **dashboard (provider add):** non-LLM search/agent providers no longer fail the model-import step with a red `Provider does not support models listing`. **Jules** (Google Labs coding agent), **linkup-search** (Linkup web search), **ollama-search** (Ollama Cloud web search — distinct from the local Ollama LLM), and **searchapi-search** (SearchAPI SERP) have no `/v1/models` endpoint, so the import surfaced a failure for expected behavior. Each now ships a small static catalog of its selectable capability ids — Linkup's `fast`/`standard`/`deep` search depths, SearchAPI's `google`/`bing`/`youtube`/… engines, a single Jules/Ollama-web-search entry — so the import step returns a usable list (`source: local_catalog`) instead of an error. Regression guard: `tests/unit/provider-models-route.test.ts`. ([#5569](https://github.com/diegosouzapw/OmniRoute/issues/5569), [#5571](https://github.com/diegosouzapw/OmniRoute/issues/5571), [#5573](https://github.com/diegosouzapw/OmniRoute/issues/5573), [#5575](https://github.com/diegosouzapw/OmniRoute/issues/5575)) + +- **dashboard (provider add):** providers without a live key/cookie validator (e.g. **LMArena (Free)**, **PiAPI**) can now be saved. The Add-connection modal treated the backend's `"Provider validation not supported"` response as a hard **Invalid** state and blocked Save entirely, leaving those providers impossible to add. The validate route now returns `unsupported: true` alongside the message, and the modal treats that as a non-blocking warning — the "Check" badge still shows "validation not supported" (informational), but Save persists the credential as-is. Regression guards: `tests/unit/ui/add-api-key-modal-unsupported-save-5565.test.tsx` (Save proceeds) and `tests/unit/providers-validate-route.test.ts` (wire-format). ([#5565](https://github.com/diegosouzapw/OmniRoute/issues/5565), [#5567](https://github.com/diegosouzapw/OmniRoute/issues/5567)) + +- **providers (codex):** fix the **Codex Responses WebSocket** path (`/v1/responses`), which regressed in v3.8.40 with a client-visible `Invalid JSON body` and bypassed the configured proxy. (1) #5591 — PR #5237 bumped the impersonation TLS profile to `chrome_149`, but `wreq-js@2.3.1` only supports up to `chrome_147`; the unknown profile produced a degenerate fingerprint and ChatGPT rejected the upstream upgrade. The Codex WS path is reverted to the proven `chrome_142` (the v3.8.39 value), and the over-bumped `grok-web`/`claude-web` profiles (masked by their circuit-breaker but silently dropping TLS impersonation) are restored to `chrome_146`. A new regression guard asserts every configured `chrome_*` profile exists in the installed `wreq-js` typings (`tests/unit/tls-profiles-valid-5591.test.mjs`). (2) #5611 — the upstream `wreq-js.websocket()` connect ignored the Proxy Registry, so a no-direct-egress Docker container failed with a DNS error; the prepare route now resolves the Global/provider proxy and threads it through to the WS connect. Regression guard in `tests/unit/responses-ws-proxy.test.mjs`. ([#5591](https://github.com/diegosouzapw/OmniRoute/issues/5591), [#5611](https://github.com/diegosouzapw/OmniRoute/issues/5611)) + +- **providers (GLM):** GLM **5.1 / 5.2** now keep the `system` role instead of having the system prompt folded into the first user turn. `roleNormalizer.ts` matched every `glm*` id with a blanket `startsWith("glm")` / `startsWith("glm-")` prefix, so the next-generation models — which z.ai documents as supporting the `system` role (GLM > 5.0) — were normalized as if they rejected it, degrading instruction-following. The matcher is now version-aware: it strips the system role only for bare `glm`, the 4.x family, and the 5.0 generation, and preserves it for `glm-5.1`/`glm-5.2` (and the Fireworks `glm-5p1` point alias). The ZenMux vendor-prefixed `z-ai/glm-*` compressed-history rule and the ERNIE rule are unchanged. Regression guards in `tests/unit/role-normalizer.test.ts`. ([#5610](https://github.com/diegosouzapw/OmniRoute/issues/5610)) + +- **Security hardening follow-ups (v3.8.15):** the `auth_token` cookie now sets an explicit 30-day `maxAge` so sessions persist as intended (Seg3); the management bootstrap warns at boot when `INITIAL_PASSWORD` is left at the insecure `CHANGEME` default (Seg2); VS Code path-token endpoints (`/api/v1/vscode/raw/[token]`) emit a once-per-process security warning since the API key travels in the URL and can leak via logs/proxies (Seg4); the system version route resolves the real global install path via `npm root -g` instead of a hardcoded `/app` (Bug3); and auto-update mode detection segment-matches `node_modules` instead of substring-matching, eliminating false "global install" positives (Bug1). + +- **fix(cli):** rename the Node process title to `omniroute` so it shows correctly in ps/htop. (thanks @waguriagentic) + +- **dashboard (model picker):** guard against null model-alias values so opening Create Combo for a custom provider node no longer crashes. `ModelSelectModal`'s custom-provider branch filtered `modelAliases` entries with a raw `fullModel.startsWith(...)`, which threw a `TypeError` whenever an alias value was `null`/`undefined` (a stale/partial entry persisted to settings). The filter/map logic is extracted into a new `buildNodeAliasModels` helper (mirroring the sibling passthrough-alias guard, #485) that requires `typeof fullModel === "string"` before calling `.startsWith`. Regression guard: `tests/unit/model-select-null-alias-guard-2247.test.ts`. (thanks @wahyuzero) + +- **fix(translator):** strip orphaned tool results (results with no matching tool call) across request formats to avoid upstream 400s. (thanks @warelik) + +- **fix(kiro):** stop injecting a placeholder user turn on trailing tool-result turns so agentic loops aren't disrupted. (thanks @jetmiky) + +- **fix(translator):** prevent doubled tool arguments in OpenAI-to-Claude responses (duplicate finish_reason guard + string tool-input passthrough). (thanks @vishalrajv) + +- **codex (agent goal streams):** protect long-running agent goal streams so extended agent runs are no longer cut off prematurely. ([#5772](https://github.com/diegosouzapw/OmniRoute/pull/5772) — thanks @nguyenxvotanminh3) + +- **sse (zero-width markers):** strip zero-width markers from streamed responses, matching the non-streaming path so streamed output is byte-clean parity. ([#5857](https://github.com/diegosouzapw/OmniRoute/pull/5857) — thanks @DKotsyuba) + +- **usage (om-usage endpoint):** restore the `om-usage` HTTP endpoint. ([#5859](https://github.com/diegosouzapw/OmniRoute/pull/5859) — thanks @Witroch4) + +- **sse (stream readiness):** tune adaptive stream-readiness timeouts so slow-first-token upstreams are handled more reliably. ([#5767](https://github.com/diegosouzapw/OmniRoute/pull/5767) — thanks @nguyenxvotanminh3) + +- **security (provider node URL):** harden provider node URL validation. ([#5760](https://github.com/diegosouzapw/OmniRoute/pull/5760) — thanks @nguyenxvotanminh3) + +- **cli (Windows doctor):** correct `rootDir` resolution in `doctor.mjs` on Windows. ([#5845](https://github.com/diegosouzapw/OmniRoute/pull/5845) — thanks @arssnndr) + +- **providers (Antigravity):** fix a 429 hang on credit exhaustion and apply a precise reset-time model lockout instead of stalling — cleaned re-implementation of #5823. ([#5846](https://github.com/diegosouzapw/OmniRoute/pull/5846) — thanks @Chewji9875 / @diegosouzapw) + +- **providers (qwen-web):** unblock the validator and chat completion — the retired endpoint is replaced and the missing SPA version header is now sent. ([#5855](https://github.com/diegosouzapw/OmniRoute/pull/5855) — thanks @janeza2) + +- **providers (kimi-web):** migrate to the `www.kimi.com` Connect-RPC API after `kimi.moonshot.cn` was retired. ([#5858](https://github.com/diegosouzapw/OmniRoute/pull/5858) — thanks @janeza2) + +- **dashboard (CSRF):** unify the dashboard CSRF origin fallback so dynamic/public origins validate correctly. ([#5856](https://github.com/diegosouzapw/OmniRoute/pull/5856) — thanks @rdself) + +- **db (health check interval):** preserve `healthCheckInterval=0` across connection create/update instead of coercing it to a default. ([#5822](https://github.com/diegosouzapw/OmniRoute/pull/5822) — thanks @atomlong) + +- **sse (claude→codex streaming):** stop the reasoning-summary drop and duplicated deltas on claude→codex streaming — reasoning snapshots are now synthesized in TRANSLATE mode and the sequence-number watermark is tracked per-stream (#5786). ([#5832](https://github.com/diegosouzapw/OmniRoute/pull/5832) — thanks @diegosouzapw) + +- **deps (runtime):** add the missing runtime dependencies `@toon-format/toon` and `safe-regex` so the published package resolves them at runtime. ([#5771](https://github.com/diegosouzapw/OmniRoute/pull/5771) — thanks @chirag127) + +- **system (Windows auto-update):** route in-app auto-update `npm` calls through the win32 shell helper so updates run correctly on Windows (#5542). ([#5797](https://github.com/diegosouzapw/OmniRoute/pull/5797) — thanks @diegosouzapw) + +- **dashboard (validation badge):** show a neutral badge for unsupported validation and make OAuth error messages clickable links (#5442, #5486). ([#5795](https://github.com/diegosouzapw/OmniRoute/pull/5795) — thanks @diegosouzapw) + +- **providers (metadata):** correct stale/broken provider metadata (#5487, #5461, #5534, #5470). ([#5790](https://github.com/diegosouzapw/OmniRoute/pull/5790) — thanks @diegosouzapw) + +- **providers (local-catalog imports):** import intentional local-catalog-only providers instead of surfacing a 502 (#5460, #5465). ([#5787](https://github.com/diegosouzapw/OmniRoute/pull/5787) — thanks @diegosouzapw) + +- **proxyfetch (failover):** skip the failover retry for non-replayable request bodies so a consumed stream isn't re-sent empty. ([#5770](https://github.com/diegosouzapw/OmniRoute/pull/5770) — thanks @Ardem2025) + +- **batch (recovery):** persist batch item checkpoints during recovery so an interrupted batch resumes from where it left off. ([#5753](https://github.com/diegosouzapw/OmniRoute/pull/5753) — thanks @ag-linden) + +- **memory (Qdrant):** enabling Qdrant now activates it as the retrieval engine (the `auto` default never selected it) and adds inline guidance (#5597). ([#5741](https://github.com/diegosouzapw/OmniRoute/pull/5741) — thanks @diegosouzapw) + +- **chat (non-streaming aggregation):** harden non-streaming SSE aggregation against malformed upstream event sequences. ([#5746](https://github.com/diegosouzapw/OmniRoute/pull/5746) — thanks @rdself) + +- **sse (cooldown parsing):** the anti-thundering-herd guard now tolerates numeric-epoch cooldown values. ([#5747](https://github.com/diegosouzapw/OmniRoute/pull/5747) — thanks @diegosouzapw) + +- **api (body size):** raise the LLM API payload limit for the responses routes so larger requests aren't rejected. ([#5652](https://github.com/diegosouzapw/OmniRoute/pull/5652) — thanks @JxnLexn) + +- **providers (HuggingChat):** fix HuggingChat web-session routing (#5592). ([#5592](https://github.com/diegosouzapw/OmniRoute/pull/5592) — thanks @backryun) + +- **sse (heap pressure):** bound the chat hot-path heap — pressure-aware admission, response cap, and clone reductions — to avoid OOM under load (#5152). ([#5425](https://github.com/diegosouzapw/OmniRoute/pull/5425) — thanks @josevictorferreira) + +- **providers (M365 Copilot):** validate M365 Copilot web credentials. ([#5432](https://github.com/diegosouzapw/OmniRoute/pull/5432) — thanks @skyzea1) + +- **providers (chatgpt-web):** restore the dot-form Pro model ids. ([#5549](https://github.com/diegosouzapw/OmniRoute/pull/5549) — thanks @Thinkscape) + +- **security (error stacks):** avoid rendering error stacks in responses. ([#5624](https://github.com/diegosouzapw/OmniRoute/pull/5624) — thanks @KooshaPari) + +- **security (linkify):** restrict `linkifyText` hrefs to an explicit `http(s)` scheme allowlist. ([#948d2d7](https://github.com/diegosouzapw/OmniRoute/commit/948d2d7f2) — thanks @diegosouzapw) + +- **translator (doubled tool args):** prevent doubled tool-call arguments in the OpenAI→Claude translation path. ([#5828](https://github.com/diegosouzapw/OmniRoute/pull/5828) — thanks @diegosouzapw) + +- **translator (orphaned tool results):** strip orphaned tool-result turns across request formats so an upstream doesn't reject a tool result with no matching call. ([#5805](https://github.com/diegosouzapw/OmniRoute/pull/5805) — thanks @diegosouzapw) + +- **translator (Gemini/Claude hardening):** re-apply lost defensive hardening for the Gemini merge path and Claude tool defaults. ([#5706](https://github.com/diegosouzapw/OmniRoute/pull/5706) — thanks @diegosouzapw) + +- **kiro (tool-result turns):** stop injecting a placeholder user turn on tool-result turns, which corrupted otherwise-valid Kiro conversations. ([#5807](https://github.com/diegosouzapw/OmniRoute/pull/5807) — thanks @diegosouzapw) + +- **providers (Kiro catalog):** add `claude-sonnet-5` to the Kiro model catalog. ([#5796](https://github.com/diegosouzapw/OmniRoute/pull/5796) — thanks @diegosouzapw) + +- **oauth (connection disambiguation):** disambiguate OAuth connections on username so two different identity providers no longer overwrite each other. ([#5803](https://github.com/diegosouzapw/OmniRoute/pull/5803) — thanks @diegosouzapw) + +- **github (Copilot prefill):** drop the trailing assistant prefill for Copilot chat, which some Copilot models rejected. ([#5802](https://github.com/diegosouzapw/OmniRoute/pull/5802) — thanks @diegosouzapw) + +- **mitm (hosts cleanup):** clean up privileged `/etc/hosts` entries on exit when possible so a crashed/interrupted run doesn't leave stale redirects behind. ([#5808](https://github.com/diegosouzapw/OmniRoute/pull/5808) — thanks @diegosouzapw) + +- **dashboard (model picker):** guard null `modelAliases` values in the model picker so a connection with no aliases no longer throws. ([#5792](https://github.com/diegosouzapw/OmniRoute/pull/5792) — thanks @diegosouzapw) + +- **dashboard (error boundaries):** add error boundaries for the Combos and MITM Proxy pages so a render error no longer blanks the whole dashboard. ([#5788](https://github.com/diegosouzapw/OmniRoute/pull/5788) — thanks @diegosouzapw) + +- **cli (process title):** rename the running process title to `omniroute`. ([#5791](https://github.com/diegosouzapw/OmniRoute/pull/5791) — thanks @diegosouzapw) + +- **compression (context-editing telemetry):** record Context Editing telemetry on the streaming path, not just the non-streaming path. ([#5761](https://github.com/diegosouzapw/OmniRoute/pull/5761) — thanks @diegosouzapw) + +- **security (v3.8.15 hardening follow-ups):** land the Seg2/Seg3/Seg4/Bug3 hardening follow-ups from the v3.8.15 security review. ([#5512](https://github.com/diegosouzapw/OmniRoute/pull/5512) — thanks @diegosouzapw) + +### 📝 Maintenance + +- **docs (architecture):** add `docs/architecture/ROUTER_BACKENDS.md` — an ADR pinning down how the routing engines (`ts` native, `bifrost`, `cliproxy`, `9router`, VibeProxy-compatible) relate to each other along two orthogonal axes (lifecycle: in-process / supervised / external vs. relay selection backend), answering the architecture questions raised in [#5603](https://github.com/diegosouzapw/OmniRoute/issues/5603) (backend interface model, why CLIProxy spawns a process, feature-flag swapping, actionable route-contract errors). The typed router-backend registry the ADR describes lands separately via [#5868](https://github.com/diegosouzapw/OmniRoute/pull/5868). ([#5891](https://github.com/diegosouzapw/OmniRoute/pull/5891)) + +- **tests (autoCombo):** stabilize the `getTaskFitnessWithSource identifies fitness_table as source for known models` unit test, which flaked whenever the models.dev capabilities DB was populated in CI: the fixture model `gpt-4o` is a real models.dev catalog id, so the fitness resolution chain returned `models_dev_tier` instead of the expected static `fitness_table` source. The fixture now uses `claude-sonnet` (a shortened alias absent from the models.dev catalog, matching the sibling resolution-chain test), which deterministically falls through to the static table — the exact `source` and score assertions are preserved (`0.95` = `FITNESS_TABLE.coding["claude-sonnet"]`). ([#5890](https://github.com/diegosouzapw/OmniRoute/pull/5890)) — thanks @KooshaPari + +- **oauth (dead-code removal):** delete the superseded legacy OAuth **service-class** hierarchy under `src/lib/oauth/services/`. The live OAuth flow runs through `src/lib/oauth/providers.ts` + `src/lib/oauth/providers/` (wired into the generic `oauth/[provider]/[action]` route); the old per-provider `class *Service extends OAuthService` implementations plus their barrel had **zero** production or test references. Removed `oauth.ts` (base class), `openai.ts`, `github.ts`, `claude.ts`, `codex.ts`, `antigravity.ts`, `qwen.ts`, `qoder.ts`, and the `index.ts` barrel (−1559 LOC). Kept the three still-live files that routes import **directly** by path: `kiro.ts` (Kiro import/exchange routes), `cursor.ts` (Cursor import route), and `codexImport.ts` (utility fns for the Codex bulk-import route). Proven safe by `typecheck:core` staying green (any live reference would fail the build) + a filesystem guard `tests/unit/oauth-legacy-services-removed.test.ts` pinning the removal against re-introduction. Salvage of the closed PR [#5039](https://github.com/diegosouzapw/OmniRoute/pull/5039). gaps v3.8.42 — T10 (5.7). + +- **refactor (god-file decomposition):** extracted pure leaf modules across db, sse, usage, api, memory, evals, models, resilience, and dashboard god-files (types/mappers/helpers/pure-transform leaves; behavior-preserving, test-guarded): db/providers, db/proxies, db/models, db/settings, usageAnalytics, migrationRunner ([#5714](https://github.com/diegosouzapw/OmniRoute/pull/5714), [#5717](https://github.com/diegosouzapw/OmniRoute/pull/5717), [#5705](https://github.com/diegosouzapw/OmniRoute/pull/5705), [#5709](https://github.com/diegosouzapw/OmniRoute/pull/5709), [#5722](https://github.com/diegosouzapw/OmniRoute/pull/5722), [#5721](https://github.com/diegosouzapw/OmniRoute/pull/5721)); sse openai-to-gemini / cursor-protobuf / rate-limit-headers / reasoning-tag ([#5824](https://github.com/diegosouzapw/OmniRoute/pull/5824), [#5794](https://github.com/diegosouzapw/OmniRoute/pull/5794), [#5736](https://github.com/diegosouzapw/OmniRoute/pull/5736), [#5734](https://github.com/diegosouzapw/OmniRoute/pull/5734)); usage families / callLogs / usageHistory / providerLimits ([#5782](https://github.com/diegosouzapw/OmniRoute/pull/5782), [#5725](https://github.com/diegosouzapw/OmniRoute/pull/5725), [#5728](https://github.com/diegosouzapw/OmniRoute/pull/5728), [#5730](https://github.com/diegosouzapw/OmniRoute/pull/5730)); api provider-models discovery / unified-catalog ([#5758](https://github.com/diegosouzapw/OmniRoute/pull/5758), [#5699](https://github.com/diegosouzapw/OmniRoute/pull/5699)); memory retrieval scoring ([#5733](https://github.com/diegosouzapw/OmniRoute/pull/5733)); evals golden-set suites ([#5740](https://github.com/diegosouzapw/OmniRoute/pull/5740)); modelsDevSync transform layer ([#5743](https://github.com/diegosouzapw/OmniRoute/pull/5743)); resilience settings split ([#5745](https://github.com/diegosouzapw/OmniRoute/pull/5745)); dashboard sidebarVisibility split ([#5683](https://github.com/diegosouzapw/OmniRoute/pull/5683)); executor shared-utility dedup + tests ([#5720](https://github.com/diegosouzapw/OmniRoute/pull/5720) — thanks @pizzav-xyz). — thanks @diegosouzapw + +- **chore (Bun script runner):** adopt Bun `1.3.10` as a locked, allow-listed **build/dev script runner** for a small set of validated TS gate/generator scripts (Node stays the published runtime): locked runtime dependency, CI script-checks + validated-scripts run under Bun, and a bun-safe pack validator. ([#5615](https://github.com/diegosouzapw/OmniRoute/pull/5615), [#5617](https://github.com/diegosouzapw/OmniRoute/pull/5617), [#5612](https://github.com/diegosouzapw/OmniRoute/pull/5612), [#5643](https://github.com/diegosouzapw/OmniRoute/pull/5643) — thanks @KooshaPari; docs [#5703](https://github.com/diegosouzapw/OmniRoute/pull/5703) — thanks @diegosouzapw) + +- **docs (sync & housekeeping):** i18n CHANGELOG mirror sync for the [3.8.43] section ([#5789](https://github.com/diegosouzapw/OmniRoute/pull/5789)); MCP tool count synced to 95 + routing-strategy count ([#5732](https://github.com/diegosouzapw/OmniRoute/pull/5732)); README faster/leaner install notes, refreshed metrics/badges, 17-strategy + Quota-Share listing, provider counts, and grammar fixes ([#5713](https://github.com/diegosouzapw/OmniRoute/pull/5713), [#5738](https://github.com/diegosouzapw/OmniRoute/pull/5738) — thanks @chirag127); security docs for banned-keyword/account-ban detection ([#5756](https://github.com/diegosouzapw/OmniRoute/pull/5756)) and the full LOCAL_ONLY route set + GHSA advisory + audit path ([#5748](https://github.com/diegosouzapw/OmniRoute/pull/5748)); relay backend-routing contract clarification ([#5621](https://github.com/diegosouzapw/OmniRoute/pull/5621) — thanks @KooshaPari); release-freeze scoped to `/generate-release` only ([#5839](https://github.com/diegosouzapw/OmniRoute/pull/5839)); `.editorconfig` repository standards ([#5879](https://github.com/diegosouzapw/OmniRoute/pull/5879) — thanks @shiva24082). — thanks @diegosouzapw + +- **test/ci (stabilization & ratchets):** guard the tsx/esm→esbuild boot transform ([#5773](https://github.com/diegosouzapw/OmniRoute/pull/5773)); align t3-web web-session metadata ([#5835](https://github.com/diegosouzapw/OmniRoute/pull/5835)); repoint the sidebar quota-share placement scan ([#5711](https://github.com/diegosouzapw/OmniRoute/pull/5711)); lightweight health probe for batch e2e ([#5651](https://github.com/diegosouzapw/OmniRoute/pull/5651) — thanks @KooshaPari); make release-green pre-flight gates visible + bounded ([#5644](https://github.com/diegosouzapw/OmniRoute/pull/5644)); stabilize nightly-mutation (tap.testFiles drift guard + anti-flake eps) ([#5682](https://github.com/diegosouzapw/OmniRoute/pull/5682)); close the QG v2 tail ([#5681](https://github.com/diegosouzapw/OmniRoute/pull/5681)); normalize check route paths on Windows ([#5613](https://github.com/diegosouzapw/OmniRoute/pull/5613) — thanks @KooshaPari); pass `sonar.projectVersion` to the SonarQube scan ([#5880](https://github.com/diegosouzapw/OmniRoute/pull/5880)); plus stryker `tap.testFiles` registration, compression-studio smoke re-anchoring, `rtk_discover` de-flake, and v3.8.43-cycle ratchet rebaselines (deadExports 225→227, complexity 1981→1982, cognitive-complexity 842→845, eslintWarnings 4121→4158→4199). — thanks @diegosouzapw + +- **refactor (oauth):** remove dead legacy OAuth service classes. ([#5838](https://github.com/diegosouzapw/OmniRoute/pull/5838) — thanks @diegosouzapw) + +### 🙌 Contributors + +Thanks to everyone whose work landed in v3.8.43: + +| Contributor | PRs / Issues | +| ------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------- | +| [@ag-linden](https://github.com/ag-linden) | #5753 | +| [@Ardem2025](https://github.com/Ardem2025) | #5770 | +| [@arssnndr](https://github.com/arssnndr) | #5845 | +| [@atomlong](https://github.com/atomlong) | #5822 | +| [@backryun](https://github.com/backryun) | #5592 | +| [@baslr](https://github.com/baslr) | direct commit / report | +| [@Chewji9875](https://github.com/Chewji9875) | #5563, #5579, #5846 | +| [@chirag127](https://github.com/chirag127) | #5738, #5771 | +| [@DKotsyuba](https://github.com/DKotsyuba) | #5857 | +| [@hartmark](https://github.com/hartmark) | #5834 | +| [@ishatiwari21](https://github.com/ishatiwari21) | #5799 | +| [@janeza2](https://github.com/janeza2) | #5855, #5858 | +| [@jetmiky](https://github.com/jetmiky) | direct commit / report | +| [@josevictorferreira](https://github.com/josevictorferreira) | #5425 | +| [@JxnLexn](https://github.com/JxnLexn) | #5652 | +| [@KooshaPari](https://github.com/KooshaPari) | #5613, #5621, #5624, #5629, #5643, #5651, #5890 | +| [@KunN-21](https://github.com/KunN-21) | direct commit / report | +| [@manhdzzz](https://github.com/manhdzzz) | direct commit / report | +| [@nguyenxvotanminh3](https://github.com/nguyenxvotanminh3) | #5760, #5767, #5772 | +| [@noir017](https://github.com/noir017) | direct commit / report | +| [@pizzav-xyz](https://github.com/pizzav-xyz) | #5720 | +| [@rdself](https://github.com/rdself) | #5746, #5856 | +| [@shiva24082](https://github.com/shiva24082) | #5879 | +| [@skyzea1](https://github.com/skyzea1) | #5432, #5701 | +| [@Stazyu](https://github.com/Stazyu) | #5557 | +| [@Thinkscape](https://github.com/Thinkscape) | #5549 | +| [@vishalrajv](https://github.com/vishalrajv) | direct commit / report | +| [@voravitl](https://github.com/voravitl) | direct commit / report | +| [@waguriagentic](https://github.com/waguriagentic) | direct commit / report | +| [@wahyuzero](https://github.com/wahyuzero) | direct commit / report | +| [@warelik](https://github.com/warelik) | direct commit / report | +| [@Witroch4](https://github.com/Witroch4) | #5731, #5859, #5863 | +| [@diegosouzapw](https://github.com/diegosouzapw) | maintainer — cycle reconciliation, release-close base-red fixes, god-file decomposition, compression/memory features | + +--- + ## [3.8.42] — 2026-06-30 ### ✨ New Features diff --git a/docs/i18n/zh-TW/README.md b/docs/i18n/zh-TW/README.md index cd191f197f..40137a2f05 100644 --- a/docs/i18n/zh-TW/README.md +++ b/docs/i18n/zh-TW/README.md @@ -626,7 +626,7 @@ PORT=20128 npm run dev **📦 pnpm** ```bash -pnpm install -g omniroute && pnpm approve-builds -g && omniroute +pnpm add -g omniroute@latest --allow-build=better-sqlite3 --allow-build=@swc/core && omniroute ``` **🐧 Arch Linux（AUR）** diff --git a/docs/openapi.yaml b/docs/openapi.yaml index c4ed685575..764ee2cb37 100644 --- a/docs/openapi.yaml +++ b/docs/openapi.yaml @@ -1,7 +1,7 @@ openapi: 3.1.0 info: title: OmniRoute API - version: 3.8.42 + version: 3.8.43 description: | OmniRoute is a local-first AI API proxy router. It provides an OpenAI-compatible endpoint that routes requests to multiple AI providers with load balancing, diff --git a/docs/ops/MATURITY_REEVAL.md b/docs/ops/MATURITY_REEVAL.md new file mode 100644 index 0000000000..f0dcc5382a --- /dev/null +++ b/docs/ops/MATURITY_REEVAL.md @@ -0,0 +1,108 @@ +--- +title: "Quality-Gate Maturity Re-evaluation (Fase 9)" +--- + +# Reavaliação de Maturidade — pós-Ondas 0–3 (Quality-Gate v2) + +> **O que é este documento.** Uma re-medição da maturidade do sistema de quality-gates +> **após** as Ondas 0–3 do programa Quality-Gate v2, comparada ao baseline registrado em +> [`QUALITY_GATE_PLAYBOOK.md`](./QUALITY_GATE_PLAYBOOK.md) (2026-06-16). Mede o que mudou, +> contra DSOMM L5 / OpenSSF Scorecard 9 / SLSA L3, separando o que é **CI-mensurável** +> (já entregue / entregável por código) do que é **processo/owner** (settings de organização). +> +> **Data:** 2026-06-30. Gerado do estado real do repositório, não da memória. +> **Régua:** OWASP DSOMM · OpenSSF Scorecard · SLSA · SonarQube "Clean as You Code". + +--- + +## 1. Veredito atualizado + +**Nota geral: A− → A ("Avançado", topo ~5%).** As **duas maiores fraquezas estruturais** +do baseline 06-16 — o *buraco fast-gates* e o *mutation-score-não-catraca* — foram **fechadas**. +Os gaps residuais para "máximo absoluto" são quase todos **owner/infra-gated** (branch-protection, +SLSA L3, CodeQL advanced); o lado-código do programa está essencialmente completo. + +| Framework de referência | Baseline 06-16 | Agora 06-30 | Movimento | Evidência | +| --- | --- | --- | --- | --- | +| **OWASP DSOMM** (5 níveis) | L3→L4 | **L4** em *Test Intensity* e *Static Depth*; L3 sólido nas demais | ▲ | mutation-ratchet bloqueante + suíte determinística no gate-de-merge | +| **OpenSSF Scorecard** | ~7–8/10 | ~7–8/10 (inalterado — gate é o **owner**) | = | falta Branch-Protection na `main` (setting do dono) + pin de actions | +| **SLSA** | L2→L3 | **L2** (encostando em L3) | = | falta builder hermético/reprodutível (infra/owner) | +| **SonarQube "Clean as You Code"** | Alinhado c/ ressalva | Alinhado c/ ressalva | = | ressalva de *sprawl* (~46+ gates) permanece — review de ROI pendente | +| **Quality-Ratchet pattern** | Exemplar | **Exemplar+** | ▲ | novo `dedicatedGate` de `mutationScore` (direction up) | +| **Mutation testing** | "Quase lá" (não-catraca) | **Catraca ativa** | ▲▲ | `check-mutation-ratchet.mjs` + baseline semeado + job nightly bloqueante | + +--- + +## 2. Deltas desde 2026-06-16 (o que as Ondas 0–3 entregaram) + +### 2.1 🔴→✅ Buraco fast-gates FECHADO (era a fraqueza estrutural #1) +O baseline alertava: `quality.yml` (PR→`release/**`) rodava **só gates de filesystem** — sem +typecheck, testes ou build —, então regressões determinísticas só explodiam no PR→`main`. +**Hoje** `.github/workflows/quality.yml` roda, no job *Fast Quality Gates*: `typecheck:core`, +**testes unitários impactados (TIA) bloqueantes com fail-safe para a suíte completa**, o +fast-path do **vitest**, e shards de unit. O gate agora roda **onde o merge acontece** (shift-left), +exatamente o princípio transversal que o playbook prescreve. + +### 2.2 🟠→✅ Mutation score virou CATRACA (era a fraqueza #3 / P0 #1) +O antídoto mais forte contra coverage-gaming estava **advisory**. **Hoje**: +- `scripts/check/check-mutation-ratchet.mjs` (advisory por default, `--ratchet` bloqueante, skip gracioso); +- `config/quality/quality-baseline.json` tem entradas `mutationScore.` semeadas (`direction: up`, `dedicatedGate`); +- `.github/workflows/nightly-mutation.yml` tem o job **"Mutation score ratchet (blocking)"** que unifica os relatórios por-batch e ratcheteia os scores merged por-módulo. + +Resultado: o score de mutação por-módulo **não pode regredir** — cobertura deixou de ser vanity-metric. + +### 2.3 ✅ Quick-wins de gate (Fase 6A/7) entregues +- **a11y axe-core "fake-green" corrigido:** `@axe-core/playwright` em devDeps; `a11y.spec.ts` com skip condicional `REQUIRE_AXE`; job no `nightly-resilience.yml`. +- **complexity varre `bin/`+`electron`:** `check-complexity.mjs` inclui esses diretórios no `ESLINT_ARGS`. +- **tracked-artifacts no pre-commit + pre-push:** `.husky/pre-commit` + `pre-push` bloqueiam artefato rastreado por engano. + +--- + +## 3. As 12 categorias — situação (delta-focada) + +| # | Categoria | Situação 06-30 | +| --- | --- | --- | +| 1 | Estilo & formatação | ✅ inalterado (Prettier+ESLint lint-staged) | +| 2 | Tipos | ✅ **reforçado** — `typecheck:core` agora também no gate PR→release | +| 3 | Testes (intensidade) | ✅ **reforçado** — mutation testing virou catraca; suíte determinística no gate-de-merge | +| 4 | Política de testes (anti-gaming) | ✅ inalterado (pr-test-policy/test-masking/pr-evidence) | +| 5 | Complexidade & saúde | ✅ **reforçado** — complexity varre bin/electron | +| 6 | Segurança estática (SAST+segredos) | 🟡 CodeQL default-setup (advanced = owner); semgrep cloud não-versionado | +| 7 | Supply-chain (deps) | ✅ inalterado (osv/audit/Trivy/Dependabot + allowlist) | +| 8 | Supply-chain (build/release) | 🟡 SLSA L2 (L3 = builder hermético, owner/infra) | +| 9 | Contratos & API | 🟡 oasdiff/osv advisory (candidatos a bloqueante-com-escopo, P1) | +| 10 | Docs & i18n (anti-rot) | ✅ **reforçado** — `fabricated-docs --strict` bloqueante (verificado exit 0) | +| 11 | Anti-alucinação / consistência | ✅ inalterado (known-symbols/fetch-targets/docs-symbols/db-rules) | +| 12 | Resiliência & domínio | ✅ inalterado (chaos/heap/k6/promptfoo/garak nightly) | + +--- + +## 4. Gap residual para "máximo absoluto" + +### 4.1 CI-mensurável / entregável por código (backlog deste programa) +- **P1 — osv/oasdiff → bloqueante com escopo certo:** osv só `CRITICAL`+fixable (two-step como o Trivy); oasdiff bloqueia breaking-change de contrato. +- **P1 — `require-tighten` bloqueante (fim de ciclo):** trava ganhos de métrica (impede afrouxar baseline sem registrar). +- **P1/P2 — review de ROI / sprawl de gates:** consolidar micro-gates de doc-sync; medir timing por-gate no `ci-summary` (combate a fadiga — ressalva do SonarQube/DORA). Os merges ROI deferidos (complexity unificada; `/api` anti-alucinação unificada) entram aqui. +- **P2 — CodeQL config commitado + semgrep versionado:** mais controle/reprodutibilidade. + +### 4.2 Processo / owner (CI não move — settings de organização) +- **Branch-protection na `main`** (sobe Scorecard, fecha o gap DSOMM). Ver [`BRANCH_PROTECTION_MAIN.md`](./BRANCH_PROTECTION_MAIN.md). +- **CodeQL Default → Advanced setup.** +- **SLSA L3** — builder hermético/reprodutível (gerador SLSA do GitHub). Stretch (diminishing returns). + +### 4.3 Explicitamente fora-de-escopo +- **DSOMM L5** é majoritariamente **org-level / processo** (não CI-codificável). +- **SLSA L4** (bit-a-bit reprodutível) é stretch declarado. + +--- + +## 5. Itens deferidos / removidos (housekeeping da cauda) + +- **`semcheck.yaml` (camada LLM de drift semântico docs↔code) — REMOVIDO.** Estava **órfão** + (nenhum workflow/script o invocava) e com contagens stale nas regras. A cobertura determinística + já existe (`check:fabricated-docs --strict` + `check:docs-counts-sync` + `check:docs-symbols`), + e a ressalva de *gate sprawl* desaconselha adicionar um gate LLM advisory de custo recorrente. + Pode ser re-introduzido no futuro como job nightly opt-in se o drift semântico virar problema real. +- **`agent-lsp` scaffold — DEFERIDO / opt-in não-ativado.** Existe como menção em docs + (`docs/architecture/QUALITY_GATES.md`, CHANGELOG) mas **sem wiring** e sem `.mcp.json.example` + no repo. Permanece como scaffold opt-in documentado; não é um gate ativo nem um gap de maturidade. diff --git a/docs/ops/TUNNELS_GUIDE.md b/docs/ops/TUNNELS_GUIDE.md index b5ade8c9a4..a168b7a469 100644 --- a/docs/ops/TUNNELS_GUIDE.md +++ b/docs/ops/TUNNELS_GUIDE.md @@ -218,6 +218,11 @@ build callback URLs against the **public** hostname, not `localhost`. Otherwise the OAuth provider redirects the user back to a URL its servers cannot reach, and the handshake fails. +Dashboard edits and settings saves do not require pinning the tunnel hostname in +`NEXT_PUBLIC_BASE_URL`. The authenticated dashboard sends same-origin unsafe +requests with a session-bound CSRF token, so ephemeral Cloudflare Quick Tunnel +hosts can still be used for normal UI management after logging in. + Set: ```bash diff --git a/docs/ops/VM_DEPLOYMENT_GUIDE.md b/docs/ops/VM_DEPLOYMENT_GUIDE.md index 71f13dff00..b8dc7a5071 100644 --- a/docs/ops/VM_DEPLOYMENT_GUIDE.md +++ b/docs/ops/VM_DEPLOYMENT_GUIDE.md @@ -119,7 +119,7 @@ REQUIRE_API_KEY=false # === URLs (change to your domain) === # Internal server-to-server base URL for scheduled jobs / self-fetches. BASE_URL=http://127.0.0.1:20128 -# Browser-facing URL used for OAuth callbacks, dashboard links, and same-origin checks. +# Browser-facing URL used for OAuth callbacks, dashboard links, and generated public URLs. NEXT_PUBLIC_BASE_URL=https://llms.seudominio.com # Optional explicit public origin override for generated public asset URLs. # OMNIROUTE_PUBLIC_BASE_URL=https://llms.seudominio.com @@ -243,11 +243,13 @@ Keep reverse-proxy stream timeouts aligned with your OmniRoute timeout env vars. `FETCH_TIMEOUT_MS` / `STREAM_IDLE_TIMEOUT_MS`, raise `proxy_read_timeout` / `proxy_send_timeout` above the same threshold. -OmniRoute uses `NEXT_PUBLIC_BASE_URL` as the canonical browser-facing origin for OAuth, -public links, and dashboard mutation origin checks. The `X-Forwarded-*` headers above are -still useful routing metadata, but they are not a replacement for setting the explicit public -URL. Only enable `OMNIROUTE_TRUST_PROXY` if OmniRoute is not directly reachable by clients and -your proxy strips/rebuilds incoming forwarded headers. +OmniRoute uses `NEXT_PUBLIC_BASE_URL` as the canonical browser-facing origin for OAuth +callbacks and generated public links. Authenticated dashboard writes use same-origin requests +plus session-bound CSRF protection, so they do not require a static public base URL. The +`X-Forwarded-*` headers above are still useful routing metadata, but they are not a replacement +for setting the explicit public URL when OAuth or generated browser links need one. Only enable +`OMNIROUTE_TRUST_PROXY` if OmniRoute is not directly reachable by clients and your proxy +strips/rebuilds incoming forwarded headers. ### 3.3 Enable and Test diff --git a/docs/reference/ENVIRONMENT.md b/docs/reference/ENVIRONMENT.md index 55f5a2847e..b307c1b1fb 100644 --- a/docs/reference/ENVIRONMENT.md +++ b/docs/reference/ENVIRONMENT.md @@ -126,6 +126,8 @@ OmniRoute uses **SQLite** (via `better-sqlite3`) for all persistence. These vari | `LIVE_WS_PORT` | `20129` | `src/server/ws/liveServer.ts` | Port for the real-time WebSocket live monitoring server. | | `LIVE_WS_HOST` | `127.0.0.1` | `src/server/ws/liveServer.ts` | Bind address for the live WebSocket server. Set to `0.0.0.0` to expose on LAN (also configure `LIVE_WS_ALLOWED_ORIGINS`). | | `LIVE_WS_ALLOWED_ORIGINS` | _(unset)_ | `src/server/ws/liveServer.ts` | Comma-separated extra origins allowed to open a live WebSocket. Loopback dashboard origins are already permitted by default. | +| `LIVE_WS_ALLOWED_HOSTS` | _(unset)_ | `src/server/ws/liveServerAllowList.ts` | Comma-separated extra hostnames allowed for live WebSocket origins. Unlike `LIVE_WS_ALLOWED_ORIGINS` (full origin URLs), matches only the host portion — useful for LAN/Tailscale setups. | +| `NEXT_PUBLIC_LIVE_WS_PUBLIC_URL` | _(unset)_ | `src/hooks/useLiveDashboard.ts` | Public URL for the live dashboard WebSocket (browser-side). Set when fronting the WS server with a reverse proxy or Cloudflare Tunnel (e.g. `wss://ws.my-ai.com/live-ws`); the browser connects there instead of `ws://hostname:20129`. | | `OMNIROUTE_ENABLE_LIVE_WS` | `true` | `src/server/ws/liveServer.ts` | Set to `0` or `false` to disable the real-time WebSocket server (enabled by default, loopback-bound). | | `OMNIROUTE_DISABLE_LIVE_WS` | `false` | `scripts/start-ws-server.mjs` | CI/harness toggle that disables the standalone live WebSocket helper script. | | `RELAY_IP_PER_MINUTE` | `30` | `src/app/api/v1/relay/chat/completions/route.ts` | Per-(token, IP) relay rate limit, requests/minute. In-memory, per instance. `0` or negative disables the IP-dimension gate (per-token DB limit still applies). | @@ -176,7 +178,11 @@ OmniRoute uses **SQLite** (via `better-sqlite3`) for all persistence. These vari | `NO_LOG_API_KEY_IDS` | _(empty)_ | `src/lib/compliance/index.ts` | Comma-separated API key IDs that bypass request logging (GDPR compliance). | | `DEFAULT_RATE_LIMIT_PER_DAY` | `1000` | `src/shared/utils/apiKeyPolicy.ts` | Fallback per-day request budget applied to API keys whose `rate_limits` column is null. Default (unset/empty/malformed) keeps the legacy 1000/day, 5000/week, 20000/month windows. Set explicitly to `0` to opt out (unlimited). Any positive integer N enables N/day, 5N/week, 20N/month. Zod-validated; invalid values log a warning and use the legacy default. | | `MAX_BODY_SIZE_BYTES` | `10485760` (10 MB) | `src/shared/middleware/bodySizeGuard.ts` | Maximum allowed request body size. Rejects payloads exceeding this limit. | -| `CORS_ORIGIN` | _(unset)_ | `src/server/cors/origins.ts` | Legacy single-origin CORS allowlist. Prefer `CORS_ALLOWED_ORIGINS` for new deployments. CORS is only for cross-origin browser API clients; same-origin dashboard requests behind a reverse proxy use `NEXT_PUBLIC_BASE_URL` / public-origin validation instead. | +| `OMNIROUTE_CHAT_LARGE_BODY_BYTES` | `262144` (256 KB) | `src/shared/middleware/chatBodyAdmission.ts` | Heap-pressure admission threshold for `POST /v1/chat/completions` (#5152). Bodies below this are always admitted and never sample the heap; at or above it the heap-pressure check applies. | +| `OMNIROUTE_CHAT_HARD_MAX_BODY_BYTES` | `52428800` (50 MB) | `src/shared/middleware/chatBodyAdmission.ts` | Chat-route hard cap. Bodies larger than this are rejected with `413` before being cloned/parsed, regardless of heap state. | +| `OMNIROUTE_CHAT_HEAP_SHED_RATIO` | `0.75` | `src/shared/middleware/chatBodyAdmission.ts` | Shed a large chat body with `503` + `Retry-After` once `heapUsed / heap_size_limit` reaches this ratio (`0 < r < 1`). Turns a process-wide V8 OOM under concurrent large compacts into a single graceful client retry; a healthy heap admits every body untouched. | +| `OMNIROUTE_MAX_NONSTREAMING_RESPONSE_BYTES` | `67108864` (64 MB) | `open-sse/handlers/chatCore/nonStreamingResponseBody.ts` | Hard cap for a non-streaming upstream response buffered fully into memory. Past this the upstream reader is cancelled and the request fails fast instead of growing an unbounded string until the heap is exhausted. | +| `CORS_ORIGIN` | _(unset)_ | `src/server/cors/origins.ts` | Legacy single-origin CORS allowlist. Prefer `CORS_ALLOWED_ORIGINS` for new deployments. CORS is only for cross-origin browser API clients; authenticated dashboard writes use same-origin requests plus session-bound CSRF protection instead. | | `CORS_ALLOWED_ORIGINS` | _(unset)_ | `src/server/cors/origins.ts` | Comma-separated CORS allowlist. No wildcard is sent unless `CORS_ALLOW_ALL=true` is explicitly configured. | | `CORS_ALLOW_ALL` | `false` | `src/server/cors/origins.ts` | Development-only escape hatch to echo any browser `Origin`. Do not enable on shared or production deployments. | | `OUTBOUND_SSRF_GUARD_ENABLED` | `true` | `src/shared/network/outboundUrlGuard.ts` | Block provider calls targeting private/loopback/link-local IP ranges. Disable only in isolated test envs. | @@ -254,10 +260,10 @@ OmniRoute provides a two-layer defense: request-side injection scanning and resp | `OMNIROUTE_CLOUD_SYNC_SECRET` | _(empty)_ | `src/lib/cloudSync.ts` | Shared secret used to verify the HMAC-SHA256 signature of Cloud Sync responses. | | `OMNIROUTE_CLOUD_SYNC_SECRETS` | `false` | `src/lib/cloudSync.ts` | Set to `true` to allow the Cloud Sync endpoint to overwrite local credentials. Default is `false`. | | `OMNIROUTE_ZED_IMPORT_LEGACY_ONE_STEP` | `false` | `src/app/api/providers/zed/import/route.ts` | Set to `true` to fall back to the v3.8.5 one-step "import everything" behavior without user confirmation. | -| `NEXT_PUBLIC_BASE_URL` | `http://localhost:20128` | OAuth, Dashboard, sync | Public-facing URL for OAuth redirect_uri, Dashboard links, generated public URLs, and same-origin browser mutation checks. **Must match your public URL behind reverse proxy.** | +| `NEXT_PUBLIC_BASE_URL` | `http://localhost:20128` | OAuth, Dashboard, sync | Public-facing URL for OAuth redirect_uri, Dashboard links, and generated public URLs. Set this to the stable public URL when OAuth callbacks or generated browser links must use a canonical reverse-proxy host. | | `NEXT_PUBLIC_CLOUD_URL` | _(empty)_ | Client-side | Client-side mirror of `CLOUD_URL`. | | `NEXT_PUBLIC_APP_URL` | _(unset)_ | `src/shared/services/cloudSyncScheduler.ts` | Legacy fallback for `NEXT_PUBLIC_BASE_URL`. | -| `OMNIROUTE_PUBLIC_BASE_URL` | _(unset)_ | Public-origin resolver, image URLs | Highest-priority browser-facing OmniRoute origin used for public URL generation and origin validation (for example `/v1/chatgpt-web/image/`). Set this when OpenWebUI or another relay reaches OmniRoute by an internal URL but the user's browser must fetch images from a LAN, tunnel, or public origin. Do **not** include `/v1`. | +| `OMNIROUTE_PUBLIC_BASE_URL` | _(unset)_ | Public-origin resolver, image URLs | Highest-priority browser-facing OmniRoute origin used for public URL generation and non-dashboard browser-origin validation (for example `/v1/chatgpt-web/image/`). Set this when OpenWebUI or another relay reaches OmniRoute by an internal URL but the user's browser must fetch images from a LAN, tunnel, or public origin. Do **not** include `/v1`. | | `OMNIROUTE_TRUST_PROXY` | _(unset)_ | `src/server/origin/publicOrigin.ts` | Optional trust mode for forwarded public-origin headers. Unset = do not trust `Forwarded` / `X-Forwarded-*` for security decisions. `true` / `loopback` trusts forwarded host/proto only from a token-stamped loopback proxy. `private` / `lan` also trusts private-LAN proxy peers. Prefer explicit `NEXT_PUBLIC_BASE_URL` in production. | | `OMNIROUTE_CGPT_WEB_IMAGE_TIMEOUT_MS` | `180000` (3 min) | `open-sse/executors/chatgpt-web.ts` | Max wait time for an async chatgpt-web image to land via the celsius WebSocket. Increase during upstream queue-deep windows. | | `OMNIROUTE_CGPT_WEB_IMAGE_CACHE_MAX_MB` | `256` | `open-sse/services/chatgptImageCache.ts` | Total in-memory byte budget (MB) for the chatgpt-web image cache serving `/v1/chatgpt-web/image/`. Lower on memory-constrained hosts; raise if image generation is heavy and clients race the 30-minute TTL. | @@ -282,11 +288,11 @@ OmniRoute provides a two-layer defense: request-side injection scanning and resp | `OMNIROUTE_CODEWHISPERER_BASE_URL` | `https://codewhisperer.us-east-1.amazonaws.com` | `open-sse/services/usage.ts` | CodeWhisperer (AWS Kiro) usage limits endpoint. Override for relays / test fixtures. | > [!IMPORTANT] -> When deploying behind a reverse proxy (nginx, Caddy), `NEXT_PUBLIC_BASE_URL` **must** be set to your public URL (e.g., `https://omniroute.example.com`). Without this, OAuth callbacks can fail because the redirect_uri won't match, generated public links can point at the internal container origin, and same-origin dashboard mutations can be rejected by browser-origin checks. +> When deploying behind a reverse proxy (nginx, Caddy), set `NEXT_PUBLIC_BASE_URL` to your stable public URL (e.g., `https://omniroute.example.com`) when OAuth callbacks or generated public links must use that hostname. Without this, OAuth callbacks can fail because the redirect_uri won't match and generated public links can point at the internal container origin. > > Keep `BASE_URL` as an internal loopback/container URL for server-to-server jobs. Do not use a browser `Origin` or public hostname for credential-bearing internal self-fetches. > -> OmniRoute centralizes public-origin validation: explicit public URL env vars are trusted first; raw `Forwarded` / `X-Forwarded-*` headers are ignored unless `OMNIROUTE_TRUST_PROXY` is enabled and the immediate proxy peer is token-stamped as trusted. Do not use CORS settings to fix same-origin dashboard requests; CORS is only for cross-origin browser clients. +> Authenticated dashboard writes do not require a static public base URL: the dashboard sends same-origin unsafe requests with a session-bound CSRF token. OmniRoute still centralizes public-origin validation for non-dashboard browser integrations: explicit public URL env vars are trusted first; raw `Forwarded` / `X-Forwarded-*` headers are ignored unless `OMNIROUTE_TRUST_PROXY` is enabled and the immediate proxy peer is token-stamped as trusted. Do not use CORS settings to fix same-origin dashboard requests; CORS is only for cross-origin browser clients. --- @@ -349,6 +355,16 @@ Controls how OmniRoute discovers and launches CLI sidecars (Claude Code, Codex, | `CLI_DEVIN_BIN` | `devin` | `open-sse/executors/devin-cli.ts` | Custom path to the Devin CLI binary (v3.8.0). Used by the Windsurf/Devin executor. | | `HERMES_HOME` | `~/.hermes` | `src/lib/cli-helper/config-generator/hermesHome.ts` | Hermes Agent home directory where OmniRoute reads/writes the Hermes CLI config. Matches the env var the Hermes PowerShell installer sets on Windows (`%LOCALAPPDATA%\hermes`). | +### CLI Profile Auto-Sync + +These feature flags are opt-in and default off. They can also be toggled from +the CLI Code dashboard. + +| Variable | Default | Source File | Description | +| --------------------------------------- | ------- | ----------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | +| `OMNIROUTE_AUTO_SYNC_CODEX_PROFILES` | `false` | `src/shared/constants/featureFlagDefinitions.ts` | After a provider model sync, automatically rewrites `~/.codex/*.config.toml` profile files from the live catalog. Requires `CLI_ALLOW_CONFIG_WRITES`; never changes the active/default Codex config, auth, Codex-lb settings, or provider choice. | +| `OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES` | `false` | `src/shared/constants/featureFlagDefinitions.ts` | After a provider model sync, automatically rewrites `~/.claude/profiles//settings.json` Claude Code profile files from the live catalog. Requires `CLI_ALLOW_CONFIG_WRITES`; never changes the active/default Claude config, auth, or provider choice. | + ### Docker Example ```bash @@ -402,6 +418,12 @@ detection above). | `OMNIROUTE_CONFIG_HOT_RELOAD_MS` | `5000` | `src/lib/config/hotReload.ts` | Polling interval (ms) for config hot-reload. Lower than `1000` is rejected. | | `OMNIROUTE_DISABLE_REDIS_AUTH_CACHE` | _(enabled)_ | `src/lib/db/apiKeys.ts` | Set `1` to bypass the Redis-backed API-key auth cache (forces DB reads). | | `OMNIROUTE_RTK_TRUST_PROJECT_FILTERS` | `0` | `open-sse/services/compression/engines/rtk/filterLoader.ts` | Trust user-managed RTK project filter rules without strict signature checks. | +| `COMPRESSION_PIPELINE_BREAKER_ENABLED` | `false` | `open-sse/services/compression/pipelineEngineBreaker.ts` | T02 stacked-pipeline per-engine circuit-breaker master switch. **Opt-in (default off)** — when on, an engine that throws repeatedly across requests is skipped (fail-open) for a cooldown; off = byte-identical legacy behavior. | +| `COMPRESSION_PIPELINE_BREAKER_THRESHOLD` | `3` | `open-sse/services/compression/pipelineEngineBreaker.ts` | Consecutive cross-request failures before an engine's breaker opens. | +| `COMPRESSION_PIPELINE_BREAKER_COOLDOWN_MS` | `30000` | `open-sse/services/compression/pipelineEngineBreaker.ts` | Milliseconds an opened engine stays skipped before a half-open probe. | +| `COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR` | `2` | `open-sse/services/compression/engines/ccr/index.ts` | T08/H8 CCR retrieval-feedback ramp: each prior retrieval of a stored block raises its effective `minChars` linearly (frequently-retrieved content compresses less; `>=3` retrievals = never compressed). `1` disables the ramp (binary skip at the threshold only). | +| `COMPRESSION_PREFIX_FREEZE_ENABLED` | `false` | `open-sse/services/compression/prefixFreeze.ts` | T08/H5 usage-observed prefix freeze master switch. **Opt-in (default off)** — when on, a system prompt observed `>=` the threshold is treated as a stable cacheable prefix and preserved from compression even for providers the static cache heuristic misses (freeze only *preserves*, never mutates). | +| `COMPRESSION_PREFIX_FREEZE_THRESHOLD` | `3` | `open-sse/services/compression/prefixFreeze.ts` | Observations of a system prompt before it is treated as a frozen stable prefix. | | `OMNIROUTE_BOOTSTRAPPED` | `false` | `src/app/(dashboard)/dashboard/page.tsx` | Set `true` by bootstrap script after initial setup. Controls setup wizard visibility. | | `OMNIROUTE_ALLOW_BODY_PROJECT_OVERRIDE` | `0` | `open-sse/executors/antigravity.ts` | Escape hatch: allow request body to override the Antigravity project field. | | `ANTIGRAVITY_CREDITS` | _(unset)_ | `open-sse/services/antigravityCredits.ts` | Override Antigravity's advertised remaining credits (testing / forced values). | @@ -571,6 +593,7 @@ REQUEST_TIMEOUT_MS (global override) │ └── FETCH_KEEPALIVE_TIMEOUT_MS (independent, default: 4000) ├─→ STREAM_IDLE_TIMEOUT_MS (inherits from REQUEST_TIMEOUT_MS, default: 600000) ├─→ STREAM_READINESS_TIMEOUT_MS (inherits from REQUEST_TIMEOUT_MS, default: 80000) +├─→ STREAM_READINESS_MAX_TIMEOUT_MS (caps adaptive readiness extensions, default: 180000) └─→ API_BRIDGE_PROXY_TIMEOUT_MS (inherits from REQUEST_TIMEOUT_MS, default: 30000) ├─→ API_BRIDGE_SERVER_REQUEST_TIMEOUT_MS (derived, default: 300000) ├── API_BRIDGE_SERVER_HEADERS_TIMEOUT_MS (default: 60000) @@ -584,6 +607,10 @@ REQUEST_TIMEOUT_MS (global override) | `FETCH_TIMEOUT_MS` | `600000` | Total HTTP request timeout for upstream provider calls. | | `STREAM_IDLE_TIMEOUT_MS` | `600000` | Max silence between SSE chunks before aborting. Extended-thinking models rarely pause >90s. | | `STREAM_READINESS_TIMEOUT_MS` | `80000` | Time to receive the first non-ping SSE event. Inherits `REQUEST_TIMEOUT_MS` when set. | +| `STREAM_READINESS_MAX_TIMEOUT_MS` | `180000` | Maximum adaptive first-event readiness window for large, tool-heavy, or high-reasoning streaming requests. | +| `OMNIROUTE_AGENT_GOAL_POLICY_ENABLED` | `true` | Kill-switch for the `/goal` heuristic. Set `false`/`0`/`off` to fully disable detection — readiness timeouts and stream recovery are never elevated by request body/headers, mitigating client-controlled timeout amplification. | +| `OMNIROUTE_AGENT_GOAL_READINESS_MAX_TIMEOUT_MS` | `600000` | Maximum first-event readiness window for detected `/goal` agent runs or requests forced with `x-omniroute-agent-goal`. | +| `OMNIROUTE_AGENT_GOAL_STREAM_RECOVERY` | `true` | Enable early stream recovery automatically for detected `/goal` agent runs. Set `false`/`0`/`off` to disable the goal-specific opt-in. This can only ADD recovery on top of the operator default — it never overrides an explicit `STREAM_RECOVERY_ENABLED`/DB settings opt-out. | | `OMNIROUTE_CODEX_DROP_NONSTANDARD_EVENTS` | _(off)_ | Strip non-standard `codex.*` SSE events (e.g. `codex.rate_limits`) that break the OpenAI SDK's `responses.stream()` with a 502. Set `true`/`1`/`yes` to enable. | | `FETCH_HEADERS_TIMEOUT_MS` | = `FETCH_TIMEOUT_MS` | Time to receive response headers. | | `FETCH_BODY_TIMEOUT_MS` | = `FETCH_TIMEOUT_MS` | Time to receive the full response body. | @@ -703,6 +730,10 @@ Embedding layer, vector store and reranking knobs for the persistent memory subs | `MEMORY_VEC_TOP_K` | `20` | Default top-K used by the `sqlite-vec` brute-force vector search inside `src/lib/memory/vectorStore.ts`. | | `MEMORY_RRF_K` | `60` | Reciprocal Rank Fusion constant `k` for hybrid FTS5 + vector retrieval (sqlite-vec recipe). | | `HF_HUB_ENDPOINT` | `https://huggingface.co` | Override Hugging Face Hub base URL used by `staticPotion.ts` (e.g. mirror endpoint for air-gapped setups). | +| `MEMORY_TYPED_DECAY_ENABLED` | `false` | TV6 typed memory decay master switch. **Opt-in (default off)** — the sweep **deletes** decayed memories. With it off, `access_count`/`last_accessed_at` are pure telemetry and nothing is ever deleted. | +| `MEMORY_TYPED_DECAY_EPISODIC_DAYS` | `30` | TTL (days) after which an unused `episodic` memory decays. `0` makes episodic immune too. Durable types (`factual`/`procedural`/`semantic`) are always immune. The decay clock re-bases on `last_accessed_at`. | +| `MEMORY_TYPED_DECAY_ACCESS_IMMUNITY` | `3` | A memory injected `>=` this many times becomes immune to decay regardless of type. `0` disables access immunity. | +| `MEMORY_TYPED_DECAY_SWEEP_INTERVAL` | `0` (disabled) | Interval (seconds) for the optional periodic decay sweep in `src/lib/memory/typedDecay.ts`. `0`/unset = no periodic sweep. Doubly opt-in: also requires `MEMORY_TYPED_DECAY_ENABLED=true`. | ### Low-RAM Docker Example @@ -740,9 +771,10 @@ Automatic model pricing data synchronization from external sources. ## 19. Model Sync (Dev) -| Variable | Default | Source File | Description | -| -------------------------- | ------------- | -------------------------- | -------------------------------------------------------- | -| `MODELS_DEV_SYNC_INTERVAL` | `86400` (24h) | `src/lib/modelsDevSync.ts` | Development-time model catalog sync interval in seconds. | +| Variable | Default | Source File | Description | +| ----------------------------------- | ------------- | ---------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | +| `MODELS_DEV_SYNC_INTERVAL` | `86400` (24h) | `src/lib/modelsDevSync.ts` | Development-time model catalog sync interval in seconds. | +| `CONTEXT_WINDOW_RECONCILE_INTERVAL` | `86400` (24h) | `src/lib/contextWindowResolver.ts` | Interval (seconds) for the self-correcting context-window reconciler (5004): pins provider-declared windows from `/models` discovery as `auto:discovery` overrides when they diverge from the catalog. Set to `0` to disable. Reuses already-synced data (no new fetch); never overwrites `manual` overrides. | --- diff --git a/docs/reference/PROVIDER_REFERENCE.md b/docs/reference/PROVIDER_REFERENCE.md index dbdee596c2..63733efa31 100644 --- a/docs/reference/PROVIDER_REFERENCE.md +++ b/docs/reference/PROVIDER_REFERENCE.md @@ -1,16 +1,16 @@ --- title: "Provider Reference" -version: 3.8.42 -lastUpdated: 2026-06-30 +version: 3.8.43 +lastUpdated: 2026-07-02 --- # Provider Reference > **Auto-generated** from `src/shared/constants/providers.ts` — do not edit by hand. > Regenerate with: `npm run gen:provider-reference` -> **Last generated:** 2026-06-30 +> **Last generated:** 2026-07-02 -Total providers: **236**. See category breakdown below. +Total providers: **237**. See category breakdown below. ## Categories @@ -64,16 +64,16 @@ Use the dashboard at `/dashboard/providers` to enable, configure, and test each | `blackbox-web` | `bb-web` | Blackbox Web (Subscription) | Web cookie | [link](https://app.blackbox.ai) | Paste your __Secure-authjs.session-token value or full cookie header from app.blackbox.ai | | `chatgpt-web` | `cgpt-web` | ChatGPT Web (Plus/Pro) | Web cookie | [link](https://chatgpt.com) | Paste your __Secure-next-auth.session-token cookie value from chatgpt.com | | `claude-web` | `cw` | Claude Web | Web cookie | [link](https://claude.ai) | Paste your session cookie from claude.ai | -| `copilot-m365-web` | `m365copilot` | Microsoft 365 Copilot (BizChat) | Web cookie | [link](https://m365.cloud.microsoft/chat) | Paste the access_token and account-specific Chathub path from the Microsoft 365 Copilot WebSocket URL. | +| `copilot-m365-web` | `m365copilot` | Microsoft 365 Copilot (BizChat) | Web cookie | [link](https://m365.cloud.microsoft/chat) | Sign in at m365.cloud.microsoft/chat, then open DevTools → Network → filter 'WS' → click the Chathub WebSocket connection. Copy both the access_token query parameter AND the account-specific Chathub path segment from its request URL (wss://…/Chathub/?…&access_token=…). It is NOT an Authorization: Bearer header on an XHR/Fetch request. The token is short-lived; this is an unofficial integration. | | `copilot-web` | `copilot` | Microsoft Copilot Web | Web cookie | [link](https://copilot.microsoft.com) | Paste your access_token from copilot.microsoft.com (or export a .har file from DevTools while logged in) | | `deepseek-web` | `ds-web` | DeepSeek Web | Web cookie | [link](https://chat.deepseek.com) | Paste your userToken from chat.deepseek.com — DevTools → Application → Local Storage → userToken | | `doubao-web` | `db` | Doubao Web (ByteDance) | Web cookie | [link](https://www.doubao.com) | Paste your session cookie from doubao.com (DevTools → Application → Cookies) | | `gemini-business` | `gembiz` | Gemini Business (Enterprise) | Web cookie | [link](https://business.gemini.google) | From your enterprise account: open business.gemini.google/home/cid/{your-cid}, then copy __Secure-1PSID and __Secure-1PSIDTS cookies from DevTools → Application → Cookies. Paste as a cookie header below. | | `gemini-web` | `gweb` | Gemini Web (Free) | Web cookie | [link](https://gemini.google.com) | Paste your __Secure-1PSID cookie value from gemini.google.com. Optionally add __Secure-1PSIDTS separated by semicolon. | | `grok-web` | `gw` | Grok Web (Subscription) | Web cookie | [link](https://grok.com) | Paste the full grok.com cookie line from DevTools → Application → Cookies. Include both `sso` and `sso-rw` (e.g. `sso=...; sso-rw=...`) — Grok's anti-bot rejects `sso` on its own. | -| `huggingchat` | `huggingchat` | HuggingChat (Free) | Web cookie | [link](https://huggingface.co/chat) | Paste your hf-chat cookie value from huggingface.co/chat (DevTools → Application → Cookies → hf-chat). Optional — works without auth for basic use. | +| `huggingchat` | `huggingchat` | HuggingChat (Free) | Web cookie | [link](https://huggingface.co/chat) | Paste the full Cookie header from huggingface.co/chat (DevTools → Network → /chat/conversation → Request Headers → Cookie). It should include hf-chat and may also include token / aws-waf-token. | | `inner-ai` | `in-ai` | Inner.ai (Subscription) | Web cookie | [link](https://app.innerai.com) | Paste your token cookie and email separated by a space: open DevTools → Application → Cookies → .innerai.com, copy the token value, then append a space and your Inner.ai login email. Example: eyJhbG... user@example.com | -| `kimi-web` | `kimi-web` | Kimi Web (Moonshot AI) | Web cookie | [link](https://kimi.moonshot.cn) | Paste your session cookie from kimi.moonshot.cn (DevTools → Application → Cookies) | +| `kimi-web` | `kimi-web` | Kimi Web (Moonshot AI) | Web cookie | [link](https://www.kimi.com) | Paste your Cookie header from www.kimi.com (must contain kimi-auth=...). Find it via DevTools → Network → request → Cookie. | | `lmarena` | `lma` | LMArena (Free) | Web cookie | [link](https://lmarena.ai) | Paste the full Cookie header from lmarena.ai (DevTools → Network → request → Cookie). The session is now split across arena-auth-prod-v1.0, .1, … — copy the whole header. Optional — works with free tier for basic comparisons. | | `muse-spark-web` | `ms-web` | Muse Spark Web (Meta AI) | Web cookie | [link](https://www.meta.ai) | Paste your ecto_1_sess value or full cookie header from meta.ai | | `perplexity-web` | `pplx-web` | Perplexity Web (Pro/Max) | Web cookie | [link](https://www.perplexity.ai) | Paste your __Secure-next-auth.session-token cookie value from perplexity.ai | @@ -92,8 +92,8 @@ Use the dashboard at `/dashboard/providers` to enable, configure, and test each | `agentrouter` | `agentrouter` | AgentRouter | API key, aggregator | [link](https://agentrouter.org) | $200 free credits on signup - multi-model routing gateway | | `ai21` | `ai21` | AI21 Labs | API key | [link](https://www.ai21.com) | $10 trial credits on signup (valid 3 months), no credit card required | | `aimlapi` | `aiml` | AI/ML API | API key, aggregator | [link](https://aimlapi.com) | Free tier paused (2026) — AI/ML API is now pay-as-you-go only (min $20 top-up); no recurring free credits. | -| `alibaba` | `ali` | Alibaba | API key | [link](https://dashscope-intl.aliyuncs.com) | — | -| `alibaba-cn` | `ali-cn` | Alibaba (China) | API key | [link](https://dashscope.aliyuncs.com) | — | +| `alibaba` | `ali` | Alibaba | API key | [link](https://bailian.console.alibabacloud.com/) | — | +| `alibaba-cn` | `ali-cn` | Alibaba (China) | API key | [link](https://dashscope.console.aliyun.com/) | — | | `anthropic` | `anthropic` | Anthropic | API key | [link](https://platform.claude.com) | — | | `api-airforce` | `af` | Api.airforce | API key | [link](https://api.airforce) | 55 free tier models including Grok-3, Claude 3.7, Qwen3, Kimi-K2, Gemini 2.5 Flash, DeepSeek-V3 | | `arcee-ai` | `arcee` | Arcee AI | API key | [link](https://arcee.ai) | Get API key at arcee.ai | @@ -110,7 +110,7 @@ Use the dashboard at `/dashboard/providers` to enable, configure, and test each | `bluesminds` | `bm` | BluesMinds | API key | [link](https://www.bluesminds.com) | Free daily pi credits — supports 200+ models including GPT-4o, GPT-4.1, Claude Sonnet 4.5, Gemini 2.0 Flash, DeepSeek V4, Qwen, Kimi K2 | | `byteplus` | `bpm` | BytePlus ModelArk | API key | [link](https://console.byteplus.com/ark) | — | | `bytez` | `bytez` | Bytez | API key | [link](https://bytez.com) | $1 free credits, refreshes every 4 weeks | -| `cablyai` | `cablyai` | CablyAI | API key, aggregator | [link](https://cablyai.com) | Bearer API key for the CablyAI OpenAI-compatible gateway. | +| `cablyai` | `cablyai` | CablyAI | API key, aggregator | [link](https://cablyai.com) | ⚠️ **DEPRECATED.** cablyai.com no longer resolves (DNS NXDOMAIN, verified 2026-06-30) — the domain is gone and every request fails with a DNS error (#5568). | | `cerebras` | `cerebras` | Cerebras | API key | [link](https://inference.cerebras.ai) | Free Trial: 1M tokens/day, 30K TPM, 5 RPM — no credit card. | | `chutes` | `chutes` | Chutes.ai | API key, aggregator | [link](https://chutes.ai) | Bearer API key for the Chutes OpenAI-compatible gateway. | | `clarifai` | `clarifai` | Clarifai | API key, enterprise | [link](https://docs.clarifai.com) | Use your Clarifai PAT or app-specific API key. OmniRoute targets the OpenAI-compatible endpoint at https://api.clarifai.com/v2/ext/openai/v1 and authenticates with Authorization: Key . | @@ -192,7 +192,7 @@ Use the dashboard at `/dashboard/providers` to enable, configure, and test each | `nscale` | `nscale` | nScale | API key | [link](https://nscale.com) | $5 free credits on signup for inference testing | | `nvidia` | `nvidia` | NVIDIA NIM | API key | [link](https://build.nvidia.com) | Free dev access: ~40 RPM, 70+ models (Kimi K2.5, GLM 4.7, DeepSeek V3.2...) | | `oci` | `oci` | OCI Generative AI | API key, enterprise | [link](https://www.oracle.com/artificial-intelligence/generative-ai) | Use your OCI Generative AI API key or IAM bearer token. Base URL can be https://inference.generativeai..oci.oraclecloud.com/openai/v1/. | -| `ollama-cloud` | `ollamacloud` | Ollama Cloud | API key | [link](https://ollama.com/settings/api-keys) | — | +| `ollama-cloud` | `ollamacloud` | Ollama Cloud | API key | [link](https://ollama.com/settings/keys) | — | | `openadapter` | `oad` | OpenAdapter | API key | [link](https://openadapter.dev) | Use your OpenAdapter API key in Authorization: Bearer sk-cv-. Fully OpenAI-compatible. API base URL: https://api.openadapter.in/v1. | | `openai` | `openai` | OpenAI | API key | [link](https://platform.openai.com) | — | | `opencode-go` | `opencode-go` | OpenCode Go | API key | [link](https://opencode.ai/go) | — | @@ -214,7 +214,7 @@ Use the dashboard at `/dashboard/providers` to enable, configure, and test each | `runwayml` | `runway` | Runway | API key, video | [link](https://docs.dev.runwayml.com) | Use your Runway API key in Authorization: Bearer . OmniRoute targets the current Runway API at https://api.dev.runwayml.com/v1 and sends the required X-Runway-Version header automatically. | | `sambanova` | `samba` | SambaNova | API key | [link](https://sambanova.ai) | $5 free credits on signup (30-day validity), no credit card required | | `sap` | `sap` | SAP Generative AI Hub | API key, enterprise | [link](https://help.sap.com/docs/sap-ai-core/sap-ai-core-service-guide/generative-ai-hub-in-sap-ai-core) | Use your SAP AI Core bearer token. Base URL can be your AI_API_URL root or a deploymentUrl from Generative AI Hub. | -| `scaleway` | `scw` | Scaleway AI | API key | [link](https://www.scaleway.com/en/ai/generative-apis) | 1M free tokens for new accounts — EU/GDPR compliant (Paris), Qwen3 235B & Llama 70B | +| `scaleway` | `scw` | Scaleway AI | API key | [link](https://www.scaleway.com/en/docs/ai-data/generative-apis/) | 1M free tokens for new accounts — EU/GDPR compliant (Paris), Qwen3 235B & Llama 70B | | `sensenova` | `sensenova` | SenseNova | API key | [link](https://platform.sensenova.cn) | Get API key at platform.sensenova.cn | | `siliconflow` | `siliconflow` | SiliconFlow | API key | [link](https://cloud.siliconflow.com) | $1 free credits plus permanently free models after identity verification | | `snowflake` | `snowflake` | Snowflake Cortex | API key, enterprise | [link](https://www.snowflake.com) | — | @@ -225,7 +225,7 @@ Use the dashboard at `/dashboard/providers` to enable, configure, and test each | `synthetic` | `synthetic` | Synthetic | API key, aggregator | [link](https://synthetic.new) | — | | `tencent` | `tencent` | Tencent Hunyuan | API key | [link](https://hunyuan.tencent.com) | Get API key at console.cloud.tencent.com | | `thebai` | `thebai` | TheB.AI | API key, aggregator | [link](https://theb.ai) | Bearer API key for the TheB.AI OpenAI-compatible gateway. | -| `together` | `together` | Together AI | API key, video | [link](https://www.together.ai) | $25 signup credits + 3 permanently free models: Llama 3.3 70B, Vision, DeepSeek-R1 distill | +| `together` | `together` | Together AI | API key, video | [link](https://www.together.ai) | — | | `tokenrouter` | `trk` | TokenRouter | API key | [link](https://tokenrouter.com) | Use your TokenRouter API key in Authorization: Bearer . Fully OpenAI-compatible. API base URL: https://api.tokenrouter.com/v1. | | `topaz` | `topaz` | Topaz | API key, image | [link](https://topazlabs.com) | — | | `udio` | `udio` | Udio | API key | [link](https://udio.com) | Paste session cookie from udio.com (Supabase auth) | @@ -247,7 +247,7 @@ Use the dashboard at `/dashboard/providers` to enable, configure, and test each | `zai` | `zai` | Z.AI | API key | [link](https://open.bigmodel.cn) | — | | `zenmux` | `zm` | ZenMux | API key | [link](https://zenmux.ai) | Use your ZenMux API key in Authorization: Bearer . ZenMux is fully OpenAI-compatible. Base URL: https://zenmux.ai/api/v1. | -## Local Providers (11) +## Local Providers (12) | ID | Alias | Name | Tags | Website | Notes | |----|-------|------|------|---------|-------| @@ -257,6 +257,7 @@ Use the dashboard at `/dashboard/providers` to enable, configure, and test each | `llama-cpp` | `llamacpp` | llama.cpp | Local, self-hosted | [link](https://github.com/ggml-org/llama.cpp) | API key optional (use any value, e.g. sk-no-key-required). Configure the llama-server OpenAI-compatible base URL (default: http://127.0.0.1:8080/v1). Note: if Llamafile is also installed, both default to port 8080 — run only one at a time or override the port. | | `llamafile` | `llamafile` | Llamafile | Local, self-hosted | [link](https://github.com/Mozilla-Ocho/llamafile) | API key optional. Configure the local Llamafile OpenAI-compatible base URL (default: http://127.0.0.1:8080/v1). | | `lm-studio` | `lmstudio` | LM Studio | Local, self-hosted | [link](https://lmstudio.ai) | API key optional. Configure the local LM Studio OpenAI-compatible base URL (default: http://localhost:1234/v1). | +| `ollama-local` | `ollama` | Ollama | Local, self-hosted | [link](https://ollama.com) | No API key required. Ollama runs locally — configure its OpenAI-compatible base URL (default: http://localhost:11434/v1) and make sure Ollama is running before connecting. | | `oobabooga` | `ooba` | oobabooga | Local, self-hosted | [link](https://github.com/oobabooga/text-generation-webui) | API key optional. Configure the local oobabooga OpenAI-compatible base URL (default: http://localhost:5000/v1). | | `sdwebui` | `sdwebui` | SD WebUI | Local | [link](https://github.com/AUTOMATIC1111/stable-diffusion-webui) | No API key required. Configure the local WebUI base URL (default: http://localhost:7860). | | `triton` | `triton` | NVIDIA Triton | Local, self-hosted | [link](https://developer.nvidia.com/triton-inference-server) | API key optional. Configure the Triton OpenAI-compatible base URL (default: http://localhost:8000/v1). | @@ -271,13 +272,13 @@ Use the dashboard at `/dashboard/providers` to enable, configure, and test each | `exa-search` | `exa-search` | Exa Search | Search | [link](https://exa.ai) | API key from dashboard.exa.ai | | `google-pse-search` | `google-pse` | Google Programmable Search | Search | [link](https://developers.google.com/custom-search/v1/overview) | Requires a Google API key and your Programmable Search Engine ID (cx) | | `linkup-search` | `linkup` | Linkup Search | Search | [link](https://docs.linkup.so) | Bearer API key from the Linkup dashboard | -| `ollama-search` | `ollama-search` | Ollama Search | Search | [link](https://ollama.com/settings/api-keys) | Same API key as Ollama Cloud (from ollama.com/settings/api-keys) | +| `ollama-search` | `ollama-search` | Ollama Search | Search | [link](https://ollama.com/settings/keys) | Same API key as Ollama Cloud (from ollama.com/settings/keys) | | `perplexity-search` | `pplx-search` | Perplexity Search | Search | [link](https://docs.perplexity.ai/guides/search-quickstart) | Same API key as Perplexity (pplx-...) | -| `searchapi-search` | `searchapi` | SearchAPI | Search | [link](https://www.searchapi.io/docs) | API key from SearchAPI (query param or Bearer auth) | +| `searchapi-search` | `searchapi` | SearchAPI | Search | [link](https://www.searchapi.io/docs/google) | API key from SearchAPI (query param or Bearer auth) | | `searxng-search` | `searxng` | SearXNG Search | Search | [link](https://docs.searxng.org) | API key is optional. Set your SearXNG base URL. Some instances may require a bearer token for access. | | `serper-search` | `serper-search` | Serper Search | Search | [link](https://serper.dev) | API key from serper.dev dashboard | | `tavily-search` | `tavily-search` | Tavily Search | Search | [link](https://tavily.com) | API key from app.tavily.com (format: tvly-...) | -| `youcom-search` | `youcom-search` | You.com Search | Search | [link](https://you.com/docs/search/overview) | X-API-Key from the You.com platform dashboard | +| `youcom-search` | `youcom-search` | You.com Search | Search | [link](https://you.com/business/api/) | X-API-Key from the You.com platform dashboard | ## Audio-only Providers (7) diff --git a/docs/reference/RELAY_BACKEND_STRATEGY.md b/docs/reference/RELAY_BACKEND_STRATEGY.md index e497d98bf0..6db4942e74 100644 --- a/docs/reference/RELAY_BACKEND_STRATEGY.md +++ b/docs/reference/RELAY_BACKEND_STRATEGY.md @@ -45,6 +45,20 @@ If you are currently comparing 9router/CLIPROXYAPI: - If a workflow needs strict sidecar behavior and lower per-request variance, use `OMNIROUTE_RELAY_BACKEND=bifrost`. - If you need sidecar resilience with graceful degradation under incident conditions, use `OMNIROUTE_RELAY_BACKEND=auto`. +## Backend boundary contract + +The stable product boundary is the OmniRoute relay API, not the dashboard implementation. The Next.js dashboard may install, configure, and supervise local services, but request routing should enter through the relay API and hand off behind that boundary. + +Long-term backend choices: + +- Keep the TypeScript relay as the in-process policy and fallback path. Auth, allowlists, request normalization, accounting, and safety checks stay here before any backend handoff. +- Use Bifrost as the preferred high-throughput Tier-1 sidecar when the deployment needs lower routing variance, centralized provider rotation, or scale-out across multiple OmniRoute replicas. +- Keep 9router and CLIPROXYAPI as embedded compatibility services. They run as supervised local processes and are useful when their provider/CLI behavior is the desired adapter, but they should not become the default Tier-1 routing engine. +- Do not make the dashboard pass arbitrary service URLs into the hot path. UI-provided URLs are configuration inputs; routing code should resolve registered, health-checked backends from server-side settings and supervisor state. +- Prefer loopback HTTP for supervised services today because the managed processes already expose HTTP-compatible APIs, the route guard can audit the boundary, and failure/fallback behavior is visible in request logs. A future SDK or socket transport is only worth adding if it measurably reduces p99 routing latency without weakening isolation or fallback semantics. + +For a very high-throughput deployment, the default answer is therefore `auto` with Bifrost enabled: use the Go sidecar on the hot path while preserving the TypeScript fallback for success rate. Use `bifrost` only when strict sidecar-only behavior is more important than graceful degradation. + ## High-throughput guidance For sustained high RPM/RPS and strict success SLO: diff --git a/docs/security/BAN_DETECTION.md b/docs/security/BAN_DETECTION.md new file mode 100644 index 0000000000..015aa366cb --- /dev/null +++ b/docs/security/BAN_DETECTION.md @@ -0,0 +1,137 @@ +--- +title: Account-Ban / Banned-Keyword Detection +--- + +# Account-Ban / Banned-Keyword Detection + +OmniRoute scans upstream error responses for signals that indicate a provider +**account is permanently dead** (suspended / deactivated / ToS-banned) and, when +matched, moves that connection into a **terminal `banned` state** so it is no +longer selected for requests. This is what the **Security → Banned Keywords** +settings card configures ("Additional keywords that trigger permanent account +ban detection. Built-in keywords always apply."). + +This page documents the built-in list, the detection flow, its scope, how to add +custom keywords safely, and how to recover a flagged connection. The terminal +state itself is part of the resilience model — see +[RESILIENCE_GUIDE](../architecture/RESILIENCE_GUIDE.md) ("Terminal states"). + +**Source of truth:** `open-sse/services/accountFallback.ts` +(`ACCOUNT_DEACTIVATED_SIGNALS`, `getMergedBannedSignals()`, `isAccountDeactivated()`). + +## Built-in keywords + +These 8 substrings always apply (case-insensitive), regardless of any custom list: + +``` +account_deactivated +account has been deactivated +account has been disabled +your account has been suspended +this account is deactivated +verify your account to continue (Antigravity / Google Cloud Code) +this service has been disabled in this account for violation (Antigravity) +this service has been disabled in this account (Antigravity) +``` + +> This list evolves as providers change their ban wording. The authoritative +> copy is `ACCOUNT_DEACTIVATED_SIGNALS` in `open-sse/services/accountFallback.ts`; +> treat the block above as a snapshot. + +Two adjacent, **separate** signal tables live in the same file and are *not* part +of banned-keyword detection: + +- `CREDITS_EXHAUSTED_SIGNALS` — billing/quota depleted (`insufficient_quota`, + `credit_balance_too_low`, `payment required`, …) → terminal `credits_exhausted`. +- `OAUTH_INVALID_TOKEN_SIGNALS` — **non-terminal**; a token refresh can recover. + +Note: common transient phrases like **`rate limit`** / `429` are handled by the +rate-limit / connection-cooldown path and are **not** ban signals. + +## Detection flow + +``` +upstream error response + → body stringified + lowercased + → isAccountDeactivated(body): getMergedBannedSignals().some(sig => body.includes(sig)) [substring match] + → match? + → connection testStatus = "banned" (permanent — 1-year cooldown, never auto-recovers) + → if setting `autoDisableBannedAccounts` is on → also isActive = false + → connection is skipped during account selection (combo QUOTA_BLOCKING statuses) +``` + +- The match is a **case-insensitive substring** search on the response **body** + (`isAccountDeactivated`, `accountFallback.ts`). +- The permanent `banned` terminalization fires on a banned-signal body at **any + HTTP status** (via `markAccountUnavailable` → `checkFallbackError`). The + narrower **`deactivated`** label (`isActive=false` when the connection has no + spare API keys) is written by the inline `chatCore.ts` path on **HTTP 401 / 403** + (classified via `classifyProviderError` → `ACCOUNT_DEACTIVATED`). Note the + `markAccountUnavailable()` path writes a *different* terminal status — + **`expired`** — for the same `ACCOUNT_DEACTIVATED` signal (via + `resolveTerminalConnectionStatus`), so the same ban can surface as either + `deactivated` or `expired` depending on which path handled the response. (The + older code comment says "when a 401 body contains these strings" — that + understates the current behavior.) +- A `banned` connection is excluded from selection everywhere terminal statuses + are filtered (`isTerminalConnectionStatus`, combo `QUOTA_BLOCKING_CONNECTION_STATUSES`). + +## Scope — which providers are scanned + +**All providers.** The check runs in the generic error-handling pipeline that +every failed upstream request flows through — it is **not** gated to +OAuth/subscription scrapers. The resulting terminal state is per **connection**, +not per provider. + +That said, the built-in *strings* are oriented toward subscription/OAuth +providers with real ban risk (ChatGPT Web, Claude Web, Codex, Muse Spark, +Antigravity). An API-key provider will only trip the detector if its error body +literally contains one of the substrings. + +## Custom banned keywords + +Add or remove keywords in **Security → Banned Keywords** (persisted as the global +`customBannedSignals` setting via `PATCH /api/settings`). They are **added to** +the built-in list — never a replacement — and hot-reload on save (and at startup) +via `setCustomBannedSignals()`. Each keyword is capped at 200 characters; there is +no array-length limit. + +**⚠ False-positive risk — choose specific phrases.** Detection is a raw substring +match on the whole response body, and a match is **permanent** (1-year cooldown, +manual recovery). A broad keyword can ban a perfectly healthy connection: + +- **Bad:** `quota`, `limit`, `error`, `denied` — appear in many transient errors. +- **Good:** full ban sentences, e.g. `your account has been suspended for`, + `account permanently banned`, `violation of our terms`. + +Prefer the longest unambiguous phrase the provider returns on a real ban. When in +doubt, watch the connection's `lastError` first, then add the exact wording. + +## Recovering a flagged connection + +Terminal `banned` / `deactivated` states **never auto-recover** (they are excluded +from the proactive-recovery tick — only `unavailable` cooldowns recover on their +own). An operator must clear them explicitly: + +1. **Re-test the connection** — the dashboard **Test** action + (`POST /api/providers/{id}/test`); a successful probe resets `testStatus` to + `active` and clears the error fields. +2. **Re-authenticate / edit credentials** — for OAuth providers, re-run the login + / refresh flow; provider create/import routes set `isActive = true`. +3. **Re-enable the connection** — if `autoDisableBannedAccounts` set + `isActive = false`, toggle it back on after fixing the account. + +There is no separate "clear ban flag" button — recovery is re-test, re-auth, or +re-enable, matching the general terminal-state rule in +[RESILIENCE_GUIDE](../architecture/RESILIENCE_GUIDE.md). + +## Source files + +| Concern | File | +| --- | --- | +| Signal tables + match | `open-sse/services/accountFallback.ts` | +| Terminalization / persistence | `src/sse/services/auth.ts` (`markAccountUnavailable`, `resolveTerminalConnectionStatus`, `clearAccountError`) | +| Inline classification | `open-sse/handlers/chatCore.ts`, `open-sse/services/errorClassifier.ts` | +| Terminal-state recovery exclusion | `src/lib/quota/connectionRecovery.ts` | +| Custom-keyword runtime load | `src/lib/config/runtimeSettings.ts` (`setCustomBannedSignals`) | +| Settings UI | `src/app/(dashboard)/dashboard/settings/components/SecurityTab.tsx` | diff --git a/docs/security/ROUTE_GUARD_TIERS.md b/docs/security/ROUTE_GUARD_TIERS.md index da7eb1d0c2..df04831a39 100644 --- a/docs/security/ROUTE_GUARD_TIERS.md +++ b/docs/security/ROUTE_GUARD_TIERS.md @@ -22,13 +22,39 @@ API key with the `manage` scope (see [Manage-scope carve-out](#manage-scope-carv These routes spawn child processes or execute runtime code. Exposing them to non-loopback traffic would allow an attacker who obtained a valid JWT (e.g., via a Cloudflared/Ngrok tunnel) to trigger process spawning — a known CVE -class (GHSA-fhh6-4qxv-rpqj). +class ([GHSA-fhh6-4qxv-rpqj](https://github.com/advisories/GHSA-fhh6-4qxv-rpqj)). -| Prefix | Reason | Bypassable by `manage`? | -| ------------------------- | --------------------------------------------------------- | ----------------------- | -| `/api/mcp/` | MCP server — spawns stdio bridges and SSE handlers | Yes | -| `/api/cli-tools/runtime/` | CLI tool runtime — executes arbitrary plugin code | No (strict-loopback) | -| `/api/services/` | Embedded services (9router, CLIProxy) — npm install+spawn | No (strict-loopback) | +**What GHSA-fhh6-4qxv-rpqj is (the attack class):** a management/agent server +exposes an endpoint that launches a subprocess (`npm install`, `node`, a browser, +a proxy, `git`, `tar`, …). If that endpoint is reachable from off-host — because +the operator put OmniRoute behind an nginx/Cloudflare/Tailscale tunnel and a JWT +leaked, or auth was misconfigured — the attacker turns "call an API" into "run a +command on the host" (remote code execution). OmniRoute closes this by enforcing a +**loopback host check unconditionally, before any auth check**, on every +spawn-capable route: a leaked token over a tunnel still can't reach the spawn. + +**The full LOCAL_ONLY set.** The authoritative source is +`LOCAL_ONLY_API_PREFIXES` / `LOCAL_ONLY_API_PATTERNS` in +`src/server/authz/routeGuard.ts`; the table below mirrors the current state. The +`check-route-guard-membership` gate enumerates every `route.ts` under the +spawn-capable prefixes and fails CI if any is not classified local-only. + +| Prefix / pattern | Why it's local-only | Manage-scope bypassable? | +| ----------------------------------- | ---------------------------------------------------------------------------------------- | ----------------------------- | +| `/api/mcp/` | MCP server — spawns stdio bridges + SSE handlers | **Yes** (only one) | +| `/api/cli-tools/runtime/` | CLI tool runtime — executes arbitrary plugin code | No — spawn-capable | +| `/api/services/` | Embedded services (9router/CLIProxy) — `npm install` + spawn | No — spawn-capable | +| `/dashboard/providers/services/` | Reverse proxy to embedded-service UIs | No | +| `/api/copilot/` | Unauthenticated LLM driver — CLI-only by default | Operator opt-in: manage/admin | +| `/api/tools/agent-bridge/` | AgentBridge — spawns MITM server + DNS edits | No — spawn-capable | +| `/api/tools/traffic-inspector/` | Traffic Inspector — http-proxy listener + system proxy | No — spawn-capable | +| `/api/plugins/`, `/api/plugins` | Plugins — load/execute via `worker_threads` + `child_process` | No — spawn-capable | +| `/api/system/version` | Auto-update (POST only; GET/HEAD/OPTIONS exempt) — spawns `git checkout` + `npm install` | No | +| `/api/db-backups/exportAll` | Spawns `tar` for the export archive | No | +| `/api/local/` | 1-click local launchers (Redis today) — spawns podman/docker | No — spawn-capable | +| `/api/headroom/start`, `/stop` | Headroom proxy lifecycle — spawns python CLI / signals PID | No — spawn-capable | +| `/api/oauth/cursor/auto-import` | `execFile("which", ["cursor"])` before importing creds | No | +| `/api/providers/{id}/login` (regex) | Launches a headful Playwright Chromium for web-cookie login | No | **Response on violation:** `403 LOCAL_ONLY` @@ -55,6 +81,35 @@ LOCAL_ONLY tier exists to prevent. | Non-loopback, Bearer with `manage` scope | `/api/cli-tools/runtime/*` | 403 LOCAL_ONLY | | Loopback, any/no Bearer | any LOCAL_ONLY | Allow (gate passes) | +#### Operator guidance & auditing + +If you run OmniRoute behind a reverse proxy or tunnel (nginx, Caddy, Cloudflare +Tunnel, Tailscale, Ngrok), the loopback check still protects the spawn-capable +routes above — a request whose client address is non-loopback is rejected with +`403 LOCAL_ONLY` **before auth runs**, so a leaked JWT can't reach a spawn. Two +operator responsibilities remain: + +- **Do not "fix" a 403 by forging the client IP as loopback.** Setting + `X-Forwarded-For: 127.0.0.1`, or a proxy that rewrites the source address to + loopback, re-opens exactly the RCE class this tier closes. Expose the + dashboard/API through the proxy — never the spawn-capable routes. +- **Keep the manage-scope bypass minimal.** Only `/api/mcp/` is bypassable, and + only with a `manage`-scoped API key. The `SPAWN_CAPABLE_PREFIXES` can never be + added to the bypass list — the zod schema rejects them and + `isLocalOnlyBypassableByManageScope` denies them at runtime (defence-in-depth), + which is what the dashboard means by "cannot be made bypassable". + +**Auditing access** — to verify nothing off-host is reaching these routes: + +- Open the **Authorization Inventory** on `/dashboard/settings/security`: it renders the + live LOCAL_ONLY prefix list, which prefixes are bypassable, and the compile-time + spawn-capable ("cannot be made bypassable") set. +- Grep your reverse-proxy / access logs for the prefixes above paired with a + non-loopback client address. Any such hit that returned `200` instead of + `403 LOCAL_ONLY` means the proxy is masking the real client IP — fix the proxy. +- A `403 LOCAL_ONLY` in OmniRoute's logs for one of these paths is the guard + working as intended, not an error to suppress. + ### Tier 2 — ALWAYS_PROTECTED **Enforced by:** `isAlwaysProtectedPath(path)` → skip `requireLogin=false` bypass diff --git a/docs/security/meta.json b/docs/security/meta.json index 14401f185e..acfb4c28c4 100644 --- a/docs/security/meta.json +++ b/docs/security/meta.json @@ -5,6 +5,7 @@ "PUBLIC_CREDS", "ERROR_SANITIZATION", "ROUTE_GUARD_TIERS", + "BAN_DETECTION", "STEALTH_GUIDE", "EGRESS_POLICY", "COMPLIANCE", diff --git a/electron/package-lock.json b/electron/package-lock.json index d3433a7923..a2073848e4 100644 --- a/electron/package-lock.json +++ b/electron/package-lock.json @@ -1,12 +1,12 @@ { "name": "omniroute-desktop", - "version": "3.8.42", + "version": "3.8.43", "lockfileVersion": 3, "requires": true, "packages": { "": { "name": "omniroute-desktop", - "version": "3.8.42", + "version": "3.8.43", "license": "MIT", "dependencies": { "electron-updater": "^6.8.9" diff --git a/electron/package.json b/electron/package.json index ac49ce5b78..0563f35ae5 100644 --- a/electron/package.json +++ b/electron/package.json @@ -1,6 +1,6 @@ { "name": "omniroute-desktop", - "version": "3.8.42", + "version": "3.8.43", "description": "OmniRoute Desktop Application", "main": "main.js", "author": { diff --git a/next.config.mjs b/next.config.mjs index 220e747be0..24bbc30de5 100644 --- a/next.config.mjs +++ b/next.config.mjs @@ -82,6 +82,16 @@ const minimalBuildAliases = isMinimalBuild } : {}; +function readTimeoutMs(...values) { + for (const value of values) { + const normalized = typeof value === "string" ? value.trim() : value; + if (normalized == null || normalized === "") continue; + const parsed = Number(normalized); + if (Number.isFinite(parsed) && parsed >= 0) return Math.floor(parsed); + } + return 600_000; +} + /** @type {import('next').NextConfig} */ const nextConfig = { distDir, @@ -114,6 +124,10 @@ const nextConfig = { // uploads (OpenAI-compatible /v1/files) routinely exceed this. Match the // 512 MB server-side cap; tune via env if needed. proxyClientMaxBodySize: process.env.NEXT_PROXY_BODY_LIMIT || "512mb", + // Next's internal router proxy defaults to 30s when this is unset. OmniRoute + // can legitimately hold non-streaming chat requests open for minutes while an + // upstream provider finishes, so reuse the existing request-timeout knobs. + proxyTimeout: readTimeoutMs(process.env.REQUEST_TIMEOUT_MS, process.env.FETCH_TIMEOUT_MS), // PR-2 of diegosouzapw/OmniRoute#3932: tree-shake barrel re-exports so // route bundles don't pull in 14 locale files, every lucide-react icon, // or the full date-fns surface when only one helper is used. diff --git a/open-sse/config/azureAi.ts b/open-sse/config/azureAi.ts index ce573d8552..fedbb99054 100644 --- a/open-sse/config/azureAi.ts +++ b/open-sse/config/azureAi.ts @@ -1,11 +1,7 @@ -import { stripTrailingSlashes } from "../utils/urlSanitize.ts"; +import { stripTrailingSlashes, normalizeBaseUrl } from "../utils/urlSanitize.ts"; export const AZURE_AI_DEFAULT_BASE_URL = "https://example-resource.services.ai.azure.com/openai/v1"; -function normalizeBaseUrl(value: string | null | undefined): string { - return stripTrailingSlashes((value || "").trim()); -} - export function normalizeAzureAiBaseUrl(value: string | null | undefined): string { const normalized = normalizeBaseUrl(value || AZURE_AI_DEFAULT_BASE_URL); if (!normalized) return AZURE_AI_DEFAULT_BASE_URL; diff --git a/open-sse/config/constants.ts b/open-sse/config/constants.ts index 6ce4e3e1f5..881ce9fbd2 100644 --- a/open-sse/config/constants.ts +++ b/open-sse/config/constants.ts @@ -22,6 +22,12 @@ export const STREAM_IDLE_TIMEOUT_MS = upstreamTimeouts.streamIdleTimeoutMs; // conservative for large prompts and slow first-byte reasoning providers. export const STREAM_READINESS_TIMEOUT_MS = upstreamTimeouts.streamReadinessTimeoutMs; +// Upper bound for adaptive stream readiness extensions (large histories, +// tool-heavy requests, high-reasoning Codex targets). Override with +// STREAM_READINESS_MAX_TIMEOUT_MS when an operator needs longer first-event +// windows for slow-thinking agent workloads. +export const STREAM_READINESS_MAX_TIMEOUT_MS = upstreamTimeouts.streamReadinessMaxTimeoutMs; + // Error code used when an upstream Antigravity request stalls before response // headers are returned. Keep it shared so executor, core normalization and // account fallback detection cannot drift. diff --git a/open-sse/config/datarobot.ts b/open-sse/config/datarobot.ts index 30a23a79ca..505f9a4299 100644 --- a/open-sse/config/datarobot.ts +++ b/open-sse/config/datarobot.ts @@ -1,4 +1,4 @@ -import { stripTrailingSlashes } from "../utils/urlSanitize.ts"; +import { stripTrailingSlashes, normalizeBaseUrl } from "../utils/urlSanitize.ts"; const DATAROBOT_API_V2_SEGMENT = "/api/v2"; const DATAROBOT_LLMGW_CHAT_PATH = "/genai/llmgw/chat/completions/"; @@ -6,10 +6,6 @@ const DATAROBOT_LLMGW_CATALOG_PATH = "/genai/llmgw/catalog/"; export const DATAROBOT_DEFAULT_BASE_URL = "https://app.datarobot.com"; -function normalizeBaseUrl(value: string | null | undefined): string { - return stripTrailingSlashes((value || "").trim()); -} - export function normalizeDataRobotBaseUrl(value: string | null | undefined): string { const normalized = normalizeBaseUrl(value || DATAROBOT_DEFAULT_BASE_URL); return normalized || DATAROBOT_DEFAULT_BASE_URL; diff --git a/open-sse/config/freeModelCatalog.data.ts b/open-sse/config/freeModelCatalog.data.ts index 791eec8c43..7074fcb3dd 100644 --- a/open-sse/config/freeModelCatalog.data.ts +++ b/open-sse/config/freeModelCatalog.data.ts @@ -195,10 +195,30 @@ export const FREE_MODEL_BUDGETS: FreeModelBudget[] = [ { provider: "hackclub", modelId: "meta-llama/llama-3.3-70b-instruct", displayName: "Llama 3.3 70B", monthlyTokens: 0, creditTokens: 0, freeType: "keyless", poolKey: "hackclub", tos: "caution" }, { provider: "hackclub", modelId: "mistralai/mistral-7b-instruct", displayName: "Mistral 7B", monthlyTokens: 0, creditTokens: 0, freeType: "keyless", poolKey: "hackclub", tos: "caution" }, { provider: "hackclub", modelId: "deepseek-ai/deepseek-coder-33b", displayName: "DeepSeek Coder 33B", monthlyTokens: 0, creditTokens: 0, freeType: "keyless", poolKey: "hackclub", tos: "caution" }, - { provider: "huggingchat", modelId: "meta-llama/Llama-3.3-70B-Instruct", displayName: "Llama 3.3 70B", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, - { provider: "huggingchat", modelId: "Qwen/Qwen2.5-72B-Instruct", displayName: "Qwen 2.5 72B", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, - { provider: "huggingchat", modelId: "mistralai/Mistral-Small-24B-Instruct-2501", displayName: "Mistral Small 24B", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, - { provider: "huggingchat", modelId: "deepseek-ai/DeepSeek-R1", displayName: "DeepSeek R1", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "baidu/ERNIE-4.5-VL-424B-A47B-Base-PT", displayName: "ERNIE 4.5 VL 424B A47B Base PT", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "CohereLabs/c4ai-command-r7b-12-2024", displayName: "Command R7B 12-2024", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "CohereLabs/command-a-reasoning-08-2025", displayName: "Command A Reasoning 08-2025", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "CohereLabs/command-a-vision-07-2025", displayName: "Command A Vision 07-2025", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "deepseek-ai/DeepSeek-V4-Pro", displayName: "DeepSeek V4 Pro", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "deepseek-ai/DeepSeek-V4-Flash", displayName: "DeepSeek V4 Flash", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "google/gemma-4-31B-it", displayName: "Gemma 4 31B", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "google/gemma-4-26B-A4B-it", displayName: "Gemma 4 26B A4B", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "inclusionAI/Ling-2.6-1T", displayName: "Ling 2.6 1T", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "meta-llama/Llama-4-Scout-17B-16E-Instruct", displayName: "Llama 4 Scout 17B 16E Instruct", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8", displayName: "Llama 4 Maverick 17B 128E Instruct FP8", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "MiniMaxAI/MiniMax-M3", displayName: "MiniMax M3", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "moonshotai/Kimi-K2.7-Code", displayName: "Kimi K2.7 Code", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "moonshotai/Kimi-K2.6", displayName: "Kimi K2.6", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4", displayName: "NVIDIA Nemotron 3 Ultra 550B A55B NVFP4", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "openai/gpt-oss-120b", displayName: "GPT-OSS 120B", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "openai/gpt-oss-20b", displayName: "GPT-OSS 20B", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "Qwen/Qwen3.5-122B-A10B", displayName: "Qwen3.5 122B A10B", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "Qwen/Qwen3.5-397B-A17B", displayName: "Qwen3.5 397B A17B", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "Qwen/Qwen3.6-27B", displayName: "Qwen3.6 27B", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "Qwen/Qwen3.6-35B-A3B", displayName: "Qwen3.6 35B A3B", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "stepfun-ai/Step-3.7-Flash", displayName: "Step 3.7 Flash", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "XiaomiMiMo/MiMo-V2.5-Pro", displayName: "MiMo V2.5 Pro", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, + { provider: "huggingchat", modelId: "zai-org/GLM-5.2", displayName: "GLM 5.2", monthlyTokens: 500000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingchat", tos: "caution" }, { provider: "huggingface", modelId: "meta-llama/llama-3.1-8b-instruct", displayName: "Llama 3.1 8B", monthlyTokens: 200000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingface", tos: "caution" }, { provider: "huggingface", modelId: "meta-llama/llama-3.2-11b-instruct", displayName: "Llama 3.2 11B", monthlyTokens: 200000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingface", tos: "caution" }, { provider: "huggingface", modelId: "mistralai/mistral-7b-instruct", displayName: "Mistral 7B", monthlyTokens: 200000, creditTokens: 0, freeType: "recurring-monthly", poolKey: "huggingface", tos: "caution" }, diff --git a/open-sse/config/imageRegistry.ts b/open-sse/config/imageRegistry.ts index 6f722da03f..bc83b37c84 100644 --- a/open-sse/config/imageRegistry.ts +++ b/open-sse/config/imageRegistry.ts @@ -548,6 +548,19 @@ export const IMAGE_PROVIDERS: Record = { ], supportedSizes: ["1024x1024", "1024x1280", "1280x1024"], }, + + // SenseNova (商汤日日新) Text-to-Image on the free Token Plan. OpenAI-compatible + // `/v1/images/generations`, so the generic OpenAI image handler routes it — same + // SenseNova api-key/connection as the chat provider. (9router#2233) + sensenova: { + id: "sensenova", + baseUrl: "https://api.sensenova.cn/v1/images/generations", + authType: "apikey", + authHeader: "bearer", + format: "openai", + models: [{ id: "sensenova-u1-fast", name: "SenseNova U1 Fast" }], + supportedSizes: ["1024x1024"], + }, }; /** diff --git a/open-sse/config/oci.ts b/open-sse/config/oci.ts index 1001ea6f2c..68d7cb5a77 100644 --- a/open-sse/config/oci.ts +++ b/open-sse/config/oci.ts @@ -1,12 +1,8 @@ -import { stripTrailingSlashes } from "../utils/urlSanitize.ts"; +import { stripTrailingSlashes, normalizeBaseUrl } from "../utils/urlSanitize.ts"; export const OCI_DEFAULT_BASE_URL = "https://inference.generativeai.us-chicago-1.oci.oraclecloud.com/openai/v1"; -function normalizeBaseUrl(value: string | null | undefined): string { - return stripTrailingSlashes((value || "").trim()); -} - export function normalizeOciBaseUrl(value: string | null | undefined): string { const normalized = normalizeBaseUrl(value || OCI_DEFAULT_BASE_URL); if (!normalized) return OCI_DEFAULT_BASE_URL; diff --git a/open-sse/config/providerModels.ts b/open-sse/config/providerModels.ts index 47fe3671e9..ac8c0aa48d 100644 --- a/open-sse/config/providerModels.ts +++ b/open-sse/config/providerModels.ts @@ -49,9 +49,16 @@ export function findModelName(aliasOrId: string, modelId: string): string { export function getModelTargetFormat(aliasOrId: string, modelId: string): string | null { const models = PROVIDER_MODELS[aliasOrId]; - if (!models) return null; - const found = models.find((m) => m.id === modelId); - return found?.targetFormat || null; + const found = models?.find((m) => m.id === modelId); + if (found?.targetFormat) return found.targetFormat; + // #5842: OpenAI "*-pro" reasoning models (o1-pro, gpt-5.x-pro) are only served by + // the native /v1/responses endpoint — /v1/chat/completions 404s ("only supported + // in v1/responses"). Curated catalog entries are tagged explicitly; this heuristic + // covers dynamically-synced ids that post-date the catalog (same spirit as the gh + // executor's /codex/i routing, 9router#102). Scoped to the openai alias so other + // providers shipping *-pro ids keep their own endpoint semantics. + if (aliasOrId === "openai" && /-pro$/i.test(modelId)) return "openai-responses"; + return null; } export function getModelStripTypes(aliasOrId: string, modelId: string): string[] { diff --git a/open-sse/config/providers/registry/chatgpt-web/index.ts b/open-sse/config/providers/registry/chatgpt-web/index.ts index 1fcb470a9d..2106b94a0b 100644 --- a/open-sse/config/providers/registry/chatgpt-web/index.ts +++ b/open-sse/config/providers/registry/chatgpt-web/index.ts @@ -9,15 +9,15 @@ export const chatgpt_webProvider: RegistryEntry = { authType: "apikey", authHeader: "cookie", models: [ - { id: "gpt-5-5-pro", name: "GPT-5.5 Pro" }, // pro tier only, standard effort - { id: "gpt-5-5-pro-extended", name: "GPT-5.5 Pro Extended" }, // pro tier only, extended effort - { id: "gpt-5-5-thinking", name: "GPT-5.5 Thinking" }, // plus, pro tier - { id: "gpt-5-5", name: "GPT-5.5 Instant" }, // free, plus, pro tier - { id: "gpt-5-4-pro", name: "GPT-5.4 Pro" }, // pro tier only - { id: "gpt-5-4-thinking", name: "GPT-5.4 Thinking" }, // plus, pro tier - { id: "gpt-5-4-t-mini", name: "GPT-5.4 Thinking Mini" }, // free-login only - { id: "gpt-5-3", name: "GPT-5.3 Instant" }, // free, free-login, plus, pro tier - { id: "gpt-5-3-mini", name: "GPT-5.3 Mini" }, // limit fallback + { id: "gpt-5.5-pro", name: "GPT-5.5 Pro" }, // pro tier only, standard effort + { id: "gpt-5.5-pro-extended", name: "GPT-5.5 Pro Extended" }, // pro tier only, extended effort + { id: "gpt-5.5-thinking", name: "GPT-5.5 Thinking" }, // plus, pro tier + { id: "gpt-5.5", name: "GPT-5.5 Instant" }, // free, plus, pro tier + { id: "gpt-5.4-pro", name: "GPT-5.4 Pro" }, // pro tier only + { id: "gpt-5.4-thinking", name: "GPT-5.4 Thinking" }, // plus, pro tier + { id: "gpt-5.4-thinking-mini", name: "GPT-5.4 Thinking Mini" }, // free-login only + { id: "gpt-5.3", name: "GPT-5.3 Instant" }, // free, free-login, plus, pro tier + { id: "gpt-5.3-mini", name: "GPT-5.3 Mini" }, // limit fallback { id: "o3", name: "o3" }, // plus ~ tier ], }; diff --git a/open-sse/config/providers/registry/huggingchat/index.ts b/open-sse/config/providers/registry/huggingchat/index.ts index 883d8b5d16..ccc95fb4a0 100644 --- a/open-sse/config/providers/registry/huggingchat/index.ts +++ b/open-sse/config/providers/registry/huggingchat/index.ts @@ -11,16 +11,133 @@ export const huggingchatProvider: RegistryEntry = { authType: "apikey", authHeader: "cookie", models: [ - { id: "meta-llama/Llama-3.3-70B-Instruct", name: "Llama 3.3 70B" }, - // Sweep 2026-06-19: + newer models surfaced on huggingface.co/chat. - { id: "Qwen/Qwen3-235B-A22B", name: "Qwen3 235B A22B", contextLength: 32768 }, + // Sweep 2026-06-30: final HuggingChat production catalog shortlist. + // Only concrete provider/model entries are registered here; router entries are excluded. { - id: "mistralai/Mistral-Small-3.1-24B-Instruct-2503", - name: "Mistral Small 3.1 24B", - contextLength: 128000, + id: "baidu/ERNIE-4.5-VL-424B-A47B-Base-PT", + name: "ERNIE 4.5 VL 424B A47B Base PT", + supportsVision: true, + }, + { + id: "CohereLabs/c4ai-command-r7b-12-2024", + name: "Command R7B 12-2024", + toolCalling: true, + }, + { + id: "CohereLabs/command-a-reasoning-08-2025", + name: "Command A Reasoning 08-2025", + toolCalling: true, + }, + { + id: "CohereLabs/command-a-vision-07-2025", + name: "Command A Vision 07-2025", + supportsVision: true, + }, + { + id: "deepseek-ai/DeepSeek-V4-Pro", + name: "DeepSeek V4 Pro", + toolCalling: true, + supportsReasoning: true, + }, + { id: "deepseek-ai/DeepSeek-V4-Flash", name: "DeepSeek V4 Flash", toolCalling: true }, + { + id: "google/gemma-4-31B-it", + name: "Gemma 4 31B", + supportsVision: true, + toolCalling: true, + }, + { + id: "google/gemma-4-26B-A4B-it", + name: "Gemma 4 26B A4B", + supportsVision: true, + toolCalling: true, + }, + { id: "inclusionAI/Ling-2.6-1T", name: "Ling 2.6 1T", toolCalling: true }, + { + id: "meta-llama/Llama-4-Scout-17B-16E-Instruct", + name: "Llama 4 Scout 17B 16E Instruct", + supportsVision: true, + toolCalling: true, + }, + { + id: "meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8", + name: "Llama 4 Maverick 17B 128E Instruct FP8", + supportsVision: true, + }, + { + id: "MiniMaxAI/MiniMax-M3", + name: "MiniMax M3", + supportsVision: true, + toolCalling: true, + supportsReasoning: true, + }, + { + id: "moonshotai/Kimi-K2.7-Code", + name: "Kimi K2.7 Code", + supportsVision: true, + toolCalling: true, + supportsReasoning: true, + }, + { + id: "moonshotai/Kimi-K2.6", + name: "Kimi K2.6", + supportsVision: true, + toolCalling: true, + }, + { + id: "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4", + name: "NVIDIA Nemotron 3 Ultra 550B A55B NVFP4", + toolCalling: true, + supportsReasoning: true, + }, + { + id: "openai/gpt-oss-120b", + name: "GPT-OSS 120B", + toolCalling: true, + supportsReasoning: true, + }, + { + id: "openai/gpt-oss-20b", + name: "GPT-OSS 20B", + toolCalling: true, + supportsReasoning: true, + }, + { + id: "Qwen/Qwen3.5-122B-A10B", + name: "Qwen3.5 122B A10B", + supportsVision: true, + toolCalling: true, + }, + { + id: "Qwen/Qwen3.5-397B-A17B", + name: "Qwen3.5 397B A17B", + supportsVision: true, + toolCalling: true, + }, + { + id: "Qwen/Qwen3.6-27B", + name: "Qwen3.6 27B", + supportsVision: true, + toolCalling: true, + }, + { + id: "Qwen/Qwen3.6-35B-A3B", + name: "Qwen3.6 35B A3B", + supportsVision: true, + toolCalling: true, + }, + { + id: "stepfun-ai/Step-3.7-Flash", + name: "Step 3.7 Flash", + supportsVision: true, + toolCalling: true, + }, + { id: "XiaomiMiMo/MiMo-V2.5-Pro", name: "MiMo V2.5 Pro", toolCalling: true }, + { + id: "zai-org/GLM-5.2", + name: "GLM 5.2", + toolCalling: true, + supportsReasoning: true, }, - { id: "Qwen/Qwen2.5-72B-Instruct", name: "Qwen 2.5 72B" }, - { id: "mistralai/Mistral-Small-24B-Instruct-2501", name: "Mistral Small 24B" }, - { id: "deepseek-ai/DeepSeek-R1", name: "DeepSeek R1" }, ], }; diff --git a/open-sse/config/providers/registry/kimi/web/index.ts b/open-sse/config/providers/registry/kimi/web/index.ts index e425c62474..c86d3ce426 100644 --- a/open-sse/config/providers/registry/kimi/web/index.ts +++ b/open-sse/config/providers/registry/kimi/web/index.ts @@ -7,11 +7,15 @@ export const kimi_webProvider: RegistryEntry = { alias: "kimi-web", format: "openai", executor: "kimi-web", - baseUrl: "https://kimi.moonshot.cn/api/chat", + // International consumer chat — the legacy `kimi.moonshot.cn` domain now + // redirects every non-CN visitor to www.kimi.com, which speaks a different + // Connect-RPC API. See `open-sse/executors/kimi-web.ts` for the wire format. + baseUrl: "https://www.kimi.com", authType: "apikey", authHeader: "cookie", models: [ { id: "kimi-default", name: "Kimi Default" }, + { id: "kimi-k2.6", name: "Kimi K2.6 (Thinking)" }, { id: "kimi-128k", name: "Kimi 128K (Long Context)" }, ], }; diff --git a/open-sse/config/providers/registry/kiro/index.ts b/open-sse/config/providers/registry/kiro/index.ts index b3a9792264..6c4d41cf6e 100644 --- a/open-sse/config/providers/registry/kiro/index.ts +++ b/open-sse/config/providers/registry/kiro/index.ts @@ -41,6 +41,12 @@ export const kiroProvider: RegistryEntry = { contextLength: 1000000, maxOutputTokens: 128000, }, + { + id: "claude-sonnet-5", + name: "Claude Sonnet 5", + contextLength: 1000000, + maxOutputTokens: 128000, + }, { id: "claude-sonnet-4.6", name: "Claude Sonnet 4.6", diff --git a/open-sse/config/providers/registry/nebius/index.ts b/open-sse/config/providers/registry/nebius/index.ts index 6e9caf7eaf..b49bd34e81 100644 --- a/open-sse/config/providers/registry/nebius/index.ts +++ b/open-sse/config/providers/registry/nebius/index.ts @@ -1,12 +1,9 @@ import type { RegistryEntry } from "../../shared.ts"; +import { buildOpenAiCompatibleRegistryEntry } from "../../shared.ts"; -export const nebiusProvider: RegistryEntry = { +export const nebiusProvider: RegistryEntry = buildOpenAiCompatibleRegistryEntry({ id: "nebius", alias: "nebius", - format: "openai", - executor: "default", baseUrl: "https://api.tokenfactory.nebius.com/v1/chat/completions", - authType: "apikey", - authHeader: "bearer", models: [{ id: "meta-llama/Llama-3.3-70B-Instruct", name: "Llama 3.3 70B Instruct" }], -}; +}); diff --git a/open-sse/config/providers/registry/ollama-cloud/index.ts b/open-sse/config/providers/registry/ollama-cloud/index.ts index b0e89994c4..bd74e0d218 100644 --- a/open-sse/config/providers/registry/ollama-cloud/index.ts +++ b/open-sse/config/providers/registry/ollama-cloud/index.ts @@ -10,7 +10,7 @@ export const ollama_cloudProvider: RegistryEntry = { authType: "apikey", authHeader: "bearer", // Note: rate limits vary by plan (free = "Light usage", Pro = more, Max = 5x Pro). - // Users can generate API keys at https://ollama.com/settings/api-keys + // Users can generate API keys at https://ollama.com/settings/keys models: [ { id: "deepseek-v4-pro", name: "DeepSeek V4 Pro", supportsReasoning: true }, { id: "deepseek-v4-flash", name: "DeepSeek V4 Flash", supportsReasoning: true }, diff --git a/open-sse/config/providers/registry/openai/index.ts b/open-sse/config/providers/registry/openai/index.ts index 46dce6e816..fcd291d254 100644 --- a/open-sse/config/providers/registry/openai/index.ts +++ b/open-sse/config/providers/registry/openai/index.ts @@ -12,9 +12,21 @@ export const openaiProvider: RegistryEntry = { defaultContextLength: 128000, models: [ { id: "gpt-5.5", name: "GPT-5.5", contextLength: 1050000 }, - { id: "gpt-5.5-pro", name: "GPT-5.5 Pro", contextLength: 1050000 }, + // #5842: *-pro reasoning models are responses-only upstream — /v1/chat/completions + // 404s ("only supported in v1/responses"). targetFormat routes them natively. + { + id: "gpt-5.5-pro", + name: "GPT-5.5 Pro", + contextLength: 1050000, + targetFormat: "openai-responses", + }, { id: "gpt-5.4", name: "GPT-5.4", contextLength: 1050000 }, - { id: "gpt-5.4-pro", name: "GPT-5.4 Pro", contextLength: 1050000 }, + { + id: "gpt-5.4-pro", + name: "GPT-5.4 Pro", + contextLength: 1050000, + targetFormat: "openai-responses", + }, { id: "gpt-5.4-mini", name: "GPT-5.4 Mini", contextLength: 400000 }, { id: "gpt-5.4-nano", name: "GPT-5.4 Nano", contextLength: 400000 }, { id: "gpt-4.1", name: "GPT-4.1", contextLength: 1047576 }, diff --git a/open-sse/config/providers/registry/sensenova/index.ts b/open-sse/config/providers/registry/sensenova/index.ts index c3f98d3e79..9407e4ef29 100644 --- a/open-sse/config/providers/registry/sensenova/index.ts +++ b/open-sse/config/providers/registry/sensenova/index.ts @@ -15,6 +15,8 @@ export const sensenovaProvider: RegistryEntry = { { id: "SenseNova-V6.5-Pro", name: "SenseNova V6.5 Pro", contextLength: 131072 }, { id: "SenseNova-V6.5-Turbo", name: "SenseNova V6.5 Turbo", contextLength: 131072 }, { id: "sensenova-6.7-flash-lite", name: "SenseNova 6.7 Flash-Lite" }, + // DeepSeek V4 Flash is served on SenseNova's free Token Plan (9router#2233). + { id: "deepseek-v4-flash", name: "DeepSeek V4 Flash" }, { id: "SenseChat-5", name: "SenseChat 5", contextLength: 131072 }, { id: "SenseChat-5-Cantonese", name: "SenseChat 5 Cantonese", contextLength: 32768 }, { id: "SenseChat-Turbo", name: "SenseChat Turbo", contextLength: 4096 }, diff --git a/open-sse/config/providers/registry/siliconflow/index.ts b/open-sse/config/providers/registry/siliconflow/index.ts index 46dd572b2d..238cf60e97 100644 --- a/open-sse/config/providers/registry/siliconflow/index.ts +++ b/open-sse/config/providers/registry/siliconflow/index.ts @@ -1,13 +1,10 @@ import type { RegistryEntry } from "../../shared.ts"; +import { buildOpenAiCompatibleRegistryEntry } from "../../shared.ts"; -export const siliconflowProvider: RegistryEntry = { +export const siliconflowProvider: RegistryEntry = buildOpenAiCompatibleRegistryEntry({ id: "siliconflow", alias: "siliconflow", - format: "openai", - executor: "default", baseUrl: "https://api.siliconflow.com/v1/chat/completions", - authType: "apikey", - authHeader: "bearer", models: [ // DeepSeek { id: "deepseek-ai/DeepSeek-V3.2", name: "DeepSeek V3.2" }, @@ -84,4 +81,4 @@ export const siliconflowProvider: RegistryEntry = { { id: "google/gemma-4-26B-A4B-it", name: "Gemma 4 26B" }, { id: "ByteDance-Seed/Seed-OSS-36B-Instruct", name: "Seed OSS 36B" }, ], -}; +}); diff --git a/open-sse/config/providers/registry/together/index.ts b/open-sse/config/providers/registry/together/index.ts index afc67211dc..c185804b08 100644 --- a/open-sse/config/providers/registry/together/index.ts +++ b/open-sse/config/providers/registry/together/index.ts @@ -1,13 +1,10 @@ import type { RegistryEntry } from "../../shared.ts"; +import { buildOpenAiCompatibleRegistryEntry } from "../../shared.ts"; -export const togetherProvider: RegistryEntry = { +export const togetherProvider: RegistryEntry = buildOpenAiCompatibleRegistryEntry({ id: "together", alias: "together", - format: "openai", - executor: "default", baseUrl: "https://api.together.xyz/v1/chat/completions", - authType: "apikey", - authHeader: "bearer", models: [ { id: "meta-llama/Llama-3.3-70B-Instruct-Turbo-Free", name: "Llama 3.3 70B Turbo (🆓 Free)" }, { id: "meta-llama/Llama-Vision-Free", name: "Llama Vision (🆓 Free)" }, @@ -20,4 +17,4 @@ export const togetherProvider: RegistryEntry = { { id: "Qwen/Qwen3-235B-A22B", name: "Qwen3 235B" }, { id: "meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8", name: "Llama 4 Maverick" }, ], -}; +}); diff --git a/open-sse/config/providers/shared.ts b/open-sse/config/providers/shared.ts index 670e18ed93..ecc4fd088d 100644 --- a/open-sse/config/providers/shared.ts +++ b/open-sse/config/providers/shared.ts @@ -143,6 +143,24 @@ export interface RegistryEntry { anonymousApiKey?: string; } +/** + * Build a standard OpenAI-compatible provider registry entry. + * Eliminates the 4-field boilerplate (format, executor, authType, authHeader) + * repeated across 40+ provider files. + */ +export function buildOpenAiCompatibleRegistryEntry( + overrides: Pick & + Partial> +): RegistryEntry { + return { + format: "openai", + executor: "default", + authType: "apikey", + authHeader: "bearer", + ...overrides, + } as RegistryEntry; +} + export interface LegacyProvider { format: string; baseUrl?: string; diff --git a/open-sse/config/sap.ts b/open-sse/config/sap.ts index 3c51fa1e3c..9430a69790 100644 --- a/open-sse/config/sap.ts +++ b/open-sse/config/sap.ts @@ -1,12 +1,8 @@ -import { stripTrailingSlashes } from "../utils/urlSanitize.ts"; +import { stripTrailingSlashes, normalizeBaseUrl } from "../utils/urlSanitize.ts"; export const SAP_DEFAULT_BASE_URL = "https://example-aicore.cfapps.eu10.hana.ondemand.com/v2/lm/deployments/example-deployment"; -function normalizeBaseUrl(value: string | null | undefined): string { - return stripTrailingSlashes((value || "").trim()); -} - function sanitizeUrl(value: string): string { try { const parsed = new URL(value); diff --git a/open-sse/config/watsonx.ts b/open-sse/config/watsonx.ts index fb9c58dfbf..0afa77de5a 100644 --- a/open-sse/config/watsonx.ts +++ b/open-sse/config/watsonx.ts @@ -1,11 +1,7 @@ -import { stripTrailingSlashes } from "../utils/urlSanitize.ts"; +import { stripTrailingSlashes, normalizeBaseUrl } from "../utils/urlSanitize.ts"; export const WATSONX_DEFAULT_BASE_URL = "https://ca-tor.ml.cloud.ibm.com/ml/gateway/v1"; -function normalizeBaseUrl(value: string | null | undefined): string { - return stripTrailingSlashes((value || "").trim()); -} - export function normalizeWatsonxBaseUrl(value: string | null | undefined): string { const normalized = normalizeBaseUrl(value || WATSONX_DEFAULT_BASE_URL); if (!normalized) return WATSONX_DEFAULT_BASE_URL; diff --git a/open-sse/executors/antigravity.ts b/open-sse/executors/antigravity.ts index a238f0feb5..74393938ed 100644 --- a/open-sse/executors/antigravity.ts +++ b/open-sse/executors/antigravity.ts @@ -1388,7 +1388,14 @@ export class AntigravityExecutor extends BaseExecutor { try { const errorBody = await response.clone().text(); const errorJson = JSON.parse(errorBody); - const errorMessage = errorJson?.error?.message || errorJson?.message || ""; + let errorMessage = errorJson?.error?.message || errorJson?.message || ""; + if (errorJson?.error?.details && Array.isArray(errorJson.error.details)) { + for (const detail of errorJson.error.details) { + if (detail?.reason) { + errorMessage += ` ${detail.reason}`; + } + } + } // 1. Try to parse explicit retry time from message const parsedRetryMs = this.parseRetryFromErrorMessage(errorMessage); diff --git a/open-sse/executors/base.ts b/open-sse/executors/base.ts index 45e2853179..96685152c2 100644 --- a/open-sse/executors/base.ts +++ b/open-sse/executors/base.ts @@ -65,6 +65,7 @@ import { stainlessRuntimeVersion, stripProxyToolPrefix, } from "./claudeIdentity.ts"; +import { withForcedResponsesUpstream } from "./forceResponsesUpstream.ts"; /** * Sanitizes a custom API path to prevent path traversal attacks. @@ -233,7 +234,10 @@ export function stripStainlessHeadersForOpenAICompat( // Normalize User-Agent: SDK-based clients send verbose product strings that some // upstreams block. Replace with a clean browser-like UA only when it looks SDK-derived. const ua = (headers["User-Agent"] || headers["user-agent"] || "").toLowerCase(); - if (ua.includes("openai") && (ua.includes("node") || ua.includes("axios") || ua.includes("undici"))) { + if ( + ua.includes("openai") && + (ua.includes("node") || ua.includes("axios") || ua.includes("undici")) + ) { setUserAgentHeader(headers, "Mozilla/5.0 (compatible; OpenAI Compatible)"); } @@ -528,15 +532,52 @@ export class BaseExecutor { return baseUrls[urlIndex] || baseUrls[0] || this.config.baseUrl || ""; } - buildHeaders( + /** + * Resolve the effective base URL for a request, preferring per-connection + * providerSpecificData.baseUrl over the static provider config baseUrl. + */ + protected resolveBaseUrl(credentials: ProviderCredentials | null, fallback?: string): string { + const psdBaseUrl = credentials?.providerSpecificData?.baseUrl; + return ( + (typeof psdBaseUrl === "string" ? psdBaseUrl : "") || fallback || this.config.baseUrl || "" + ); + } + + /** + * Resolve the effective API key via extra-keys round-robin rotation. + * Mutates `credentials.providerSpecificData.selectedKeyId` on rotation. + */ + protected resolveEffectiveKey(credentials: ProviderCredentials): string | undefined { + const extraKeys = + (credentials.providerSpecificData?.extraApiKeys as string[] | undefined) ?? []; + const selectedKeyId = (credentials.providerSpecificData as Record | undefined) + ?.selectedKeyId as string | undefined; + let effectiveKey = credentials.apiKey; + if (extraKeys.length > 0 && credentials.connectionId && credentials.apiKey) { + const resolved = resolveKeyForRequest( + credentials.connectionId, + credentials.apiKey, + extraKeys, + selectedKeyId ?? null + ); + effectiveKey = resolved?.key ?? credentials.apiKey; + if (resolved && credentials.providerSpecificData) { + (credentials.providerSpecificData as Record).selectedKeyId = + resolved.keyId; + } + } + return effectiveKey; + } + + /** + * Build the common header preamble shared by BaseExecutor and DefaultExecutor: + * Content-Type, config.headers, per-provider User-Agent env override, and + * resolved effective key (via extra-keys round-robin). + */ + protected buildHeadersPreamble( credentials: ProviderCredentials, - stream = true, - clientHeaders?: Record | null, - model?: string, - health?: Record - ): Record { - void clientHeaders; - void model; + stream: boolean + ): { headers: Record; effectiveKey: string | undefined } { const headers: Record = { "Content-Type": "application/json", ...this.config.headers, @@ -553,28 +594,25 @@ export class BaseExecutor { } } + const effectiveKey = this.resolveEffectiveKey(credentials); + void stream; + return { headers, effectiveKey }; + } + + buildHeaders( + credentials: ProviderCredentials, + stream = true, + clientHeaders?: Record | null, + model?: string, + health?: Record + ): Record { + void clientHeaders; + void model; + const { headers, effectiveKey } = this.buildHeadersPreamble(credentials, stream); + if (credentials.accessToken) { headers["Authorization"] = `Bearer ${credentials.accessToken}`; } else if (credentials.apiKey) { - const extraKeys = - (credentials.providerSpecificData?.extraApiKeys as string[] | undefined) ?? []; - const selectedKeyId = ( - credentials.providerSpecificData as Record | undefined - )?.selectedKeyId as string | undefined; - let effectiveKey = credentials.apiKey; - if (extraKeys.length > 0 && credentials.connectionId) { - const resolved = resolveKeyForRequest( - credentials.connectionId, - credentials.apiKey, - extraKeys, - selectedKeyId ?? null - ); - effectiveKey = resolved?.key ?? credentials.apiKey; - if (resolved && credentials.providerSpecificData) { - (credentials.providerSpecificData as Record).selectedKeyId = - resolved.keyId; - } - } headers["Authorization"] = `Bearer ${effectiveKey}`; } @@ -863,9 +901,14 @@ export class BaseExecutor { const strippedFields = new Set(); for (let urlIndex = 0; urlIndex < fallbackCount; urlIndex++) { - const url = this.buildUrl(model, stream, urlIndex, activeCredentials); - const headers = this.buildHeaders(activeCredentials, stream, clientHeaders, model); - applyConfiguredUserAgent(headers, activeCredentials?.providerSpecificData); + const requestCredentials = withForcedResponsesUpstream( + this.provider, + body, + activeCredentials + ); + const url = this.buildUrl(model, stream, urlIndex, requestCredentials); + const headers = this.buildHeaders(requestCredentials, stream, clientHeaders, model); + applyConfiguredUserAgent(headers, requestCredentials?.providerSpecificData); // Strip OpenAI SDK (X-Stainless-*) metadata + normalize SDK-derived User-Agent // on OpenAI-compatible passthrough requests — some upstream gateways 403 on them. @@ -878,7 +921,7 @@ export class BaseExecutor { } const ccRequestDefaults = isClaudeCodeCompatible(this.provider) - ? getClaudeCodeCompatibleRequestDefaults(activeCredentials?.providerSpecificData) + ? getClaudeCodeCompatibleRequestDefaults(requestCredentials?.providerSpecificData) : {}; const shouldForwardExtendedContext = extendedContext && @@ -894,7 +937,7 @@ export class BaseExecutor { model, body, stream, - activeCredentials + requestCredentials ); let transformedBody = sanitizeReasoningEffortForProvider( rawTransformedBody, @@ -1110,14 +1153,11 @@ export class BaseExecutor { const seed = activeCredentials?.accessToken || activeCredentials?.apiKey || "anon"; const psd = activeCredentials?.providerSpecificData as - | Record - | undefined; + Record | undefined; let identitySource: - | "upstream-metadata" - | "upstream-header" - | "synthesized" - | "synthesized-cloaked" = "synthesized"; + "upstream-metadata" | "upstream-header" | "synthesized" | "synthesized-cloaked" = + "synthesized"; let sessionId: string; let deviceId: string; let accountUUID: string; diff --git a/open-sse/executors/chatgpt-web.ts b/open-sse/executors/chatgpt-web.ts index 135c9d450c..87a4bd3f29 100644 --- a/open-sse/executors/chatgpt-web.ts +++ b/open-sse/executors/chatgpt-web.ts @@ -79,13 +79,13 @@ function deviceIdFor(cookie: string): string { return id; } -// OmniRoute model ID → ChatGPT internal slug. OmniRoute uses dot-form IDs -// (e.g. "gpt-5.3-instant"), ChatGPT's web routes use dash-form -// (e.g. "gpt-5-3-instant"). The slug catalog comes from -// /backend-api/models on a logged-in account; "gpt-5-4-t-mini" is ChatGPT's -// abbreviated slug for "GPT-5.4 Thinking Mini". +// OmniRoute model ID → ChatGPT internal slug. The public ChatGPT Web catalog +// keeps OmniRoute's historical dot-form IDs (e.g. "gpt-5.5-pro"), while +// ChatGPT's backend routes use dash-form slugs (e.g. "gpt-5-5-pro"). The slug +// catalog comes from /backend-api/models on a logged-in account; +// "gpt-5-4-t-mini" is ChatGPT's abbreviated slug for "GPT-5.4 Thinking Mini". const MODEL_MAP: Record = { - // Current ChatGPT Web slugs (these are what the provider catalog exposes). + // ChatGPT backend slugs are also accepted directly for power users / tests. "gpt-5-5-pro": "gpt-5-5-pro", "gpt-5-5-pro-extended": "gpt-5-5-pro", "gpt-5-5-thinking": "gpt-5-5-thinking", @@ -96,7 +96,7 @@ const MODEL_MAP: Record = { "gpt-5-3": "gpt-5-3", "gpt-5-3-mini": "gpt-5-3-mini", - // Legacy OmniRoute dot-form aliases kept for direct provider/model callers. + // Public OmniRoute dot-form ids exposed by the provider catalog. "gpt-5.5-pro": "gpt-5-5-pro", "gpt-5.5-pro-extended": "gpt-5-5-pro", "gpt-5.5-thinking": "gpt-5-5-thinking", diff --git a/open-sse/executors/commandCode.ts b/open-sse/executors/commandCode.ts index 70bebeedf1..afd9811dd4 100644 --- a/open-sse/executors/commandCode.ts +++ b/open-sse/executors/commandCode.ts @@ -1,5 +1,6 @@ import { randomUUID } from "node:crypto"; +import { isVisionModelId } from "@/shared/constants/visionModels"; import { REGISTRY } from "../config/providerRegistry.ts"; import { BaseExecutor, mergeUpstreamExtraHeaders, type ExecuteInput } from "./base.ts"; @@ -55,6 +56,99 @@ function normalizeContentText(content: unknown): string { .join("\n"); } +/** + * Model id patterns for Command Code models that have `text, vision` + * capability per the official CC model registry, but are NOT caught + * by the shared {@link isVisionModelId} heuristic. Kept as a local + * set because these are CC-specific model IDs (vendor-prefix shapes + * like "moonshotai/Kimi-K2.6" or CC aliases like "gpt-5.4-mini"). + * + * Source: Command Code /alpha/generate model registry (docs). + */ +const CC_VISION_MODEL_PATTERNS: readonly RegExp[] = [ + // Open Source + /kimi-k2/i, // moonshotai/Kimi-K2.6, Kimi-K2.7-Code, Kimi-K2.5 + /qwen3\.\d/i, // Qwen/Qwen3.6-Plus, Qwen/Qwen3.7-Plus + /step-?3/i, // stepfun/Step-3.7-Flash + // Anthropic + /claude-fable/i, // claude-fable-5 (not covered by claude-opus/sonnet/haiku-4) + // OpenAI + /gpt-5/i, // gpt-5.5, gpt-5.4, gpt-5.3-codex, gpt-5.4-mini + // Sakana + /fugu/i, // sakana/fugu-ultra +]; + +/** + * Whether a model id routed through the Command Code executor is + * vision-capable. Checks Mimo-specific rules first, then CC-specific + * patterns, then falls through to the shared {@link isVisionModelId} + * heuristic (which covers minimax-m3, claude-3/4 families, gemini, + * gpt-4o/4.1, mistral-medium-3, and general "-vision" / "multimodal"). + */ +function isCommandCodeVisionModel(model?: string | null): boolean { + if (!model) return false; + // mimo-v2.5-pro is text-only — exclude before any positive check + if (/(?:^|\/)mimo-v2\.5-pro$/i.test(model)) return false; + // Only mimo-v2.5 and mimo-v2-omni accept images per Xiaomi vendor docs + if (/(?:^|\/)mimo-v2\.5$/i.test(model)) return true; + if (/(?:^|\/)mimo-v2-omni$/i.test(model)) return true; + // CC-specific patterns: Kimi K2, Qwen 3.x, Stepfun, Claude Fable, + // GPT-5, Sakana Fugu — not covered by the shared heuristic + if (CC_VISION_MODEL_PATTERNS.some((pattern) => pattern.test(model))) return true; + // Fall through: minimax-m3, claude-3/4, gemini-2/3, gpt-4o, -vision, multimodal + return isVisionModelId(model); +} + +/** + * Extract the image URL from an OpenAI-compatible or Command Code + * content part, returning undefined for non-image parts. + * + * OpenAI-compatible: { type: "image_url", image_url: { url: "..." } } + * Command Code CLI: { type: "image", image: "..." } + */ +function extractImageUrl(part: JsonRecord): string | undefined { + if (part.type === "image") return stringValue(part.image); + if (part.type === "image_url") { + if (isRecord(part.image_url)) return stringValue(part.image_url.url); + return stringValue(part.image_url); + } + return undefined; +} + +/** + * Convert an OpenAI-format content array to Command Code's internal + * CLI format. For vision-capable models (MiniMax M3, MiMo v2.5, etc.) + * this also preserves image parts alongside text. + */ +function convertUserContentParts(content: unknown, isVisionModel: boolean): string | unknown[] { + // For non-vision models or string content, extract text only. + if (!isVisionModel || typeof content === "string") { + return normalizeContentText(content); + } + + const parts: unknown[] = []; + for (const part of asRecordArray(content)) { + if (part.type === "text") { + const text = stringValue(part.text); + if (text) parts.push({ type: "text", text }); + continue; + } + const imgUrl = extractImageUrl(part); + if (imgUrl) { + parts.push({ type: "image", image: imgUrl }); + continue; + } + // Always drop tool_use / tool_result / thinking parts from user + // messages (Command Code doesn't accept them for role:"user"). + } + + // When every part was stripped, fall back to empty text so the + // message is still valid JSON (Command Code rejects empty content). + if (parts.length === 0) parts.push({ type: "text", text: "" }); + + return parts; +} + function convertTools(tools: unknown): unknown[] { return asRecordArray(tools).map((tool) => { const fn = isRecord(tool.function) ? tool.function : tool; @@ -86,11 +180,15 @@ function completeToolCallIds(messages: JsonRecord[]): Set { return new Set([...callIds].filter((id) => resultIds.has(id))); } -function convertMessages(messages: unknown): { system: string; messages: unknown[] } { +function convertMessages( + messages: unknown, + model?: string | null +): { system: string; messages: unknown[] } { const source = asRecordArray(messages); const pairedToolCallIds = completeToolCallIds(source); const out: unknown[] = []; const system: string[] = []; + const isVision = isCommandCodeVisionModel(model); for (const message of source) { const role = stringValue(message.role); @@ -101,7 +199,7 @@ function convertMessages(messages: unknown): { system: string; messages: unknown } if (role === "user") { - out.push({ role: "user", content: normalizeContentText(message.content) }); + out.push({ role: "user", content: convertUserContentParts(message.content, isVision) }); continue; } @@ -175,9 +273,6 @@ const COMMAND_CODE_PASSTHROUGH_FIELDS = [ function buildCommandCodeBody(model: string, body: unknown, stream = false): JsonRecord { const input = isRecord(body) ? body : {}; - const converted = convertMessages(input.messages); - const explicitSystem = typeof input.system === "string" ? input.system : ""; - const system = [converted.system, explicitSystem].filter(Boolean).join("\n\n"); // Payload rules may rewrite `body.model` (e.g. deepseek-v4-pro-max → // deepseek/deepseek-v4-pro for the command-code provider). Prefer the @@ -185,6 +280,10 @@ function buildCommandCodeBody(model: string, body: unknown, stream = false): Jso const resolvedModel = typeof input.model === "string" && input.model.trim().length > 0 ? input.model : model; + const converted = convertMessages(input.messages, resolvedModel); + const explicitSystem = typeof input.system === "string" ? input.system : ""; + const system = [converted.system, explicitSystem].filter(Boolean).join("\n\n"); + const params: JsonRecord = { model: resolvedModel, messages: converted.messages, diff --git a/open-sse/executors/copilot-m365-connection.ts b/open-sse/executors/copilot-m365-connection.ts index 5c648f4b99..e5545de3f9 100644 --- a/open-sse/executors/copilot-m365-connection.ts +++ b/open-sse/executors/copilot-m365-connection.ts @@ -86,6 +86,38 @@ export function newChatSessionId(): string { return randomBytes(16).toString("hex"); } +function parsePastedCredential( + raw: string +): Partial> { + const value = raw.trim(); + const parts: Record = {}; + + for (const segment of value.split(/[;\n]/)) { + const separator = segment.indexOf("="); + if (separator <= 0) continue; + const key = segment.slice(0, separator).trim(); + const partValue = segment.slice(separator + 1).trim(); + if (key && partValue) parts[key] = partValue; + } + + if (/^wss:\/\/substrate\.office\.com\/m365Copilot\/Chathub\//i.test(value)) { + try { + const url = new URL(value); + parts.access_token ||= url.searchParams.get("access_token") || ""; + parts.chathubPath ||= decodeURIComponent( + url.pathname.split("/m365Copilot/Chathub/")[1] || "" + ); + } catch { + // Keep any key/value fields already parsed from the pasted text. + } + } + + return { + accessToken: parts.access_token || parts.accessToken, + chathubPath: parts.chathubPath || parts.userTenant, + }; +} + /** * Read the pasted credential bits. The individual access_token is opaque (JWE), * so it is consumed verbatim. The Chathub path (`user@tenant`) is pasted @@ -95,8 +127,14 @@ export function resolveConnectionParams( credentials: ProviderCredentials | undefined ): M365ConnectionParams | { error: string } { const psd = (credentials?.providerSpecificData ?? {}) as JsonRecord; + const parsedApiKey = + typeof credentials?.apiKey === "string" ? parsePastedCredential(credentials.apiKey) : {}; const accessToken = - (typeof credentials?.apiKey === "string" && credentials.apiKey) || + parsedApiKey.accessToken || + (typeof credentials?.apiKey === "string" && + credentials.apiKey && + !credentials.apiKey.includes("access_token=") && + credentials.apiKey) || (typeof psd.accessToken === "string" && psd.accessToken) || (typeof psd.access_token === "string" && psd.access_token) || ""; @@ -104,6 +142,7 @@ export function resolveConnectionParams( return { error: "Missing M365 Copilot access_token. Paste it as the provider credential." }; } const chathubPath = + parsedApiKey.chathubPath || (typeof psd.chathubPath === "string" && psd.chathubPath) || (typeof psd.userTenant === "string" && psd.userTenant) || ""; diff --git a/open-sse/executors/default.ts b/open-sse/executors/default.ts index 21ebd0f07d..cbdb8d90bf 100644 --- a/open-sse/executors/default.ts +++ b/open-sse/executors/default.ts @@ -1,12 +1,7 @@ -import { BaseExecutor, setUserAgentHeader, type ExecuteInput } from "./base.ts"; +import { BaseExecutor, type ExecuteInput } from "./base.ts"; import { PROVIDERS, OAUTH_ENDPOINTS } from "../config/constants.ts"; import { getAccessToken } from "../services/tokenRefresh.ts"; -import { - getRotatingApiKey, - getValidApiKey, - resolveKeyForRequest, -} from "../services/apiKeyRotator.ts"; -import type { KeyHealth } from "../services/apiKeyRotator.ts"; + import { buildClaudeCodeCompatibleHeaders, CLAUDE_CODE_COMPATIBLE_DEFAULT_CHAT_PATH, @@ -14,6 +9,7 @@ import { } from "../services/claudeCodeCompatible.ts"; import { getGigachatAccessToken } from "../services/gigachatAuth.ts"; import { getRegistryEntry } from "../config/providerRegistry.ts"; +import { getModelTargetFormat } from "../config/providerModels.ts"; import { mergeClientAnthropicBeta, normalizeAnthropicHeaderVariants, @@ -38,6 +34,14 @@ import { LOCAL_PROVIDERS } from "@/shared/constants/providers"; import { isForbiddenCustomHeaderName } from "@/shared/constants/upstreamHeaders"; import { getClaudeCodeCompatibleRequestDefaults } from "@/lib/providers/requestDefaults"; import { buildClineHeaders } from "@/shared/utils/clineAuth"; +import { normalizeBaseUrl } from "../utils/urlSanitize.ts"; +import { + normalizeHerokuChatUrl, + normalizeDatabricksChatUrl, + normalizeSnowflakeChatUrl, + normalizeGigachatChatUrl, +} from "@/lib/providers/validation/urlHelpers"; +import { forwardOpencodeClientHeaders } from "../utils/opencodeHeaders.ts"; import type { PoolConfig } from "../services/sessionPool/types.ts"; @@ -83,28 +87,12 @@ function applyCustomHeaders(headers: Record, rawCustomHeaders: u } } -function normalizeBaseUrl(baseUrl) { - return (baseUrl || "").trim().replace(/\/$/, ""); -} - function normalizeBailianMessagesUrl(baseUrl) { const normalized = normalizeBaseUrl(baseUrl).replace(/\?beta=true$/, ""); const messagesUrl = normalized.endsWith("/messages") ? normalized : `${normalized}/messages`; return messagesUrl; } -function normalizeHerokuChatUrl(baseUrl) { - const normalized = normalizeBaseUrl(baseUrl); - if (normalized.endsWith("/v1/chat/completions")) return normalized; - return `${normalized}/v1/chat/completions`; -} - -function normalizeDatabricksChatUrl(baseUrl) { - const normalized = normalizeBaseUrl(baseUrl); - if (normalized.endsWith("/chat/completions")) return normalized; - return `${normalized}/chat/completions`; -} - function normalizeDataRobotChatUrl(baseUrl) { return buildDataRobotChatUrl(baseUrl); } @@ -130,18 +118,6 @@ function normalizeXiaomiMimoChatUrl(baseUrl) { return `${normalized}/chat/completions`; } -function normalizeSnowflakeChatUrl(baseUrl) { - const normalized = normalizeBaseUrl(baseUrl) - .replace(/\/cortex\/inference:complete$/, "") - .replace(/\/api\/v2$/, ""); - return `${normalized}/api/v2/cortex/inference:complete`; -} - -function normalizeGigachatChatUrl(baseUrl) { - const normalized = normalizeBaseUrl(baseUrl).replace(/\/chat\/completions$/, ""); - return `${normalized}/chat/completions`; -} - function normalizeOpenAIChatUrl(baseUrl) { const normalized = normalizeBaseUrl(baseUrl); if ( @@ -186,8 +162,9 @@ export class DefaultExecutor extends BaseExecutor { const normalized = baseUrl.replace(/\/$/, ""); const customPath = typeof psd?.chatPath === "string" && psd.chatPath ? psd.chatPath : null; if (customPath) return `${normalized}${customPath}`; + const forceResponses = psd?._omnirouteForceResponsesUpstream === true; const path = - getOpenAICompatibleType(this.provider, psd) === "responses" + forceResponses || getOpenAICompatibleType(this.provider, psd) === "responses" ? "/responses" : "/chat/completions"; return `${normalized}${path}`; @@ -206,20 +183,38 @@ export class DefaultExecutor extends BaseExecutor { return `${normalized}${customPath || "/messages"}`; } switch (this.provider) { + case "openai": { + // #5842: responses-only models (o1-pro / gpt-5.x-pro) 404 on + // /v1/chat/completions ("only supported in v1/responses"). Route them to + // the native /responses endpoint — the per-model targetFormat (registry tag + // + the -pro heuristic in getModelTargetFormat) is the single source of + // truth, keeping the URL in lockstep with the chatCore body translation. + // Mirrors the gh executor's targetFormat-driven routing (9router#102). + const customBaseUrl = + typeof credentials?.providerSpecificData?.baseUrl === "string" && + credentials.providerSpecificData.baseUrl.trim() + ? (credentials.providerSpecificData.baseUrl as string) + : null; + const chatUrl = customBaseUrl ? normalizeOpenAIChatUrl(customBaseUrl) : this.config.baseUrl; + if (getModelTargetFormat("openai", model) === "openai-responses") { + return chatUrl.replace(/\/chat\/completions\/?$/, "/responses"); + } + return chatUrl; + } case "bailian-coding-plan": { - const baseUrl = credentials?.providerSpecificData?.baseUrl || this.config.baseUrl; + const baseUrl = this.resolveBaseUrl(credentials); return normalizeBailianMessagesUrl(baseUrl); } case "heroku": { - const baseUrl = credentials?.providerSpecificData?.baseUrl || this.config.baseUrl; + const baseUrl = this.resolveBaseUrl(credentials); return normalizeHerokuChatUrl(baseUrl); } case "databricks": { - const baseUrl = credentials?.providerSpecificData?.baseUrl || this.config.baseUrl; + const baseUrl = this.resolveBaseUrl(credentials); return normalizeDatabricksChatUrl(baseUrl); } case "datarobot": { - const baseUrl = credentials?.providerSpecificData?.baseUrl || this.config.baseUrl; + const baseUrl = this.resolveBaseUrl(credentials); return normalizeDataRobotChatUrl(baseUrl); } case "azure-ai": { @@ -229,11 +224,11 @@ export class DefaultExecutor extends BaseExecutor { forceResponses || credentials?.providerSpecificData?.apiType === "responses" ? "responses" : "chat"; - const baseUrl = credentials?.providerSpecificData?.baseUrl || this.config.baseUrl; + const baseUrl = this.resolveBaseUrl(credentials); return normalizeAzureAiChatUrl(baseUrl, apiType); } case "watsonx": { - const baseUrl = credentials?.providerSpecificData?.baseUrl || this.config.baseUrl; + const baseUrl = this.resolveBaseUrl(credentials); return normalizeWatsonxChatUrl(baseUrl); } case "oci": { @@ -243,33 +238,34 @@ export class DefaultExecutor extends BaseExecutor { forceResponses || credentials?.providerSpecificData?.apiType === "responses" ? "responses" : "chat"; - const baseUrl = credentials?.providerSpecificData?.baseUrl || this.config.baseUrl; + const baseUrl = this.resolveBaseUrl(credentials); return normalizeOciChatUrl(baseUrl, apiType); } case "sap": { - const baseUrl = credentials?.providerSpecificData?.baseUrl || this.config.baseUrl; + const baseUrl = this.resolveBaseUrl(credentials); return normalizeSapChatUrl(baseUrl); } case "xiaomi-mimo": { - const baseUrl = credentials?.providerSpecificData?.baseUrl || this.config.baseUrl; + const baseUrl = this.resolveBaseUrl(credentials); return normalizeXiaomiMimoChatUrl(baseUrl); } case "snowflake": { - const baseUrl = credentials?.providerSpecificData?.baseUrl || this.config.baseUrl; + const baseUrl = this.resolveBaseUrl(credentials); return normalizeSnowflakeChatUrl(baseUrl); } case "gigachat": { - const baseUrl = credentials?.providerSpecificData?.baseUrl || this.config.baseUrl; + const baseUrl = this.resolveBaseUrl(credentials); return normalizeGigachatChatUrl(baseUrl); } case "maritalk": { - const baseUrl = credentials?.providerSpecificData?.baseUrl || this.config.baseUrl; + const baseUrl = this.resolveBaseUrl(credentials); return buildMaritalkChatUrl(baseUrl); } case "siliconflow": { - const baseUrl = credentials?.providerSpecificData?.baseUrl || this.config.baseUrl; + const baseUrl = this.resolveBaseUrl(credentials); return normalizeOpenAIChatUrl(baseUrl); } + case "ollama-local": case "llama-cpp": case "lm-studio": case "modal": @@ -293,7 +289,7 @@ export class DefaultExecutor extends BaseExecutor { } case "zai": case "glm-coding-apikey": { - const zaiBaseUrl = credentials?.providerSpecificData?.baseUrl || this.config.baseUrl; + const zaiBaseUrl = this.resolveBaseUrl(credentials); return `${zaiBaseUrl}?beta=true`; } case "claude": @@ -333,40 +329,7 @@ export class DefaultExecutor extends BaseExecutor { } buildHeaders(credentials, stream = true, clientHeaders?: Record | null) { - const headers = { "Content-Type": "application/json", ...this.config.headers }; - - // Allow per-provider User-Agent override via environment variable. - const providerId = this.config?.id || this.provider; - if (providerId) { - const envKey = `${providerId.toUpperCase().replace(/[^A-Z0-9]/g, "_")}_USER_AGENT`; - const envUA = process.env[envKey]?.trim(); - if (envUA) { - headers["User-Agent"] = envUA; - if ("user-agent" in headers) { - headers["user-agent"] = envUA; - } - } - } - - // T07: resolve extra keys round-robin locally since DefaultExecutor overrides BaseExecutor buildHeaders - const extraKeys = - (credentials.providerSpecificData?.extraApiKeys as string[] | undefined) ?? []; - const selectedKeyId = (credentials.providerSpecificData as Record | undefined) - ?.selectedKeyId as string | undefined; - let effectiveKey = credentials.apiKey; - if (extraKeys.length > 0 && credentials.connectionId && credentials.apiKey) { - const resolved = resolveKeyForRequest( - credentials.connectionId, - credentials.apiKey, - extraKeys, - selectedKeyId ?? null - ); - effectiveKey = resolved?.key ?? credentials.apiKey; - if (resolved && credentials.providerSpecificData) { - (credentials.providerSpecificData as Record).selectedKeyId = - resolved.keyId; - } - } + const { headers, effectiveKey } = this.buildHeadersPreamble(credentials, stream); switch (this.provider) { case "gemini": @@ -544,25 +507,7 @@ export class DefaultExecutor extends BaseExecutor { // Forward client request metadata headers (from OpenCode or similar clients) // Allowlist-based: only specific x-opencode-* headers and User-Agent are forwarded if (clientHeaders) { - const clientUA = clientHeaders["User-Agent"] || clientHeaders["user-agent"]; - if (clientUA) { - setUserAgentHeader(headers, clientUA); - } - - const opencodeHeaderKeys = [ - "x-opencode-session", - "x-opencode-request", - "x-opencode-project", - "x-opencode-client", - ]; - for (const headerName of opencodeHeaderKeys) { - const value = Object.entries(clientHeaders).find( - ([key]) => key.toLowerCase() === headerName.toLowerCase() - )?.[1]; - if (value) { - headers[headerName] = value; - } - } + forwardOpencodeClientHeaders(headers, clientHeaders); // #3974: merge the client's negotiated anthropic-beta (allowlisted) into the // outbound set. The registry's static ANTHROPIC_BETA_CLAUDE_OAUTH lacks @@ -596,8 +541,7 @@ export class DefaultExecutor extends BaseExecutor { const record = body as Record; const rf = record.response_format as - | { type?: string; json_schema?: { schema?: unknown } } - | undefined; + { type?: string; json_schema?: { schema?: unknown } } | undefined; if (rf?.type !== "json_schema" || !rf.json_schema?.schema) return body; const schemaJson = JSON.stringify(rf.json_schema.schema, null, 2); @@ -691,7 +635,8 @@ export class DefaultExecutor extends BaseExecutor { // injection when `thinking` / `enable_thinking` is set. Skip injection in // those cases instead of unconditionally adding `stream_options`. const defaultsRecord = withDefaults as Record; - const bodyDisablesStreamOptions = defaultsRecord.stream !== undefined && defaultsRecord.stream !== true; + const bodyDisablesStreamOptions = + defaultsRecord.stream !== undefined && defaultsRecord.stream !== true; const qwenBlocksStreamOptions = this.provider === "qwen" && (Boolean(defaultsRecord.thinking) || Boolean(defaultsRecord.enable_thinking)); @@ -768,8 +713,7 @@ export class DefaultExecutor extends BaseExecutor { // ../translator/paramSupport.ts so adding one means editing one table. if (typeof withDefaults === "object" && withDefaults !== null) { const bodyRecord = withDefaults as Record; - const outboundModel = - typeof bodyRecord.model === "string" ? bodyRecord.model : model; + const outboundModel = typeof bodyRecord.model === "string" ? bodyRecord.model : model; stripUnsupportedParams(this.provider, outboundModel, bodyRecord); } diff --git a/open-sse/executors/forceResponsesUpstream.ts b/open-sse/executors/forceResponsesUpstream.ts new file mode 100644 index 0000000000..0c545d980e --- /dev/null +++ b/open-sse/executors/forceResponsesUpstream.ts @@ -0,0 +1,60 @@ +import { getOpenAICompatibleType } from "../services/provider.ts"; + +/** + * Force OpenAI-compatible upstreams onto the native `/responses` endpoint. + * + * A Responses-API-shaped request (`input` / `previous_response_id` / + * `max_output_tokens` / `reasoning`) that carries MCP (`namespace`) or + * `tool_search*` tools loses the Codex deferred tool-discovery mechanism when + * OmniRoute downgrades it to `/chat/completions` — so the MCP namespaces never + * surface to the model and `apply_patch` is mis-handled (#5483). Detecting that + * shape lets the executor pass it through natively instead of downgrading. + */ + +type CredentialsLike = { providerSpecificData?: Record | null }; + +function isRecord(value: unknown): value is Record { + return !!value && typeof value === "object" && !Array.isArray(value); +} + +export function shouldForceResponsesUpstream( + provider: string, + body: unknown, + credentials: CredentialsLike | null +): boolean { + if (!provider.startsWith("openai-compatible-")) return false; + if (!isRecord(body)) return false; + + const providerSpecificData = credentials?.providerSpecificData ?? null; + if (providerSpecificData?._omnirouteForceResponsesUpstream === true) return true; + if (getOpenAICompatibleType(provider, providerSpecificData) === "responses") return false; + + const hasResponsesShape = + body.input !== undefined || + body.previous_response_id !== undefined || + body.max_output_tokens !== undefined || + body.reasoning !== undefined; + if (!hasResponsesShape) return false; + + const tools = Array.isArray(body.tools) ? body.tools : []; + return tools.some((toolValue) => { + if (!isRecord(toolValue)) return false; + const toolType = typeof toolValue.type === "string" ? toolValue.type : ""; + return toolType === "namespace" || /^tool_search/.test(toolType); + }); +} + +export function withForcedResponsesUpstream( + provider: string, + body: unknown, + credentials: T +): T { + if (!shouldForceResponsesUpstream(provider, body, credentials)) return credentials; + return { + ...credentials, + providerSpecificData: { + ...credentials.providerSpecificData, + _omnirouteForceResponsesUpstream: true, + }, + } as T; +} diff --git a/open-sse/executors/github.ts b/open-sse/executors/github.ts index f2507bd9cf..5271500d31 100644 --- a/open-sse/executors/github.ts +++ b/open-sse/executors/github.ts @@ -149,6 +149,18 @@ export class GithubExecutor extends BaseExecutor { ); } + // GitHub Copilot's /chat/completions endpoint rejects a conversation that ends + // with an assistant message: "This model does not support assistant message + // prefill. The conversation must end with a user message." (HTTP 400). Anthropic + // clients such as newest Claude Desktop send a trailing assistant turn as a + // prefill seed — the Anthropic API honors it, but Copilot does not. Drop it here, + // scoped to the GitHub executor only (the shared translator/contextManager and + // other providers that DO honor prefill are untouched). + // Port of 9router#2143 (author: Manuel ). + if (Array.isArray(modifiedBody.messages)) { + modifiedBody.messages = this.dropTrailingAssistantPrefill(modifiedBody.messages); + } + // Config-driven strip of params unsupported by the target provider/model. // For GitHub Copilot this removes Claude-style `thinking` and // `reasoning_effort` for Claude models that reject them upstream @@ -188,6 +200,19 @@ export class GithubExecutor extends BaseExecutor { return { ...msg, content: cleanContent.length > 0 ? cleanContent : null }; } + // Remove trailing assistant message(s). GitHub Copilot's /chat/completions endpoint + // can't honor an assistant prefill and 400s unless the conversation ends with a + // non-assistant (user/tool) message. Never empties the array — an assistant-only + // conversation keeps its last message. No-op (same array reference) when the + // conversation already ends with a non-assistant message. + // Port of 9router#2143 (author: Manuel ). + dropTrailingAssistantPrefill(messages: any): any { + if (!Array.isArray(messages) || messages.length === 0) return messages; + let end = messages.length; + while (end > 1 && messages[end - 1]?.role === "assistant") end--; + return end === messages.length ? messages : messages.slice(0, end); + } + async execute(input: ExecuteInput) { const result = await super.execute(input); if (!result || !result.response) return result; diff --git a/open-sse/executors/huggingchat.ts b/open-sse/executors/huggingchat.ts index 5d2bb91a74..46db72840e 100644 --- a/open-sse/executors/huggingchat.ts +++ b/open-sse/executors/huggingchat.ts @@ -6,7 +6,8 @@ * * API flow: * 1. POST /chat/conversation { model } -> { conversationId } - * 2. POST /chat/conversation/{id} (multipart: data = JSON{inputs}, optional files) + * 2. GET /chat/api/v2/conversations/{id} -> { rootMessageId } + * 3. POST /chat/conversation/{id} (multipart: data = JSON{inputs, id}, optional files) * -> JSONL stream of MessageUpdate objects * * Streaming format (JSONL, not SSE): @@ -24,16 +25,18 @@ import { type ExecuteInput, } from "./base.ts"; import { FETCH_TIMEOUT_MS } from "../config/constants.ts"; +import { buildErrorBody, sanitizeErrorMessage } from "../utils/error.ts"; import { normalizeSessionCookieHeader } from "@/lib/providers/webCookieAuth"; const HUGGINGFACE_BASE = "https://huggingface.co"; const CONVERSATION_URL = `${HUGGINGFACE_BASE}/chat/conversation`; +const API_CONVERSATIONS_URL = `${HUGGINGFACE_BASE}/chat/api/v2/conversations`; const DEFAULT_COOKIE_NAME = "hf-chat"; const USER_AGENT = "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/149.0.0.0 Safari/537.36"; -const DEFAULT_MODEL = "meta-llama/Llama-3.3-70B-Instruct"; +const DEFAULT_MODEL = "baidu/ERNIE-4.5-VL-424B-A47B-Base-PT"; // -- Helpers ----------------------------------------------------------------- @@ -41,6 +44,10 @@ function normalizeHuggingChatCookieHeader(apiKey: string): string { return normalizeSessionCookieHeader(apiKey, DEFAULT_COOKIE_NAME); } +function isEncryptedCredentialBlob(value: unknown): boolean { + return typeof value === "string" && value.trim().startsWith("enc:v1:"); +} + function extractTextFromContent(content: unknown): string { if (typeof content === "string") return content.trim(); if (!Array.isArray(content)) return ""; @@ -109,6 +116,135 @@ function estimateTokens(text: string): number { return Math.max(1, Math.ceil((text || "").length / 4)); } +function getLocalTimezone(): string { + try { + return Intl.DateTimeFormat().resolvedOptions().timeZone || "UTC"; + } catch { + return "UTC"; + } +} + +async function readUpstreamErrorDetails(response: Response): Promise<{ + message: string | null; + details: unknown; +}> { + const contentType = response.headers.get("content-type") || ""; + const text = await response.text().catch(() => ""); + if (!text) return { message: null, details: null }; + + if (contentType.includes("json")) { + try { + const parsed = JSON.parse(text) as Record; + const message = + typeof parsed.message === "string" + ? parsed.message + : typeof parsed.error === "string" + ? parsed.error + : parsed.error && + typeof parsed.error === "object" && + typeof (parsed.error as Record).message === "string" + ? String((parsed.error as Record).message) + : null; + return { message: message ? sanitizeErrorMessage(message) : null, details: parsed }; + } catch { + // Fall through to text handling below. + } + } + + return { message: sanitizeErrorMessage(text), details: { body: text } }; +} + +function unwrapSuperjsonPayload(value: unknown): unknown { + if (!value || typeof value !== "object" || Array.isArray(value)) return value; + const record = value as Record; + return record.json && typeof record.json === "object" ? record.json : value; +} + +function extractInitialParentMessageId(value: unknown): string | null { + const payload = unwrapSuperjsonPayload(value); + if (!payload || typeof payload !== "object" || Array.isArray(payload)) return null; + + const record = payload as Record; + if (typeof record.rootMessageId === "string" && record.rootMessageId.trim()) { + return record.rootMessageId; + } + + const messages = Array.isArray(record.messages) ? record.messages : []; + const lastMessage = messages.at(-1); + if (lastMessage && typeof lastMessage === "object") { + const id = (lastMessage as Record).id; + if (typeof id === "string" && id.trim()) return id; + } + + return null; +} + +async function fetchInitialParentMessageId( + conversationId: string, + headers: Record, + signal: AbortSignal +): Promise { + const res = await fetch(`${API_CONVERSATIONS_URL}/${conversationId}`, { + method: "GET", + headers, + signal, + }); + if (!res.ok) return null; + + const text = await res.text().catch(() => ""); + if (!text) return null; + + try { + return extractInitialParentMessageId(JSON.parse(text)); + } catch { + return null; + } +} + +function splitCombinedSetCookieHeader(header: string): string[] { + return header + .split(/,(?=\s*[^;,=\s]+=)/) + .map((value) => value.trim()) + .filter(Boolean); +} + +function getSetCookieHeaders(headers: Headers): string[] { + const maybeGetSetCookie = (headers as Headers & { getSetCookie?: () => string[] }).getSetCookie; + if (typeof maybeGetSetCookie === "function") { + return maybeGetSetCookie.call(headers).filter(Boolean); + } + + const combined = headers.get("set-cookie"); + return combined ? splitCombinedSetCookieHeader(combined) : []; +} + +function parseSetCookiePair(setCookie: string): { name: string; value: string } | null { + const pair = setCookie.split(";", 1)[0]?.trim() || ""; + const eq = pair.indexOf("="); + if (eq <= 0) return null; + return { name: pair.slice(0, eq).trim(), value: pair.slice(eq + 1) }; +} + +function mergeCookieHeaderWithSetCookie(cookieHeader: string, setCookieHeaders: string[]): string { + const cookieMap = new Map(); + + for (const part of cookieHeader.split(";")) { + const trimmed = part.trim(); + if (!trimmed) continue; + const eq = trimmed.indexOf("="); + if (eq <= 0) continue; + cookieMap.set(trimmed.slice(0, eq).trim(), trimmed.slice(eq + 1)); + } + + for (const setCookie of setCookieHeaders) { + const parsed = parseSetCookiePair(setCookie); + if (!parsed || !parsed.value) continue; + cookieMap.set(parsed.name, parsed.value); + } + + return [...cookieMap.entries()].map(([name, value]) => `${name}=${value}`).join("; "); +} + function parseJsonlLine(line: string): { token?: string; done?: boolean; @@ -340,8 +476,7 @@ export class HuggingChatExecutor extends BaseExecutor { }> { const { model, body, stream, credentials, signal, log, upstreamExtraHeaders } = input; const messages = (body as Record).messages as - | Array> - | undefined; + Array> | undefined; if (!messages || !Array.isArray(messages) || messages.length === 0) { return { @@ -357,7 +492,26 @@ export class HuggingChatExecutor extends BaseExecutor { }; } - const cookieHeader = normalizeHuggingChatCookieHeader(credentials.apiKey || ""); + if (isEncryptedCredentialBlob(credentials.apiKey)) { + return { + response: new Response( + JSON.stringify({ + error: { + message: + "HuggingChat credentials are encrypted but STORAGE_ENCRYPTION_KEY is not loaded. " + + "Restore the encryption key or re-save the HuggingChat cookie.", + type: "auth_error", + }, + }), + { status: 401, headers: { "Content-Type": "application/json" } } + ), + url: CONVERSATION_URL, + headers: {}, + transformedBody: body, + }; + } + + let cookieHeader = normalizeHuggingChatCookieHeader(credentials.apiKey || ""); if (!cookieHeader) { return { response: new Response( @@ -419,6 +573,7 @@ export class HuggingChatExecutor extends BaseExecutor { if (!createRes.ok) { const status = createRes.status; + const upstreamError = await readUpstreamErrorDetails(createRes); let message = `HuggingChat conversation creation failed (HTTP ${status})`; if (status === 401 || status === 403) { message = @@ -427,9 +582,12 @@ export class HuggingChatExecutor extends BaseExecutor { } else if (status === 429) { message = "HuggingChat rate limited. Wait a moment and retry."; } + if (upstreamError.message) { + message = `${message}: ${upstreamError.message}`; + } return { response: new Response( - JSON.stringify({ error: { message, type: "upstream_error", code: `HTTP_${status}` } }), + JSON.stringify(buildErrorBody(status, message, upstreamError.details)), { status, headers: { "Content-Type": "application/json" } } ), url: CONVERSATION_URL, @@ -440,6 +598,9 @@ export class HuggingChatExecutor extends BaseExecutor { const createData = (await createRes.json()) as Record; conversationId = createData.conversationId as string; + const createSetCookieHeaders = getSetCookieHeaders(createRes.headers); + cookieHeader = mergeCookieHeaderWithSetCookie(cookieHeader, createSetCookieHeaders); + baseHeaders.Cookie = cookieHeader; if (!conversationId) { return { @@ -457,8 +618,6 @@ export class HuggingChatExecutor extends BaseExecutor { transformedBody: body, }; } - - log?.debug?.("HUGGINGCHAT", `Created conversation: ${conversationId}`); } catch (err) { const message = err instanceof Error ? err.message : String(err); log?.error?.("HUGGINGCHAT", `Conversation creation failed: ${message}`); @@ -476,15 +635,41 @@ export class HuggingChatExecutor extends BaseExecutor { } // -- Step 2: Send message ----------------------------------------------- + const parentMessageId = await fetchInitialParentMessageId( + conversationId, + baseHeaders, + combinedSignal + ); + if (!parentMessageId) { + return { + response: new Response( + JSON.stringify({ + error: { + message: "HuggingChat did not return an initial parent message id", + type: "upstream_error", + }, + }), + { status: 502, headers: { "Content-Type": "application/json" } } + ), + url: `${API_CONVERSATIONS_URL}/${conversationId}`, + headers: baseHeaders, + transformedBody: body, + }; + } + const messageUrl = `${CONVERSATION_URL}/${conversationId}`; const formData = new FormData(); - formData.append( - "data", - JSON.stringify({ - inputs, - id: crypto.randomUUID(), - }) - ); + const sendDataPayload: Record = { + inputs, + is_retry: false, + is_continue: false, + generationId: crypto.randomUUID(), + selectedMcpServerNames: [], + selectedMcpServers: [], + timezone: getLocalTimezone(), + id: parentMessageId, + }; + formData.append("data", JSON.stringify(sendDataPayload)); mergeUpstreamExtraHeaders(baseHeaders, upstreamExtraHeaders); @@ -508,12 +693,13 @@ export class HuggingChatExecutor extends BaseExecutor { ), url: messageUrl, headers: baseHeaders, - transformedBody: body, + transformedBody: sendDataPayload, }; } if (!upstreamResponse.ok) { const status = upstreamResponse.status; + const upstreamError = await readUpstreamErrorDetails(upstreamResponse); let message = `HuggingChat returned HTTP ${status}`; if (status === 401 || status === 403) { message = "HuggingChat auth failed -- session cookie may be expired."; @@ -522,14 +708,17 @@ export class HuggingChatExecutor extends BaseExecutor { } else if (status === 404) { message = `HuggingChat model not found: ${resolvedModel}. Check the model ID.`; } + if (upstreamError.message) { + message = `${message}: ${upstreamError.message}`; + } return { response: new Response( - JSON.stringify({ error: { message, type: "upstream_error", code: `HTTP_${status}` } }), + JSON.stringify(buildErrorBody(status, message, upstreamError.details)), { status, headers: { "Content-Type": "application/json" } } ), url: messageUrl, headers: baseHeaders, - transformedBody: body, + transformedBody: sendDataPayload, }; } @@ -543,7 +732,7 @@ export class HuggingChatExecutor extends BaseExecutor { ), url: messageUrl, headers: baseHeaders, - transformedBody: body, + transformedBody: sendDataPayload, }; } @@ -586,7 +775,7 @@ export class HuggingChatExecutor extends BaseExecutor { }), url: messageUrl, headers: baseHeaders, - transformedBody: body, + transformedBody: sendDataPayload, }; } @@ -617,7 +806,7 @@ export class HuggingChatExecutor extends BaseExecutor { ), url: messageUrl, headers: baseHeaders, - transformedBody: body, + transformedBody: sendDataPayload, }; } } diff --git a/open-sse/executors/kimi-web.ts b/open-sse/executors/kimi-web.ts index 6249480286..f45e9d1072 100644 --- a/open-sse/executors/kimi-web.ts +++ b/open-sse/executors/kimi-web.ts @@ -1,55 +1,261 @@ /** - * KimiWebExecutor — Moonshot AI Chat via kimi.moonshot.cn + * KimiWebExecutor — Moonshot AI Chat via www.kimi.com (international) * - * Routes requests through Kimi's consumer chat API. - * Chinese market provider with strong long-context support. + * Routes requests through Kimi's consumer chat API on the international domain. + * Originally this executor targeted `kimi.moonshot.cn` (mainland-CN consumer + * chat). That domain now redirects every visitor outside CN to + * `https://www.kimi.com/`, which speaks a completely different API surface: * - * Endpoint: POST https://kimi.moonshot.cn/api/chat - * Auth: Session cookie from kimi.moonshot.cn + * - Endpoint: POST /apiv2/kimi.gateway.chat.v1.ChatService/Chat + * - Protocol: Connect-RPC (unary envelope framing — 5-byte header + JSON) + * - Auth: `Authorization: Bearer ` + `Cookie: kimi-auth=` + * - Body: Connect-framed `{scenario, message:{role,blocks:[{text:{content}}]}, + * options:{thinking,enable_plugin}}` + * - Response: Connect-framed stream of events carrying deltas with one of + * `mask: "block.text.content"` (answer) or + * `mask: "block.think.content"` (reasoning), emitted via + * `op: "set"` (initial) and `op: "append"` (incremental). + * + * Cookie handling: the user pastes their full Cookie header from www.kimi.com. + * We extract the `kimi-auth` JWT from it (it is the only cookie the upstream + * actually consults) and use it both as the Bearer token and as the Cookie we + * send back, so we don't leak the user's analytics cookies (Ga, CF, HM, ...). + * + * The `x-msh-*` / `x-traffic-id` / `x-msh-shield-data` headers the SPA sends + * are NOT required — verified by stripping them one at a time against a live + * session; the upstream returns the same response either way. */ import { BaseExecutor, type ExecuteInput } from "./base.ts"; -import { makeExecutorErrorResult as makeErrorResult, normalizeCookie } from "../utils/error.ts"; +import { makeExecutorErrorResult as makeErrorResult, sanitizeErrorMessage } from "../utils/error.ts"; +import { extractKimiJwt } from "@/lib/providers/webCookieAuth"; -const BASE_URL = "https://kimi.moonshot.cn"; -const CHAT_URL = `${BASE_URL}/api/chat`; +export { extractKimiJwt }; + +const BASE_URL = "https://www.kimi.com"; +const CHAT_URL = `${BASE_URL}/apiv2/kimi.gateway.chat.v1.ChatService/Chat`; const USER_AGENT = "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/149.0.0.0 Safari/537.36"; +const DEFAULT_SCENARIO = "SCENARIO_K2D5"; + +/** Wrap a JSON message in the 5-byte Connect streaming envelope (flags + length). */ +export function frameConnectMessage(json: string): Uint8Array { + const payload = new TextEncoder().encode(json); + const framed = new Uint8Array(5 + payload.length); + framed[0] = 0; // flags: 0 = uncompressed + const len = payload.length; + framed[1] = (len >>> 24) & 0xff; + framed[2] = (len >>> 16) & 0xff; + framed[3] = (len >>> 8) & 0xff; + framed[4] = len & 0xff; + framed.set(payload, 5); + return framed; +} + +interface ConnectFrame { + flags: number; + message: Record | null; +} + +/** + * ponytail: cap a single Connect frame at 8 MiB. Kimi's largest legitimate + * event is well under 1 KiB (a delta or stage transition); anything bigger + * means the upstream is misbehaving or an attacker controls the response and + * is trying to OOM the proxy by sending a header claiming a huge length. + * The non-streaming accumulator would otherwise grow unbounded. If you ever + * see this tripping in production, raise the ceiling and add a regression + * test — but never remove it. + */ +const MAX_FRAME_LEN = 8 * 1024 * 1024; + +/** + * Decode one Connect frame from a stream buffer. + * Returns: + * - `consumed: 0` if there isn't enough data yet (need more bytes) + * - `consumed: -1` if the frame header claims a length above MAX_FRAME_LEN + * (caller must treat this as a stream-fatal protocol error) + * - `consumed: N` + the parsed frame otherwise + */ +export function decodeConnectFrame(buf: Uint8Array, byteOffset: number): { consumed: number; frame: ConnectFrame | null } { + if (byteOffset + 5 > buf.length) return { consumed: 0, frame: null }; + const flags = buf[byteOffset]; + const len = + (buf[byteOffset + 1] << 24) | + (buf[byteOffset + 2] << 16) | + (buf[byteOffset + 3] << 8) | + buf[byteOffset + 4]; + // Sign-extend the high bit back to negative when len was read as signed. + const msgLen = len < 0 ? len + 0x100000000 : len; + if (msgLen > MAX_FRAME_LEN) return { consumed: -1, frame: null }; + if (byteOffset + 5 + msgLen > buf.length) return { consumed: 0, frame: null }; + + const payload = buf.subarray(byteOffset + 5, byteOffset + 5 + msgLen); + let message: Record | null = null; + if (msgLen > 0) { + try { + message = JSON.parse(new TextDecoder().decode(payload)); + } catch { + message = null; + } + } + return { consumed: 5 + msgLen, frame: { flags, message } }; +} + +type DeltaKind = "text" | "think" | null; + +/** + * Extract a content delta + kind from a Connect frame message. + * + * The chat stream uses two ops against two masks: + * - `op: "set"` on `block.text` / `block.think` → first chunk + * - `op: "append"` on `block.text.content` / `block.think.content` → subsequent chunks + * + * Anything else (heartbeats, chat/message metadata, stage transitions) is + * suppressed; we only surface text to the client. + */ +export function extractDelta(msg: Record | null): { kind: DeltaKind; text: string } | null { + if (!msg) return null; + const op = String(msg.op ?? ""); + const mask = String(msg.mask ?? ""); + const block = (msg.block ?? {}) as Record; + + // `op: append` carries a delta string under `block..content`. + if (op === "append") { + if (mask === "block.text.content") { + const text = String(((block.text ?? {}) as Record).content ?? ""); + return text ? { kind: "text", text } : null; + } + if (mask === "block.think.content") { + const text = String(((block.think ?? {}) as Record).content ?? ""); + return text ? { kind: "think", text } : null; + } + return null; + } + + // `op: set` on `block.text` / `block.think` carries the initial content. + if (op === "set") { + if (mask === "block.text") { + const text = String(((block.text ?? {}) as Record).content ?? ""); + return text ? { kind: "text", text } : null; + } + if (mask === "block.think") { + const text = String(((block.think ?? {}) as Record).content ?? ""); + return text ? { kind: "think", text } : null; + } + } + return null; +} + +export function isEndOfStream(msg: Record | null): boolean { + if (!msg) return false; + // Assistant message flipped to COMPLETED. + const message = (msg.message ?? null) as Record | null; + if (message && String(message.status ?? "") === "MESSAGE_STATUS_COMPLETED" && String(message.role ?? "") === "assistant") { + return true; + } + return false; +} + +/** + * Fold a multi-turn OpenAI `messages` array into a single Kimi user turn. + * + * Limitations (kimi-web is a single-turn consumer chat, not an agentic API): + * - `tool` and `function` role messages are silently dropped — Kimi's web + * chat has no concept of tool results, so agentic flows should use the + * `kimi-coding` (api.kimi.com) provider instead. + * - Assistant `tool_calls` and image content parts are stringified into + * text, which loses structure. Acceptable for free-text continuation, + * unacceptable for tool-round-trip — same workaround: use kimi-coding. + */ +export function foldMessages(messages: Array<{ role: string; content: unknown }>): string { + let system = ""; + let user = ""; + for (const m of messages) { + const text = typeof m.content === "string" ? m.content : JSON.stringify(m.content ?? ""); + if (m.role === "system") { + system += (system ? "\n\n" : "") + text; + } else if (m.role === "user") { + // Kimi's web chat is single-turn; keep only the latest user content but + // preserve prior assistant text for continuity when present. + user = user ? `${user}\n\n${text}` : text; + } else if (m.role === "assistant") { + user = user ? `${user}\n\nAssistant: ${text}` : `Assistant: ${text}`; + } + } + return system ? `${system}\n\n${user}` : user; +} + export class KimiWebExecutor extends BaseExecutor { constructor() { - super("kimi-web", { id: "kimi-web", baseUrl: "https://kimi.moonshot.cn" }); + super("kimi-web", { id: "kimi-web", baseUrl: BASE_URL }); + } + + private buildKimiHeaders(jwt: string): Record { + const headers: Record = { + "Content-Type": "application/connect+json", + Accept: "*/*", + "User-Agent": USER_AGENT, + Origin: BASE_URL, + Referer: `${BASE_URL}/`, + "connect-protocol-version": "1", + }; + if (jwt) { + headers["Authorization"] = `Bearer ${jwt}`; + headers["Cookie"] = `kimi-auth=${jwt}`; + } + return headers; + } + + private buildRequestBody(prompt: string, wantThinking: boolean): string { + return JSON.stringify({ + scenario: DEFAULT_SCENARIO, + tools: [{ type: "TOOL_TYPE_SEARCH", search: {} }, { type: "TOOL_TYPE_CRON_JOB" }], + message: { + role: "user", + blocks: [{ message_id: "", text: { content: prompt } }], + scenario: DEFAULT_SCENARIO, + }, + options: { thinking: wantThinking, enable_plugin: true }, + }); } async execute(input: ExecuteInput) { const { body, credentials, signal, stream: wantStream } = input; const bodyObj = (body || {}) as Record; - const rawCookie = normalizeCookie(String(credentials?.apiKey ?? "").trim()); - const messages = (bodyObj.messages as Array<{ role: string; content: string }>) || []; + const rawCredential = String(credentials?.apiKey ?? "").trim(); + const jwt = extractKimiJwt(rawCredential); + if (!jwt) { + return makeErrorResult( + 400, + "Missing Kimi session — paste the full Cookie header from www.kimi.com (must contain kimi-auth=) or just the JWT itself.", + body, + CHAT_URL + ); + } + + const messages = (bodyObj.messages as Array<{ role: string; content: unknown }>) || []; const modelId = (bodyObj.model as string) || "kimi-default"; + // Decide thinking intent. A user sending `reasoning_effort: "none"` is + // explicit — honour it even when the model id suggests a thinking variant. + // Otherwise thinking models (kimi-k2.6 etc.) default to thinking on. + const modelWantsThinking = /k2\.6|k2-6|think/i.test(modelId); + const wantThinking = bodyObj.reasoning_effort === "none" ? false : modelWantsThinking; - const reqBody = { - messages: messages.map((m) => ({ role: m.role, content: m.content })), - model: modelId, - stream: wantStream, - max_tokens: (bodyObj.max_tokens as number) || 4096, - }; + const prompt = foldMessages(messages); + const reqBody = this.buildRequestBody(prompt, wantThinking); + const reqHeaders = this.buildKimiHeaders(jwt); - const reqHeaders: Record = { - "Content-Type": "application/json", - "User-Agent": USER_AGENT, - Accept: wantStream ? "text/event-stream" : "application/json", - Referer: `${BASE_URL}/`, - Origin: BASE_URL, - }; - if (rawCookie) reqHeaders.Cookie = rawCookie; + // Connect framing wraps the JSON body in a 5-byte envelope. Without it the + // upstream returns `invalid_argument` for every request. + const framedBody = frameConnectMessage(reqBody); let upstream: Response; try { upstream = await fetch(CHAT_URL, { method: "POST", headers: reqHeaders, - body: JSON.stringify(reqBody), + body: new Uint8Array(framedBody), signal, }); } catch (err) { @@ -63,93 +269,171 @@ export class KimiWebExecutor extends BaseExecutor { if (!upstream.ok) { const errText = await upstream.text().catch(() => ""); - return makeErrorResult(upstream.status, `Kimi error: ${errText}`, body, CHAT_URL); + return makeErrorResult(upstream.status, `Kimi error: ${sanitizeErrorMessage(errText)}`, body, CHAT_URL); } - if (!wantStream) { - const data = (await upstream.json()) as Record; - const content = - (data?.choices as Array<{ message?: { content?: string } }>)?.[0]?.message?.content || - (data?.content as string) || - ""; + const encoder = new TextEncoder(); + const id = `chatcmpl-kimi-${Date.now()}`; + const created = Math.floor(Date.now() / 1000); + + const emitChunk = ( + controller: ReadableStreamDefaultController, + delta: Record, + finish: string | null = null + ) => { + const chunk = { + id, + object: "chat.completion.chunk", + created, + model: modelId, + choices: [{ index: 0, delta, finish_reason: finish }], + }; + controller.enqueue(encoder.encode(`data: ${JSON.stringify(chunk)}\n\n`)); + }; + + // The upstream is a Connect-framed stream regardless of whether the + // client asked for SSE — Kimi always streams. For non-streaming clients + // we buffer the full response below. + const sourceStream = upstream.body ?? new ReadableStream({ start: (c) => c.close() }); + + if (wantStream) { + const outStream = new ReadableStream({ + async start(controller) { + const reader = sourceStream.getReader(); + let buffer = new Uint8Array(0); + let emittedRole = false; + try { + while (true) { + const { done, value } = await reader.read(); + if (done) break; + if (value) { + const merged = new Uint8Array(buffer.length + value.length); + merged.set(buffer, 0); + merged.set(value, buffer.length); + buffer = merged; + + let offset = 0; + while (offset < buffer.length) { + const { consumed, frame } = decodeConnectFrame(buffer, offset); + if (consumed === -1) { + // Frame header claims a length above MAX_FRAME_LEN — stream-fatal. + controller.error(new Error("Kimi Connect frame exceeded MAX_FRAME_LEN")); + return; + } + if (consumed === 0) break; // need more bytes + offset += consumed; + if (!frame?.message) continue; + + const delta = extractDelta(frame.message); + if (delta) { + if (!emittedRole) { + emittedRole = true; + emitChunk(controller, { role: "assistant", content: "" }); + } + if (delta.kind === "think") { + emitChunk(controller, { reasoning_content: delta.text }); + } else { + emitChunk(controller, { content: delta.text }); + } + } + if (isEndOfStream(frame.message)) { + emitChunk(controller, {}, "stop"); + controller.enqueue(encoder.encode("data: [DONE]\n\n")); + controller.close(); + return; + } + } + // Compact the buffer. + buffer = buffer.subarray(offset); + } + } + // Stream ended without an explicit COMPLETED marker — flush a stop. + if (!emittedRole) { + emitChunk(controller, { role: "assistant", content: "" }); + } + emitChunk(controller, {}, "stop"); + controller.enqueue(encoder.encode("data: [DONE]\n\n")); + controller.close(); + } catch (err) { + if (!signal?.aborted) { + try { + controller.error(err); + } catch { + /* controller already closed */ + } + } + } + }, + }); + return { - response: new Response( - JSON.stringify({ - id: `chatcmpl-kimi-${Date.now()}`, - object: "chat.completion", - created: Math.floor(Date.now() / 1000), - model: modelId, - choices: [{ index: 0, message: { role: "assistant", content }, finish_reason: "stop" }], - }), - { headers: { "Content-Type": "application/json" } } - ), + response: new Response(outStream, { + headers: { + "Content-Type": "text/event-stream", + "Cache-Control": "no-cache", + Connection: "keep-alive", + }, + }), url: CHAT_URL, headers: reqHeaders, - transformedBody: reqBody, + transformedBody: JSON.parse(reqBody), }; } - // Streaming - const encoder = new TextEncoder(); - const decoder = new TextDecoder(); - const stream = new ReadableStream({ - async start(controller) { - const reader = upstream.body?.getReader(); - if (!reader) { - controller.close(); - return; - } - let buffer = ""; - try { - while (true) { - const { done, value } = await reader.read(); - if (done) break; - buffer += decoder.decode(value, { stream: true }); - const lines = buffer.split("\n"); - buffer = lines.pop() || ""; - for (const line of lines) { - if (!line.startsWith("data:")) continue; - const data = line.slice(5).trim(); - if (data === "[DONE]") { - controller.enqueue(encoder.encode("data: [DONE]\n\n")); - continue; - } - try { - const parsed = JSON.parse(data); - const text = parsed.choices?.[0]?.delta?.content || ""; - if (text) { - const chunk = { - id: `chatcmpl-kimi-${Date.now()}`, - object: "chat.completion.chunk", - created: Math.floor(Date.now() / 1000), - model: modelId, - choices: [{ index: 0, delta: { content: text }, finish_reason: null }], - }; - controller.enqueue(encoder.encode(`data: ${JSON.stringify(chunk)}\n\n`)); - } - } catch {} - } - } - } catch (err) { - if (!signal?.aborted) controller.error(err); - } finally { - controller.enqueue(encoder.encode("data: [DONE]\n\n")); - controller.close(); - } - }, - }); + // Non-streaming: collect all deltas into a single chat.completion JSON. + let answer = ""; + let reasoning = ""; + const reader = sourceStream.getReader(); + let buffer = new Uint8Array(0); + try { + while (true) { + const { done, value } = await reader.read(); + if (done) break; + if (!value) continue; + const merged = new Uint8Array(buffer.length + value.length); + merged.set(buffer, 0); + merged.set(value, buffer.length); + buffer = merged; + let offset = 0; + while (offset < buffer.length) { + const { consumed, frame } = decodeConnectFrame(buffer, offset); + if (consumed === -1) break; // oversized frame — abort, return what we have + if (consumed === 0) break; + offset += consumed; + if (!frame?.message) continue; + const delta = extractDelta(frame.message); + if (delta) { + if (delta.kind === "think") reasoning += delta.text; + else answer += delta.text; + } + if (isEndOfStream(frame.message)) { + offset = buffer.length; // drain + break; + } + } + buffer = buffer.subarray(offset); + } + } catch { + /* best-effort — return what we have */ + } + + const message: Record = { role: "assistant", content: answer }; + if (reasoning) message.reasoning_content = reasoning; + const completion = { + id, + object: "chat.completion", + created, + model: modelId, + choices: [{ index: 0, message, finish_reason: "stop" }], + }; return { - response: new Response(stream, { - headers: { - "Content-Type": "text/event-stream", - "Cache-Control": "no-cache", - Connection: "keep-alive", - }, + response: new Response(JSON.stringify(completion), { + headers: { "Content-Type": "application/json" }, }), url: CHAT_URL, headers: reqHeaders, - transformedBody: reqBody, + transformedBody: JSON.parse(reqBody), }; } } diff --git a/open-sse/executors/opencode.ts b/open-sse/executors/opencode.ts index 6fccf7117d..60bde9d37c 100644 --- a/open-sse/executors/opencode.ts +++ b/open-sse/executors/opencode.ts @@ -1,10 +1,4 @@ -import { randomUUID } from "crypto"; -import { - BaseExecutor, - setUserAgentHeader, - type ExecuteInput, - type ProviderCredentials, -} from "./base.ts"; +import { BaseExecutor, type ExecuteInput, type ProviderCredentials } from "./base.ts"; import { PROVIDERS } from "../config/constants.ts"; import { getModelTargetFormat } from "../config/providerModels.ts"; import { @@ -12,6 +6,7 @@ import { isThinkingMessageModel, } from "../utils/reasoningContentInjector.ts"; import { runWithProxyContext } from "../utils/proxyFetch.ts"; +import { forwardOpencodeClientHeaders } from "../utils/opencodeHeaders.ts"; /** * Per-account proxy configuration, persisted by NoAuthAccountCard under @@ -42,6 +37,22 @@ interface OpencodeAccountState { const OPENCODE_COOLDOWN_BASE_MS = 5_000; const OPENCODE_COOLDOWN_MAX_MS = 60_000; +const EFFORT_LEVELS = ["low", "medium", "high", "max"] as const; + +/** + * Parse a DeepSeek V4 Pro model string with an effort-level suffix. + * e.g. "deepseek-v4-pro-low" → { baseModel: "deepseek-v4-pro", effort: "low" } + * Returns null if the model doesn't match the pattern. + */ +function parseDeepSeekEffortLevel(model: string): { baseModel: string; effort: string } | null { + const m = String(model || ""); + const matchedLevel = EFFORT_LEVELS.find((level) => m.endsWith(`-${level}`)); + if (!matchedLevel) return null; + const baseModel = m.slice(0, -matchedLevel.length - 1); + if (baseModel.toLowerCase() !== "deepseek-v4-pro") return null; + return { baseModel: "deepseek-v4-pro", effort: matchedLevel }; +} + export class OpencodeExecutor extends BaseExecutor { _requestFormat: string | null = null; @@ -159,7 +170,9 @@ export class OpencodeExecutor extends BaseExecutor { log?.info?.( "OPENCODE", `dispatch via account ${masked} (idx ${attempt + 1}/${this.accounts.length})` + - (account.proxy ? ` through proxy ${account.proxy.host}:${account.proxy.port}` : " direct") + (account.proxy + ? ` through proxy ${account.proxy.host}:${account.proxy.port}` + : " direct") ); // Pin egress to this account's proxy for the whole BaseExecutor dispatch @@ -173,10 +186,7 @@ export class OpencodeExecutor extends BaseExecutor { const status = result.response.status; if (status === 429) { this.markCooldown(account); - log?.warn?.( - "OPENCODE", - `Rate limited (429) on account ${masked}, rotating to next…` - ); + log?.warn?.("OPENCODE", `Rate limited (429) on account ${masked}, rotating to next…`); continue; } @@ -239,59 +249,9 @@ export class OpencodeExecutor extends BaseExecutor { } if (clientHeaders) { - const clientUA = clientHeaders["User-Agent"] || clientHeaders["user-agent"]; - if (clientUA) { - setUserAgentHeader(headers, clientUA); - } - - // Forward OpenCode request metadata headers from client - const findClientHeader = (name: string) => - Object.entries(clientHeaders).find( - ([key]) => key.toLowerCase() === name.toLowerCase() - )?.[1]; - - const opencodeHeaderKeys = [ - "x-opencode-session", - "x-opencode-request", - "x-opencode-project", - "x-opencode-client", - ]; - for (const headerName of opencodeHeaderKeys) { - const value = findClientHeader(headerName); - if (value) { - headers[headerName] = value; - } - } - - // #4022: OpenCode CLI only emits x-opencode-* headers when the provider id - // starts with "opencode". For a custom-named provider (e.g. "omniroute") it - // instead sends x-session-affinity / X-Session-Id, which both carry the same - // OpenCode sessionID. Map that session id onto x-opencode-session so session - // continuity to the opencode.ai upstream works regardless of how the user - // named the provider. Scoped to this executor (opencode.ai/zen upstreams - // only) — the generic DefaultExecutor intentionally does NOT do this, to - // avoid leaking the client session id to arbitrary third-party upstreams. - if (!headers["x-opencode-session"]) { - const sessionAffinity = - findClientHeader("x-session-affinity") || findClientHeader("x-session-id"); - if (sessionAffinity) { - headers["x-opencode-session"] = sessionAffinity; - - // #4465: a custom-named provider only reaches this fallback because the - // OpenCode CLI did NOT emit the x-opencode-* set (it only does so when the - // provider id starts with "opencode"). It therefore also dropped - // x-opencode-request, a per-request correlation id. Synthesize one so these - // users are not disadvantaged versus opencode-prefixed providers on the - // opencode.ai upstream. x-opencode-client / x-opencode-project are NOT - // fabricated: their valid values are opencode-internal and inventing them - // could be rejected upstream — they remain forward-only above. Scoped to this - // executor (opencode.ai/zen) and only to the fallback path, so the direct - // OpenCode CLI flow (which controls its own request id) is untouched. - if (!headers["x-opencode-request"]) { - headers["x-opencode-request"] = randomUUID(); - } - } - } + forwardOpencodeClientHeaders(headers, clientHeaders, { + synthesizeRequestId: true, + }); } void model; @@ -316,16 +276,11 @@ export class OpencodeExecutor extends BaseExecutor { } if (modifiedBody && typeof modifiedBody === "object" && !Array.isArray(modifiedBody)) { const mb = modifiedBody as Record; - const m = String(model || ""); - const effortLevels = ["low", "medium", "high", "max"] as const; - const matchedLevel = effortLevels.find((level) => m.endsWith(`-${level}`)); - if (matchedLevel) { - const base = m.slice(0, -matchedLevel.length - 1); - if (base.toLowerCase() === "deepseek-v4-pro") { - mb.model = "deepseek-v4-pro"; - if (mb.reasoning_effort === undefined) { - mb.reasoning_effort = matchedLevel; - } + const parsed = parseDeepSeekEffortLevel(model); + if (parsed) { + mb.model = parsed.baseModel; + if (mb.reasoning_effort === undefined) { + mb.reasoning_effort = parsed.effort; } } } diff --git a/open-sse/executors/qwen-web.ts b/open-sse/executors/qwen-web.ts index b72e92c77d..a61072d283 100644 --- a/open-sse/executors/qwen-web.ts +++ b/open-sse/executors/qwen-web.ts @@ -43,6 +43,13 @@ const USER_AGENT = const BX_VERSION = "2.5.36"; const BX_UMIDTOKEN_FALLBACK = "T2gA0000000000000000000000000000000000000000"; +// Qwen SPA version — required by the v2 chat completion endpoint. Without this +// header the upstream returns HTTP 200 with `{"success":false,"data":{"code":"Bad_Request"}}` +// for every completion request, even with a valid session. The version string is +// the SPA build identifier shipped in the React client's `version` request header. +// Pinned from a live capture (2026-07); bump if Qwen ships a breaking change. +const QWEN_SPA_VERSION = "0.2.66"; + const MODEL_ALIASES: Record = { // Legacy OmniRoute ids → current upstream catalog (GET /api/models). "qwen-plus": "qwen3.7-plus", @@ -96,6 +103,7 @@ export class QwenWebExecutor extends BaseExecutor { Origin: BASE_URL, Referer: chatId ? `${BASE_URL}/c/${chatId}` : `${BASE_URL}/`, source: "web", + version: QWEN_SPA_VERSION, "x-request-id": uuid(), "bx-v": BX_VERSION, "bx-umidtoken": BX_UMIDTOKEN_FALLBACK, diff --git a/open-sse/executors/t3-chat-web.ts b/open-sse/executors/t3-chat-web.ts index c3119fcca7..d7bbfcfd62 100644 --- a/open-sse/executors/t3-chat-web.ts +++ b/open-sse/executors/t3-chat-web.ts @@ -14,7 +14,7 @@ */ import { BaseExecutor, type ExecuteInput } from "./base.ts"; -import { sanitizeErrorMessage } from "../utils/error.ts"; +import { errorResponse } from "../utils/error.ts"; import { prepareToolMessages, buildToolAwareResult } from "../translator/webTools.ts"; // ─── Constants ─────────────────────────────────────────────────────────────── @@ -110,16 +110,7 @@ export function validateT3Credentials(creds: T3ChatCredentials | null | undefine } function buildErrorResponse(status: number, message: string): Response { - return new Response( - JSON.stringify({ - error: { - message: sanitizeErrorMessage(message), - type: "upstream_error", - code: `HTTP_${status}`, - }, - }), - { status, headers: { "Content-Type": "application/json" } } - ); + return errorResponse(status, message); } /** diff --git a/open-sse/handlers/chatCore.ts b/open-sse/handlers/chatCore.ts index 8f9a810cf3..4535402de7 100644 --- a/open-sse/handlers/chatCore.ts +++ b/open-sse/handlers/chatCore.ts @@ -37,6 +37,7 @@ import { import { buildStreamingResponseHeaders, materializeDeduplicatedExecutionResult, + stripNextMiddlewareControlHeaders, stripStaleForwardingHeaders, } from "./chatCore/responseHeaders.ts"; import { @@ -62,6 +63,7 @@ import { checkHeapPressureGuard } from "../utils/heapPressure.ts"; import { normalizeHeaders } from "../utils/headers.ts"; import { resolveChatCoreRequestFormat } from "./chatCore/requestFormat.ts"; import { resolveChatCoreTargetFormat } from "./chatCore/targetFormat.ts"; +import { defaultClaudeToolType } from "./chatCore/claudeToolDefaults.ts"; import { injectSystemPrompt, injectCustomSystemPrompt } from "../services/systemPrompt.ts"; import { translateRequest, needsTranslation } from "../translator/index.ts"; import { FORMATS } from "../translator/formats.ts"; @@ -74,11 +76,9 @@ import { withBodyTimeout, } from "../utils/stream.ts"; import { ensureStreamReadiness } from "../utils/streamReadiness.ts"; -import { - resolveSuppressThinkClose, - THINKING_MARKER_HEADER, -} from "../utils/thinkCloseMarker.ts"; +import { resolveSuppressThinkClose, THINKING_MARKER_HEADER } from "../utils/thinkCloseMarker.ts"; import { resolveStreamReadinessTimeout } from "../utils/streamReadinessPolicy.ts"; +import { resolveAgentGoalPolicy } from "../utils/agentGoalPolicy.ts"; import { createStreamController } from "../utils/streamHandler.ts"; import * as streamFailure from "../utils/streamFailureFinalization.ts"; import { createSseHeartbeatTransform, shapeForClientFormat } from "../utils/sseHeartbeat.ts"; @@ -126,12 +126,16 @@ import { FETCH_BODY_TIMEOUT_MS, PROVIDER_MAX_TOKENS, STREAM_IDLE_TIMEOUT_MS, + STREAM_READINESS_MAX_TIMEOUT_MS, STREAM_READINESS_TIMEOUT_MS, ANTIGRAVITY_PRE_RESPONSE_TIMEOUT_CODE, STREAM_RECOVERY, } from "../config/constants.ts"; import { createRecoverableStream, makeContinuationBody } from "../services/streamRecovery.ts"; -import { resolveResilienceSettings } from "@/lib/resilience/settings"; +import { + resolveResilienceSettings, + isStreamRecoveryExplicitlyConfigured, +} from "@/lib/resilience/settings"; import { classifyProviderError, PROVIDER_ERROR_TYPES, @@ -450,6 +454,13 @@ export async function handleChatCore({ if (pluginGate.body) { body = pluginGate.body; } + const agentGoalPolicy = resolveAgentGoalPolicy(body, clientRawRequest?.headers ?? null); + if (agentGoalPolicy.detected) { + log?.debug?.( + "AGENT_GOAL", + `long-running goal mode enabled: readinessMax=${agentGoalPolicy.readinessMaxTimeoutMs}ms streamRecovery=${agentGoalPolicy.streamRecoveryEnabled}` + ); + } let effectiveServiceTier: EffectiveServiceTier = "standard"; // Codex service-tier resolvers extracted to chatCore/serviceTier.ts (#3501); bind the per-request @@ -1143,8 +1154,7 @@ export async function handleChatCore({ } // Phase 4A: unified output styles (supersedes cavemanOutputMode via the back-compat shim). let outputStyleResult: - | import("../services/compression/outputStyles/apply.ts").OutputStylesResult - | null = null; + import("../services/compression/outputStyles/apply.ts").OutputStylesResult | null = null; if (config.enabled) { try { const { resolveOutputStyleSelection } = @@ -1196,8 +1206,8 @@ export async function handleChatCore({ ? ((compressionInputBody as Record).max_tokens as number) : null; let adaptiveTelemetry: - | import("../services/compression/adaptiveCompression/types.ts").AdaptiveTelemetry - | null = null; + import("../services/compression/adaptiveCompression/types.ts").AdaptiveTelemetry | null = + null; const compressionPlan = selectCompressionPlan( config, compressionComboKey, @@ -1876,6 +1886,16 @@ export async function handleChatCore({ } } + // Claude: strict Anthropic-compatible gateways (e.g. MiniMax) reject tool + // definitions that omit the required `type` discriminator with HTTP 400. Default + // a missing `type` to "custom" before dispatch, mirroring Anthropic's own + // inference, so legacy Claude-format tool payloads survive strict gateways (#2195). + if (targetFormat === FORMATS.CLAUDE && Array.isArray(translatedBody.tools)) { + translatedBody.tools = defaultClaudeToolType( + translatedBody.tools + ) as typeof translatedBody.tools; + } + // Extract toolNameMap for response translation (Claude OAuth) const translatedToolNameMap = translatedBody._toolNameMap; const nativeClaudeToolNameMap = isClaudePassthrough @@ -2116,8 +2136,7 @@ export async function handleChatCore({ let onPipelineStreamError: streamFailure.PipelineStreamErrorHandler | null = null; let onClientDisconnectFinalize: - | ((event: { reason: string; duration: number }) => boolean) - | null = null; + ((event: { reason: string; duration: number }) => boolean) | null = null; // Create stream controller for disconnect detection const streamController = createStreamController({ @@ -2442,8 +2461,21 @@ export async function handleChatCore({ // Reuse the request-consolidated settings read (see line ~2076) — no // second DB/cache hit. Default OFF when the setting is absent. const sr = resolveResilienceSettings(settings).streamRecovery; - streamRecoveryEnabled = sr.enabled; + // Fail-closed: the agent-goal-policy heuristic may only ADD recovery + // when the operator has no explicit configuration. If the operator + // explicitly configured stream recovery (env var or DB/settings + // override), that value always wins — the goal policy must never + // re-enable recovery the operator explicitly turned off. + const operatorExplicit = isStreamRecoveryExplicitlyConfigured(settings); + const goalOverride = !operatorExplicit && agentGoalPolicy.streamRecoveryEnabled; + streamRecoveryEnabled = sr.enabled || goalOverride; continueMidStreamEnabled = sr.continueMidStream === true; + if (goalOverride && !sr.enabled) { + log?.info?.( + "AGENT_GOAL", + `agentGoalPolicy override: stream recovery enabled for goal request requestId=${traceId} model=${modelToCall || model || requestedModel || "unknown"}` + ); + } } catch { streamRecoveryEnabled = false; continueMidStreamEnabled = false; @@ -2587,6 +2619,7 @@ export async function handleChatCore({ const headersObj = normalizeHeaders(rawResult.response.headers); const responseHeaders = new Headers(headersObj); stripStaleForwardingHeaders(responseHeaders); + stripNextMiddlewareControlHeaders(responseHeaders); const contentType = (responseHeaders.get("content-type") || "").toLowerCase(); const payload = await readNonStreamingResponseBody( rawResult.response, @@ -3916,6 +3949,9 @@ export async function handleChatCore({ provider, model, body: (finalBody || translatedBody) as Record | null | undefined, + maxTimeoutMs: agentGoalPolicy.detected + ? Math.max(STREAM_READINESS_MAX_TIMEOUT_MS, agentGoalPolicy.readinessMaxTimeoutMs) + : STREAM_READINESS_MAX_TIMEOUT_MS, }); if (streamReadinessPolicy.timeoutMs !== streamReadinessPolicy.baseTimeoutMs) { log?.debug?.( @@ -4038,6 +4074,20 @@ export async function handleChatCore({ } effectiveServiceTier = resolveReportedServiceTier(streamResponseBody) ?? effectiveServiceTier; + // Context Editing telemetry (streaming): the reconstructed stream body now carries + // context_management.applied_edits from the final message_delta snapshot. Mirror the + // non-streaming hook so streaming context-clear savings also surface under engine + // "context-editing" in compression analytics. Best-effort, Claude-only. + if (normalizedStreamStatus === 200) { + recordContextEditingTelemetryHook({ + contextEditingEnabled, + provider, + responseBody: streamResponseBody, + skillRequestId, + log, + }); + } + streamFailure.finalizeStreamRequestLog({ pendingRequestId, model, diff --git a/open-sse/handlers/chatCore/claudeToolDefaults.ts b/open-sse/handlers/chatCore/claudeToolDefaults.ts new file mode 100644 index 0000000000..bf758b6a5a --- /dev/null +++ b/open-sse/handlers/chatCore/claudeToolDefaults.ts @@ -0,0 +1,27 @@ +// Claude (Anthropic Messages) tool-definition normalization for outbound requests. + +type UnknownRecord = Record; + +/** + * Claude's tool schema requires every tool to carry an explicit `type` discriminator + * (e.g. "custom", "computer_20241022", "bash_20241022"). Anthropic's own API infers + * "custom" when it's omitted, but strict Anthropic-compatible gateways (e.g. MiniMax) + * enforce the documented schema and reject payloads whose tools lack `type` with + * HTTP 400. Default a missing `type` to "custom" so legacy Claude-format tool + * definitions survive strict gateways, while leaving any tool that already declares a + * type (incl. built-in tool types) untouched. (port from 9router#2195) + * + * Non-array input is returned unchanged; defaulted entries are new objects so the + * caller's original tool objects are not mutated. Non-object array entries (null, + * primitives, arrays) are passed through untouched rather than wrapped — spreading + * a primitive would fabricate a garbage tool (e.g. `{ type: "custom", '0': 'h' }`). + */ +export function defaultClaudeToolType(tools: unknown): unknown { + if (!Array.isArray(tools)) return tools; + return tools.map((tool) => { + if (tool && typeof tool === "object" && !Array.isArray(tool)) { + return (tool as UnknownRecord).type ? tool : { type: "custom", ...(tool as UnknownRecord) }; + } + return tool; + }); +} diff --git a/open-sse/handlers/chatCore/logTruncation.ts b/open-sse/handlers/chatCore/logTruncation.ts index 5f4171948b..e2a4b51c96 100644 --- a/open-sse/handlers/chatCore/logTruncation.ts +++ b/open-sse/handlers/chatCore/logTruncation.ts @@ -63,6 +63,14 @@ export function cloneBoundedChatLogPayload(value: unknown, depth = 0): unknown { * MAX_LOG_BODY_CHARS, return a lightweight summary instead of the full clone. * This prevents persistAttemptLogs from holding multi-MB references to * translatedBody across 17 call sites per request. + * + * When the summarized object carries a `tools` definition, re-attach it + * (bounded via `cloneBoundedChatLogPayload`) so the request-details view can + * still show which tools were available even though the rest of the payload + * — including the message history that triggered this summary — is dropped. + * The reused helper already caps array length, object keys, nesting depth, + * and string length, so this stays a small, bounded addition rather than a + * separately-budgeted multi-KB/MB re-clone. */ export function truncateForLog(value: unknown): Record | null | undefined { if (value === null || value === undefined) return value as null | undefined; @@ -80,5 +88,6 @@ export function truncateForLog(value: unknown): Record | null | if (Array.isArray(obj.messages)) summary.messageCount = obj.messages.length; if (Array.isArray(obj.contents)) summary.contentCount = obj.contents.length; if (typeof obj.stream === "boolean") summary.stream = obj.stream; + if (Array.isArray(obj.tools)) summary.tools = cloneBoundedChatLogPayload(obj.tools); return summary; } diff --git a/open-sse/handlers/chatCore/nonStreamingResponseBody.ts b/open-sse/handlers/chatCore/nonStreamingResponseBody.ts index c55c7c0261..da6e8c6acc 100644 --- a/open-sse/handlers/chatCore/nonStreamingResponseBody.ts +++ b/open-sse/handlers/chatCore/nonStreamingResponseBody.ts @@ -16,48 +16,135 @@ import { type NonStreamingSseTerminalState, } from "./nonStreamingSse.ts"; +/** + * Thrown when a non-streaming upstream body exceeds the hard cap. Buffering an unbounded + * SSE/NDJSON or JSON response in non-streaming mode was an OOM path (`rawBody += chunk` + * with no ceiling): a single multi-hundred-MB upstream response could fill the V8 heap. + * Callers treat this like any other upstream error rather than crashing the process. + */ +export class NonStreamingResponseTooLargeError extends Error { + readonly bytesSeen: number; + readonly maxBytes: number; + constructor(bytesSeen: number, maxBytes: number) { + super( + `Upstream non-streaming response exceeded the ${maxBytes}-byte cap (saw at least ${bytesSeen} bytes)` + ); + this.name = "NonStreamingResponseTooLargeError"; + this.bytesSeen = bytesSeen; + this.maxBytes = maxBytes; + } +} + +const DEFAULT_MAX_NONSTREAMING_RESPONSE_BYTES = 64 * 1024 * 1024; // 64 MB + +/** + * Hard cap for a non-streaming response buffered fully into memory. Generous by default so + * legitimate large completions pass; bounds only pathological/runaway upstream bodies. + * Override with `OMNIROUTE_MAX_NONSTREAMING_RESPONSE_BYTES`. + */ +export const MAX_NONSTREAMING_RESPONSE_BYTES = (() => { + const parsed = Number.parseInt(String(process.env.OMNIROUTE_MAX_NONSTREAMING_RESPONSE_BYTES), 10); + return Number.isFinite(parsed) && parsed > 0 ? parsed : DEFAULT_MAX_NONSTREAMING_RESPONSE_BYTES; +})(); + export async function readNonStreamingResponseBody( response: Response, contentType: string, - upstreamStream: boolean + upstreamStream: boolean, + maxBytes: number = MAX_NONSTREAMING_RESPONSE_BYTES ): Promise { if ( !upstreamStream || !response.body || (!contentType.includes("text/event-stream") && !contentType.includes("application/x-ndjson")) ) { + // Reject before buffering when the upstream declares an over-cap Content-Length. + const declared = Number.parseInt(response.headers.get("content-length") ?? "", 10); + if (Number.isFinite(declared) && declared > maxBytes) { + throw new NonStreamingResponseTooLargeError(declared, maxBytes); + } return withBodyTimeout(response.text()); } - const reader = response.body.getReader(); + return drainNonStreamingSseBody(response.body, maxBytes); +} + +/** + * Drain an SSE/NDJSON stream consumed in non-streaming mode into a single string, bounded + * by `maxBytes` (cancels the upstream and throws {@link NonStreamingResponseTooLargeError} + * past the cap) and by the body timeout, cancelling early on a terminal SSE signal. + */ +type NonStreamingChunk = { kind: "done" } | { kind: "skip" } | { kind: "chunk"; value: Uint8Array }; + +function cancelNonStreamingReader( + reader: ReadableStreamDefaultReader, + reason: unknown +): void { + try { + void reader.cancel(reason).catch(() => {}); + } catch { + // The caller is already unwinding or returning a complete terminal response. + } +} + +/** Read the next chunk under the body-timeout deadline, normalizing end/empty cases. */ +async function readNextNonStreamingChunk( + reader: ReadableStreamDefaultReader, + deadline: number +): Promise { + const timeoutMs = deadline > 0 ? deadline - Date.now() : 0; + if (deadline > 0 && timeoutMs <= 0) { + throw createBodyTimeoutError(FETCH_BODY_TIMEOUT_MS); + } + const { done, value } = await readStreamChunkWithTimeout(reader, timeoutMs); + if (done) return { kind: "done" }; + if (!value) return { kind: "skip" }; + return { kind: "chunk", value }; +} + +async function drainNonStreamingSseBody( + body: ReadableStream, + maxBytes: number +): Promise { + const reader = body.getReader(); const decoder = new TextDecoder(); const terminalState: NonStreamingSseTerminalState = { currentEvent: "", pendingLine: "", }; let rawBody = ""; + let bytesSeen = 0; const deadline = FETCH_BODY_TIMEOUT_MS > 0 ? Date.now() + FETCH_BODY_TIMEOUT_MS : 0; + let cancelRequested = false; + const requestCancel = (reason: unknown) => { + if (cancelRequested) return; + cancelRequested = true; + cancelNonStreamingReader(reader, reason); + }; try { while (true) { - const timeoutMs = deadline > 0 ? deadline - Date.now() : 0; - if (deadline > 0 && timeoutMs <= 0) { - throw createBodyTimeoutError(FETCH_BODY_TIMEOUT_MS); + const next = await readNextNonStreamingChunk(reader, deadline); + if (next.kind === "done") break; + if (next.kind === "skip") continue; + + // Bound the buffer: cancel the upstream and fail fast past the cap rather than + // growing `rawBody` until the V8 heap is exhausted. + bytesSeen += next.value.byteLength; + if (bytesSeen > maxBytes) { + requestCancel("non-streaming response exceeded byte cap"); + throw new NonStreamingResponseTooLargeError(bytesSeen, maxBytes); } - const { done, value } = await readStreamChunkWithTimeout(reader, timeoutMs); - if (done) break; - if (!value) continue; - - const decodedChunk = decoder.decode(value, { stream: true }); + const decodedChunk = decoder.decode(next.value, { stream: true }); rawBody += decodedChunk; if (appendNonStreamingSseTerminalSignal(terminalState, decodedChunk)) { - await reader.cancel("non-streaming bridge consumed terminal SSE event").catch(() => {}); + requestCancel("non-streaming bridge consumed terminal SSE event"); break; } } } catch (error) { - await reader.cancel(error).catch(() => {}); + requestCancel(error); throw error; } finally { rawBody += decoder.decode(); diff --git a/open-sse/handlers/chatCore/nonStreamingSse.ts b/open-sse/handlers/chatCore/nonStreamingSse.ts index 23e3da9517..5d7e642fc6 100644 --- a/open-sse/handlers/chatCore/nonStreamingSse.ts +++ b/open-sse/handlers/chatCore/nonStreamingSse.ts @@ -64,7 +64,9 @@ export function isTruthyStreamBody(body: unknown): boolean { return !!body && typeof body === "object" && (body as { stream?: unknown }).stream === true; } -export function isEventStreamAccepted(headers: Record | Headers | null | undefined) { +export function isEventStreamAccepted( + headers: Record | Headers | null | undefined +) { return (getHeaderValueCaseInsensitive(headers, "accept") || "") .toLowerCase() .includes("text/event-stream"); @@ -88,19 +90,32 @@ const NON_STREAMING_SSE_TERMINAL_TYPES = new Set([ "response.incomplete", ]); +function isNonStreamingSseTerminalType(eventType: string): boolean { + return NON_STREAMING_SSE_TERMINAL_TYPES.has(eventType); +} + export type NonStreamingSseTerminalState = { currentEvent: string; pendingLine: string; }; +function hasClaudeTerminalMessageDelta(parsed: unknown, eventType: string): boolean { + if (eventType !== "message_delta" || !parsed || typeof parsed !== "object") return false; + const delta = (parsed as { delta?: unknown }).delta; + if (!delta || typeof delta !== "object") return false; + const stopReason = (delta as { stop_reason?: unknown }).stop_reason; + return typeof stopReason === "string" ? stopReason.length > 0 : stopReason != null; +} + function processNonStreamingSseTerminalLine( state: NonStreamingSseTerminalState, rawLine: string ): boolean { const trimmed = rawLine.trim(); if (!trimmed || trimmed.startsWith(":")) { + const terminalEventOnly = !trimmed && isNonStreamingSseTerminalType(state.currentEvent); if (!trimmed) state.currentEvent = ""; - return false; + return terminalEventOnly; } if (trimmed.startsWith("event:")) { @@ -119,8 +134,11 @@ function processNonStreamingSseTerminalLine( // terminate with `[DONE]` (handled above), so parsing every one of them here is pure // waste that compounds into the CPU-runaway on large buffered responses. Skip the // JSON.parse unless the line could actually be a typed terminal. - if (!data.includes('"type"')) { - return NON_STREAMING_SSE_TERMINAL_TYPES.has(state.currentEvent); + if ( + !data.includes('"type"') && + !(state.currentEvent === "message_delta" && data.includes("stop_reason")) + ) { + return isNonStreamingSseTerminalType(state.currentEvent); } try { @@ -129,7 +147,9 @@ function processNonStreamingSseTerminalLine( parsed && typeof parsed === "object" && typeof parsed.type === "string" ? parsed.type : state.currentEvent; - return NON_STREAMING_SSE_TERMINAL_TYPES.has(eventType); + return ( + isNonStreamingSseTerminalType(eventType) || hasClaudeTerminalMessageDelta(parsed, eventType) + ); } catch { // Keep reading malformed data so the parser can report a useful upstream error. return false; diff --git a/open-sse/handlers/chatCore/responseHeaders.ts b/open-sse/handlers/chatCore/responseHeaders.ts index 3d490babc6..3c09538716 100644 --- a/open-sse/handlers/chatCore/responseHeaders.ts +++ b/open-sse/handlers/chatCore/responseHeaders.ts @@ -11,13 +11,61 @@ const STREAMING_RESPONSE_HEADER_DENYLIST = new Set([ "transfer-encoding", ]); +/** + * Prefix of Next.js internal middleware control headers. + * + * When an upstream provider is itself hosted behind a Next.js middleware + * (e.g. synthetic.new), a perfectly successful `200 OK` response can still + * carry Next's own control headers such as `x-middleware-rewrite`, + * `x-middleware-next`, `x-middleware-override-headers`, + * `x-middleware-set-cookie`, and the `x-middleware-request-*` family. + * + * OmniRoute forwards upstream response headers verbatim. If we re-emit those + * headers from an App Router route handler, Next 16's `app-route` runtime + * interprets `x-middleware-rewrite` as a `NextResponse.rewrite()` call and + * throws `NextResponse.rewrite() was used in a app route handler` — turning a + * successful upstream call into a 500. This is provider-agnostic proxy + * hygiene: any upstream behind Next middleware can leak these headers. + * + * See issue #5849. + */ +const NEXTJS_MIDDLEWARE_HEADER_PREFIX = "x-middleware-"; + +/** + * True when `headerName` is a Next.js internal middleware control header that + * must never be forwarded from a proxied upstream response. + */ +export function isNextMiddlewareControlHeader(headerName: string): boolean { + return headerName.toLowerCase().startsWith(NEXTJS_MIDDLEWARE_HEADER_PREFIX); +} + +/** + * Strip the whole `x-middleware-*` family (see {@link isNextMiddlewareControlHeader}) + * from a `Headers` instance. Used on the non-streaming JSON path alongside + * {@link stripStaleForwardingHeaders}. + */ +export function stripNextMiddlewareControlHeaders(headers: Headers): void { + const toDelete: string[] = []; + headers.forEach((_value, key) => { + if (isNextMiddlewareControlHeader(key)) { + toDelete.push(key); + } + }); + for (const key of toDelete) { + headers.delete(key); + } +} + export function buildStreamingResponseHeaders( providerHeaders: Headers, meta: Parameters[0] ): Record { const forwardedHeaders: [string, string][] = []; providerHeaders.forEach((value, key) => { - if (!STREAMING_RESPONSE_HEADER_DENYLIST.has(key.toLowerCase())) { + if ( + !STREAMING_RESPONSE_HEADER_DENYLIST.has(key.toLowerCase()) && + !isNextMiddlewareControlHeader(key) + ) { forwardedHeaders.push([key, value]); } }); diff --git a/open-sse/handlers/chatCore/upstreamBody.ts b/open-sse/handlers/chatCore/upstreamBody.ts index 51d6c74c11..1a302552d8 100644 --- a/open-sse/handlers/chatCore/upstreamBody.ts +++ b/open-sse/handlers/chatCore/upstreamBody.ts @@ -16,7 +16,6 @@ import { } from "../../services/payloadRules.ts"; import { getEffectiveToolLimit } from "../../services/toolLimitDetector.ts"; import { providerSupportsCaching } from "../../utils/cacheControlPolicy.ts"; -import { MAX_TOOLS_LIMIT } from "../../config/constants.ts"; import { FORMATS } from "../../translator/formats.ts"; type LoggerLike = { debug?: (...args: unknown[]) => void } | null | undefined; @@ -39,13 +38,13 @@ function buildAppliedRulesSummary( .join(", "); } -function truncateToolList(bodyToSend: Body, provider: string | null | undefined, log?: LoggerLike): Body { +function truncateToolList( + bodyToSend: Body, + provider: string | null | undefined, + log?: LoggerLike +): Body { const effectiveToolLimit = getEffectiveToolLimit(provider); - if ( - effectiveToolLimit < MAX_TOOLS_LIMIT && - Array.isArray(bodyToSend.tools) && - bodyToSend.tools.length > effectiveToolLimit - ) { + if (Array.isArray(bodyToSend.tools) && bodyToSend.tools.length > effectiveToolLimit) { const truncatedTools = bodyToSend.tools.slice(0, effectiveToolLimit); bodyToSend = { ...bodyToSend, tools: truncatedTools }; log?.debug?.( @@ -64,8 +63,7 @@ function backfillQwenOAuthUser( credentials: CredentialsLike, log?: LoggerLike ): Body { - const hasValidQwenUser = - typeof bodyToSend.user === "string" && bodyToSend.user.trim().length > 0; + const hasValidQwenUser = typeof bodyToSend.user === "string" && bodyToSend.user.trim().length > 0; const isQwenOAuthRequest = provider === "qwen" && !credentials?.apiKey && diff --git a/open-sse/handlers/responseSanitizer.ts b/open-sse/handlers/responseSanitizer.ts index aaa216569b..04d25871a6 100644 --- a/open-sse/handlers/responseSanitizer.ts +++ b/open-sse/handlers/responseSanitizer.ts @@ -3,6 +3,14 @@ import { getReadableReasoningValue, } from "../utils/reasoningFields.ts"; import { normalizeOpenAICompatibleFinishReason } from "../utils/finishReason.ts"; +import { + collapseExcessiveNewlines, + extractThinkingFromContent, +} from "./responseSanitizer/reasoning.ts"; +export { + extractThinkingFromContent, + shouldParseTextualReasoningTags, +} from "./responseSanitizer/reasoning.ts"; /** * Response Sanitizer — Normalizes LLM responses to strict OpenAI SDK format. @@ -191,144 +199,6 @@ function containsTextualToolCallContent(content: unknown): boolean { ); } -const REASONING_TAG_NAMES = ["think", "thinking", "thought", "internal_thought"]; -const REASONING_TAG_PATTERN = REASONING_TAG_NAMES.join("|"); -// Matches complete /// blocks. -const THINK_TAG_REGEX = new RegExp( - `<(${REASONING_TAG_PATTERN})\\b[^>]*>([\\s\\S]*?)<\\/\\1>`, - "gi" -); -const REASONING_CLOSE_TAG_REGEX = new RegExp(``, "i"); -const REASONING_TAG_FRAGMENT_REGEX = new RegExp(`]*>`, "gi"); -const CONTENT_OPEN_TAG_REGEX = /]*>/i; -// Matches an unclosed reasoning tag at the end of a message. Some providers can -// emit malformed/open reasoning wrappers (for example "]*)?(?:>|\\r?\\n)([\\s\\S]*)$`, - "i" -); - -// #638, #727: Collapse runs of 2+ consecutive newlines into \n\n -// Tool call responses from thinking models often accumulate excessive newlines -const EXCESSIVE_NEWLINES = /\n{2,}/g; -function collapseExcessiveNewlines(text: string): string { - return text.replace(EXCESSIVE_NEWLINES, "\n\n"); -} - -function cleanReasoningFragment(text: string): string { - return text.replace(REASONING_TAG_FRAGMENT_REGEX, "").trim(); -} - -function splitClosingOnlyReasoningPrefix(text: string): { - content: string; - thinking: string | null; -} | null { - const closeMatch = text.match(REASONING_CLOSE_TAG_REGEX); - if (!closeMatch || closeMatch.index === undefined || closeMatch.index === 0) return null; - const closeIndex = closeMatch.index; - - const suffix = text.slice(closeIndex + closeMatch[0].length); - if (!CONTENT_OPEN_TAG_REGEX.test(suffix)) return null; - - const thinking = cleanReasoningFragment(text.slice(0, closeIndex)); - if (!thinking) return null; - return { content: suffix.trim(), thinking }; -} - -function movePrefixBeforeContentTagToThinking(cleaned: string, thinkingParts: string[]): string { - const contentMatch = cleaned.match(CONTENT_OPEN_TAG_REGEX); - if (!contentMatch || contentMatch.index === undefined || contentMatch.index <= 0) return cleaned; - const contentIndex = contentMatch.index; - - const prefix = cleanReasoningFragment(cleaned.slice(0, contentIndex)); - if (prefix) thinkingParts.unshift(prefix); - return cleaned.slice(contentIndex); -} - -/** - * Extract blocks from text content and return separated parts. - * @returns {{ content: string, thinking: string | null }} - */ -export function extractThinkingFromContent(text: string): { - content: string; - thinking: string | null; -} { - if (!text || typeof text !== "string") { - return { content: text || "", thinking: null }; - } - - const thinkingParts: string[] = []; - let hasThinkTags = false; - - let cleaned = text.replace(THINK_TAG_REGEX, (_match, _tagName, thinkContent) => { - hasThinkTags = true; - const trimmed = thinkContent.trim(); - if (trimmed) { - thinkingParts.push(trimmed); - } - return ""; - }); - - if (!hasThinkTags) { - const closingOnly = splitClosingOnlyReasoningPrefix(cleaned); - if (closingOnly) { - return closingOnly; - } - } - - const unclosedMatch = cleaned.match(UNCLOSED_REASONING_TAG_REGEX); - if (unclosedMatch?.index !== undefined) { - hasThinkTags = true; - const reasoning = String(unclosedMatch[2] || "").trim(); - if (reasoning) thinkingParts.push(reasoning); - const prefix = cleaned.slice(0, unclosedMatch.index); - cleaned = /^(?:\s|§\d+§)*$/.test(prefix) ? "" : prefix; - } - - if (!hasThinkTags) { - return { content: text, thinking: null }; - } - - cleaned = movePrefixBeforeContentTagToThinking(cleaned, thinkingParts); - - return { - content: cleaned.trim(), - thinking: thinkingParts.length > 0 ? thinkingParts.join("\n\n") : null, - }; -} - -function normalizeReasoningRouteId(value: unknown): string { - return typeof value === "string" ? value.toLowerCase() : ""; -} - -function isAntigravityReasoningRoute(providerId: string, modelId: string): boolean { - return ( - providerId.includes("antigravity") || - providerId === "agy" || - modelId.includes("antigravity/") || - modelId.startsWith("agy/") - ); -} - -function isTextualReasoningTagNativeRoute(providerId: string, modelId: string): boolean { - const routeId = `${providerId}/${modelId}`; - return ( - /deepseek[-_/]?r1\b/.test(routeId) || - /r1[-_/]?distill\b/.test(routeId) || - /(?:^|[/:_-])qwq(?:[/._:-]|$)/.test(routeId) - ); -} - -export function shouldParseTextualReasoningTags(provider?: unknown, model?: unknown): boolean { - const providerId = normalizeReasoningRouteId(provider); - const modelId = normalizeReasoningRouteId(model); - return ( - !isAntigravityReasoningRoute(providerId, modelId) && - isTextualReasoningTagNativeRoute(providerId, modelId) - ); -} - /** * Sanitize a non-streaming OpenAI ChatCompletion response. * Strips non-standard fields and normalizes required fields. @@ -724,7 +594,7 @@ function sanitizeResponsesStreamingOutputItem(item: unknown): JsonRecord | null return { ...partRecord, type: toString(partRecord.type) || "summary_text", - text: collapseExcessiveNewlines(toString(partRecord.text) || ""), + text: collapseExcessiveNewlines(stripZeroWidthText(toString(partRecord.text) || "")), }; }) .filter((part) => part !== null) @@ -754,7 +624,7 @@ function sanitizeResponsesStreamingOutputItem(item: unknown): JsonRecord | null type: "function_call_output", output: typeof itemRecord.output === "string" - ? collapseExcessiveNewlines(itemRecord.output) + ? collapseExcessiveNewlines(stripZeroWidthText(itemRecord.output)) : JSON.stringify(itemRecord.output ?? ""), }; } @@ -770,10 +640,39 @@ function sanitizeResponsesStreamingOutput(output: unknown): JsonRecord[] { .filter((item): item is JsonRecord => item !== null); } +// Native Responses streaming events that carry raw model text directly on the +// root `delta` field. These must get the same zero-width-joiner stripping as the +// non-streaming path so agent words (opencode/cursor/aider) are not corrupted. +// Scoped as an allow-list on purpose: `response.function_call_arguments.delta` +// carries tool-call argument JSON that must pass through byte-exact. +const RESPONSES_STREAMING_TEXT_DELTA_EVENTS = new Set([ + "response.output_text.delta", + "response.reasoning_summary_text.delta", + "response.reasoning_text.delta", +]); + +// Matching `*.done` events that carry the finalized text on the root `text` field. +const RESPONSES_STREAMING_TEXT_DONE_EVENTS = new Set([ + "response.output_text.done", + "response.reasoning_summary_text.done", + "response.reasoning_text.done", +]); + function sanitizeResponsesStreamingEvent(parsedRecord: JsonRecord): JsonRecord { const sanitized: JsonRecord = { ...parsedRecord }; const eventType = toString(parsedRecord.type) || ""; + // Root-level text events (output_text / reasoning_summary_text / reasoning_text) + // carry the model text directly on the event, not under item/output. Strip ZWJ + // there too. Only touch string values and only the allow-listed event types — + // never function-call argument events. + if (RESPONSES_STREAMING_TEXT_DELTA_EVENTS.has(eventType) && typeof sanitized.delta === "string") { + sanitized.delta = stripZeroWidthText(sanitized.delta); + } + if (RESPONSES_STREAMING_TEXT_DONE_EVENTS.has(eventType) && typeof sanitized.text === "string") { + sanitized.text = stripZeroWidthText(sanitized.text); + } + if (parsedRecord.item !== undefined) { const sanitizedItem = sanitizeResponsesStreamingOutputItem(parsedRecord.item); if (sanitizedItem) { @@ -1087,10 +986,18 @@ export function sanitizeStreamingChunk(parsed: unknown): unknown { if (deltaRecord.content !== undefined) { delta.content = typeof deltaRecord.content === "string" - ? collapseExcessiveNewlines(deltaRecord.content) + ? collapseExcessiveNewlines(stripZeroWidthText(deltaRecord.content)) : deltaRecord.content; } copyOpenAICompatibleReasoningFields(deltaRecord, delta); + // Parity with the non-streaming path: strip the zero-width joiners that the + // request side injects into agent words. copyOpenAICompatibleReasoningFields + // is shared, so strip locally on the fields it writes (string values only). + for (const reasoningKey of ["reasoning_content", "reasoning", "reasoning_text"]) { + if (typeof delta[reasoningKey] === "string") { + delta[reasoningKey] = stripZeroWidthText(delta[reasoningKey] as string); + } + } if (deltaRecord.tool_calls !== undefined) { delta.tool_calls = Array.isArray(deltaRecord.tool_calls) ? deltaRecord.tool_calls.map((tc) => { diff --git a/open-sse/handlers/responseSanitizer/reasoning.ts b/open-sse/handlers/responseSanitizer/reasoning.ts new file mode 100644 index 0000000000..cf15fc2fc6 --- /dev/null +++ b/open-sse/handlers/responseSanitizer/reasoning.ts @@ -0,0 +1,143 @@ +export const REASONING_TAG_NAMES = ["think", "thinking", "thought", "internal_thought"]; +export const REASONING_TAG_PATTERN = REASONING_TAG_NAMES.join("|"); +// Matches complete /// blocks. +export const THINK_TAG_REGEX = new RegExp( + `<(${REASONING_TAG_PATTERN})\\b[^>]*>([\\s\\S]*?)<\\/\\1>`, + "gi" +); +export const REASONING_CLOSE_TAG_REGEX = new RegExp(``, "i"); +export const REASONING_TAG_FRAGMENT_REGEX = new RegExp( + `]*>`, + "gi" +); +export const CONTENT_OPEN_TAG_REGEX = /]*>/i; +// Matches an unclosed reasoning tag at the end of a message. Some providers can +// emit malformed/open reasoning wrappers (for example "]*)?(?:>|\\r?\\n)([\\s\\S]*)$`, + "i" +); + +// #638, #727: Collapse runs of 2+ consecutive newlines into \n\n +// Tool call responses from thinking models often accumulate excessive newlines +export const EXCESSIVE_NEWLINES = /\n{2,}/g; +export function collapseExcessiveNewlines(text: string): string { + return text.replace(EXCESSIVE_NEWLINES, "\n\n"); +} + +export function cleanReasoningFragment(text: string): string { + return text.replace(REASONING_TAG_FRAGMENT_REGEX, "").trim(); +} + +export function splitClosingOnlyReasoningPrefix(text: string): { + content: string; + thinking: string | null; +} | null { + const closeMatch = text.match(REASONING_CLOSE_TAG_REGEX); + if (!closeMatch || closeMatch.index === undefined || closeMatch.index === 0) return null; + const closeIndex = closeMatch.index; + + const suffix = text.slice(closeIndex + closeMatch[0].length); + if (!CONTENT_OPEN_TAG_REGEX.test(suffix)) return null; + + const thinking = cleanReasoningFragment(text.slice(0, closeIndex)); + if (!thinking) return null; + return { content: suffix.trim(), thinking }; +} + +export function movePrefixBeforeContentTagToThinking( + cleaned: string, + thinkingParts: string[] +): string { + const contentMatch = cleaned.match(CONTENT_OPEN_TAG_REGEX); + if (!contentMatch || contentMatch.index === undefined || contentMatch.index <= 0) return cleaned; + const contentIndex = contentMatch.index; + + const prefix = cleanReasoningFragment(cleaned.slice(0, contentIndex)); + if (prefix) thinkingParts.unshift(prefix); + return cleaned.slice(contentIndex); +} + +/** + * Extract blocks from text content and return separated parts. + * @returns {{ content: string, thinking: string | null }} + */ +export function extractThinkingFromContent(text: string): { + content: string; + thinking: string | null; +} { + if (!text || typeof text !== "string") { + return { content: text || "", thinking: null }; + } + + const thinkingParts: string[] = []; + let hasThinkTags = false; + + let cleaned = text.replace(THINK_TAG_REGEX, (_match, _tagName, thinkContent) => { + hasThinkTags = true; + const trimmed = thinkContent.trim(); + if (trimmed) { + thinkingParts.push(trimmed); + } + return ""; + }); + + if (!hasThinkTags) { + const closingOnly = splitClosingOnlyReasoningPrefix(cleaned); + if (closingOnly) { + return closingOnly; + } + } + + const unclosedMatch = cleaned.match(UNCLOSED_REASONING_TAG_REGEX); + if (unclosedMatch?.index !== undefined) { + hasThinkTags = true; + const reasoning = String(unclosedMatch[2] || "").trim(); + if (reasoning) thinkingParts.push(reasoning); + const prefix = cleaned.slice(0, unclosedMatch.index); + cleaned = /^(?:\s|§\d+§)*$/.test(prefix) ? "" : prefix; + } + + if (!hasThinkTags) { + return { content: text, thinking: null }; + } + + cleaned = movePrefixBeforeContentTagToThinking(cleaned, thinkingParts); + + return { + content: cleaned.trim(), + thinking: thinkingParts.length > 0 ? thinkingParts.join("\n\n") : null, + }; +} + +export function normalizeReasoningRouteId(value: unknown): string { + return typeof value === "string" ? value.toLowerCase() : ""; +} + +export function isAntigravityReasoningRoute(providerId: string, modelId: string): boolean { + return ( + providerId.includes("antigravity") || + providerId === "agy" || + modelId.includes("antigravity/") || + modelId.startsWith("agy/") + ); +} + +export function isTextualReasoningTagNativeRoute(providerId: string, modelId: string): boolean { + const routeId = `${providerId}/${modelId}`; + return ( + /deepseek[-_/]?r1\b/.test(routeId) || + /r1[-_/]?distill\b/.test(routeId) || + /(?:^|[/:_-])qwq(?:[/._:-]|$)/.test(routeId) + ); +} + +export function shouldParseTextualReasoningTags(provider?: unknown, model?: unknown): boolean { + const providerId = normalizeReasoningRouteId(provider); + const modelId = normalizeReasoningRouteId(model); + return ( + !isAntigravityReasoningRoute(providerId, modelId) && + isTextualReasoningTagNativeRoute(providerId, modelId) + ); +} diff --git a/open-sse/handlers/sseParser.ts b/open-sse/handlers/sseParser.ts index 94a14e810e..d2e634e12a 100644 --- a/open-sse/handlers/sseParser.ts +++ b/open-sse/handlers/sseParser.ts @@ -370,8 +370,7 @@ export function parseSSEToClaudeResponse(rawSSE, fallbackModel) { type: "thinking", index, thinking: toString(contentBlock.thinking), - signature: - typeof contentBlock.signature === "string" ? contentBlock.signature : undefined, + signature: toString(contentBlock.signature) || undefined, }); } else if (blockType === "tool_use") { blocks.set(index, { @@ -416,12 +415,17 @@ export function parseSSEToClaudeResponse(rawSSE, fallbackModel) { continue; } - if (deltaType === "thinking_delta" || typeof delta.thinking === "string") { + const isThinkingDelta = deltaType === "thinking_delta" || typeof delta.thinking === "string"; + const isSignatureDelta = + deltaType === "signature_delta" || typeof delta.signature === "string"; + if (isThinkingDelta || isSignatureDelta) { const thinking = existing && existing.type === "thinking" ? existing : { type: "thinking", index, thinking: "", signature: undefined }; - thinking.thinking += toString(delta.thinking); + if (isThinkingDelta) thinking.thinking += toString(delta.thinking); + const signature = toString(delta.signature); + if (signature) thinking.signature = `${thinking.signature || ""}${signature}`; blocks.set(index, thinking); continue; } @@ -454,35 +458,27 @@ export function parseSSEToClaudeResponse(rawSSE, fallbackModel) { if (!sawClaudeEvent) return null; - const content = [...blocks.values()] - .sort((a, b) => a.index - b.index) - .flatMap((block) => { - if (block.type === "text") { - return block.text ? [{ type: "text", text: block.text }] : []; - } - if (block.type === "thinking") { - return block.thinking - ? [ - { - type: "thinking", - thinking: block.thinking, - ...(block.signature ? { signature: block.signature } : {}), - }, - ] - : []; + const content = []; + for (const block of [...blocks.values()].sort((a, b) => a.index - b.index)) { + if (block.type === "text") { + if (block.text) content.push({ type: "text", text: block.text }); + continue; + } + if (block.type === "thinking") { + const hasSignature = typeof block.signature === "string" && block.signature.length > 0; + if (block.thinking || hasSignature) { + content.push({ + type: "thinking", + thinking: block.thinking || "", + ...(hasSignature ? { signature: block.signature } : {}), + }); } + continue; + } - const parsedInput = - block.inputJson.trim().length > 0 ? tryParseJson(block.inputJson) : block.input; - return [ - { - type: "tool_use", - id: block.id, - name: block.name, - input: parsedInput, - }, - ]; - }); + const input = block.inputJson.trim().length > 0 ? tryParseJson(block.inputJson) : block.input; + content.push({ type: "tool_use", id: block.id, name: block.name, input }); + } return { id: messageId || `msg_${Date.now()}`, diff --git a/open-sse/mcp-server/__tests__/httpAuthContext.test.ts b/open-sse/mcp-server/__tests__/httpAuthContext.test.ts index b90a7b30d6..db451bd2d4 100644 --- a/open-sse/mcp-server/__tests__/httpAuthContext.test.ts +++ b/open-sse/mcp-server/__tests__/httpAuthContext.test.ts @@ -108,6 +108,44 @@ describe("MCP HTTP auth context", () => { } }); + it("forwarded caller auth wins over the OMNIROUTE_API_KEY env fallback (#5819)", async () => { + const fetchMock = vi.fn().mockResolvedValue({ + ok: true, + json: async () => ({ combos: [] }), + }); + vi.stubGlobal("fetch", fetchMock); + vi.stubEnv("OMNIROUTE_API_KEY", "env-fallback-key"); + + const [clientTransport, serverTransport] = InMemoryTransport.createLinkedPair(); + const server = createMcpServer(); + await server.connect(serverTransport); + const client = new Client({ name: "test-client", version: "1.0.0" }); + await client.connect(clientTransport); + + try { + const request = new Request("http://localhost/api/mcp/stream", { + headers: { Authorization: "Bearer manage-key" }, + }); + await withMcpHttpAuthContext(request, () => + client.callTool({ name: "omniroute_list_combos", arguments: {} }) + ); + + // The per-caller forwarded identity must win over the static env fallback. + expect(fetchMock).toHaveBeenCalledWith( + expect.stringContaining("/api/combos"), + expect.objectContaining({ + headers: expect.objectContaining({ Authorization: "Bearer manage-key" }), + }) + ); + const sentHeaders = fetchMock.mock.calls[0][1].headers as Record; + expect(sentHeaders.Authorization).not.toBe("Bearer env-fallback-key"); + } finally { + await client.close(); + vi.unstubAllEnvs(); + vi.unstubAllGlobals(); + } + }); + it("forwards request auth through advanced tool apiFetch", async () => { const fetchMock = vi.fn().mockResolvedValue({ ok: true, diff --git a/open-sse/mcp-server/mcpCallerIdentity.ts b/open-sse/mcp-server/mcpCallerIdentity.ts new file mode 100644 index 0000000000..7d24e370b9 --- /dev/null +++ b/open-sse/mcp-server/mcpCallerIdentity.ts @@ -0,0 +1,53 @@ +/** + * #5649 — resolve the MCP caller's API-key **principal id** for content stores + * (CCR) that are keyed by principal. + * + * The CCR store keys blocks by `String(apiKeyInfo.id)` at compression time + * (`chatCore` → `apiKeyInfo = getApiKeyMetadata(rawKey)`). MCP tool retrieval must + * resolve the SAME id or the block is not found. On the MCP HTTP transports + * (SSE / Streamable HTTP) the raw key lives in `httpAuthContext`'s + * AsyncLocalStorage (set by `withMcpHttpAuthContext`), NOT in the tool handler's + * `extra.authInfo` (OmniRoute authenticates with API keys, not OAuth client ids — + * so `extra.authInfo.clientId` is never populated and the caller resolved to + * "anonymous", producing a cross-principal store-key miss). + * + * Resolving through the same `getApiKeyMetadata` lookup keeps cross-tenant IDOR + * isolation intact: a different key → a different id → a miss; no key → undefined + * → the anonymous (`__anon__`) bucket, which only matches unauthenticated stores. + */ +import { getMcpHttpAuthHeadersForInternalFetch } from "./httpAuthContext.ts"; +import { extractApiKey } from "../../src/sse/services/auth.ts"; +import { getApiKeyMetadata } from "../../src/lib/db/apiKeys.ts"; + +type ApiKeyLookup = (rawKey: string) => Promise<{ id?: string | number | null } | null>; + +/** + * Pure resolver: given the request auth headers and a key→metadata lookup, return + * the principal id (as a string) or `undefined`. Separated from the AsyncLocalStorage + * read so it is unit-testable without a live transport or DB. + */ +export async function resolvePrincipalFromHeaders( + headers: Record, + lookup: ApiKeyLookup = getApiKeyMetadata +): Promise { + // Nothing to resolve without an Authorization / x-api-key header. + if (!headers.Authorization && !headers["x-api-key"]) return undefined; + const rawKey = extractApiKey({ headers: new Headers(headers) }, { allowUrl: false }); + if (!rawKey) return undefined; + try { + const meta = await lookup(rawKey); + return meta?.id != null && meta.id !== "" ? String(meta.id) : undefined; + } catch { + // Fail closed: an unresolved principal can only reach the anonymous bucket. + return undefined; + } +} + +/** + * Resolve the current MCP HTTP caller's API-key principal id from the ambient + * `httpAuthContext`. Returns `undefined` off the HTTP transport (stdio) or when the + * request carries no API key. + */ +export function resolveMcpCallerApiKeyId(): Promise { + return resolvePrincipalFromHeaders(getMcpHttpAuthHeadersForInternalFetch()); +} diff --git a/open-sse/mcp-server/server.ts b/open-sse/mcp-server/server.ts index bc0d31a8d1..554f1ab1d8 100644 --- a/open-sse/mcp-server/server.ts +++ b/open-sse/mcp-server/server.ts @@ -215,8 +215,10 @@ async function omniRouteFetch(path: string, options: RequestInit = {}): Promise< const apiKey = getOmniRouteApiKey(); const headers: Record = { "Content-Type": "application/json", - ...getMcpHttpAuthHeadersForInternalFetch(), + // Static env key is only a fallback; the per-caller MCP identity forwarded via + // withMcpHttpAuthContext must win over it (#5819). ...(apiKey ? { Authorization: `Bearer ${apiKey}` } : {}), + ...getMcpHttpAuthHeadersForInternalFetch(), ...((options.headers as Record) || {}), }; diff --git a/open-sse/mcp-server/tools/advancedTools.ts b/open-sse/mcp-server/tools/advancedTools.ts index 0efa428126..e9fbdc22b7 100644 --- a/open-sse/mcp-server/tools/advancedTools.ts +++ b/open-sse/mcp-server/tools/advancedTools.ts @@ -38,8 +38,10 @@ async function apiFetch(path: string, options: RequestInit = {}): Promise = { "Content-Type": "application/json", - ...getMcpHttpAuthHeadersForInternalFetch(), + // Static env key is only a fallback; the per-caller MCP identity forwarded via + // withMcpHttpAuthContext must win over it (#5819). ...(OMNIROUTE_API_KEY ? { Authorization: `Bearer ${OMNIROUTE_API_KEY}` } : {}), + ...getMcpHttpAuthHeadersForInternalFetch(), ...((options.headers as Record) || {}), }; const response = await fetch(url, { ...options, headers, signal: AbortSignal.timeout(30000) }); diff --git a/open-sse/mcp-server/tools/compressionTools.ts b/open-sse/mcp-server/tools/compressionTools.ts index 22ff308855..4d1df0f459 100644 --- a/open-sse/mcp-server/tools/compressionTools.ts +++ b/open-sse/mcp-server/tools/compressionTools.ts @@ -243,7 +243,14 @@ import { compressionComboStatsInput, } from "../schemas/tools.ts"; import { handleCcrRetrieve } from "../../services/compression/engines/ccr/index.ts"; +import { + listRtkCommandSamples, + discoverRepeatedNoise, + suggestFilter, + commandToId, +} from "../../services/compression/engines/rtk/index.ts"; import { resolveCallerScopeContext } from "../scopeEnforcement.ts"; +import { resolveMcpCallerApiKeyId } from "../mcpCallerIdentity.ts"; const ccrRetrieveInput = z.object({ hash: z @@ -313,6 +320,48 @@ export async function handleCompressionComboStats( }; } +// T07 — RTK learn/discover exposed via MCP (read-only; suggestions only). Mines the opt-in +// raw-output sample store, exactly like the /api/context/rtk/{discover,learn} routes. +const rtkDiscoverInput = z.object({ + limit: z.number().int().positive().max(2000).optional().describe("Max samples to scan (default 500)"), +}); + +const rtkLearnInput = z.object({ + command: z.string().min(1).max(500).describe("The command to learn an RTK filter draft for"), + limit: z.number().int().positive().max(2000).optional().describe("Max samples to scan (default 500)"), +}); + +function resolveSampleLimit(limit?: number): number { + if (!Number.isFinite(limit) || !limit || limit <= 0) return 500; + return Math.min(2000, Math.floor(limit)); +} + +export async function handleRtkDiscover( + args: z.infer +): Promise<{ sampleCount: number; candidates: ReturnType }> { + const start = Date.now(); + const samples = listRtkCommandSamples({ limit: resolveSampleLimit(args.limit) }); + const candidates = discoverRepeatedNoise(samples); + const result = { sampleCount: samples.length, candidates }; + await logToolCall("omniroute_rtk_discover", args, result, Date.now() - start, true); + return result; +} + +export async function handleRtkLearn( + args: z.infer +): Promise<{ command: string; sampleCount: number; filter: ReturnType }> { + const start = Date.now(); + const command = args.command.trim(); + const targetId = commandToId(command); + const matching = listRtkCommandSamples({ limit: resolveSampleLimit(args.limit) }).filter( + (sample) => commandToId(sample.command) === targetId + ); + const filter = suggestFilter(command, matching); + const result = { command, sampleCount: matching.length, filter }; + await logToolCall("omniroute_rtk_learn", args, result, Date.now() - start, true); + return result; +} + export const compressionTools = { omniroute_compression_status: { name: "omniroute_compression_status", @@ -362,11 +411,41 @@ export const compressionTools = { "Scope: read:compression. Always available (sticky-on).", scopes: ["read:compression"], inputSchema: ccrRetrieveInput, - handler: (args: z.infer, extra?: McpToolExtraLike) => { - // Derive caller identity from MCP auth context so the retrieve is scoped to the - // same principal that stored the block. This closes the cross-tenant IDOR (HIGH). + handler: async (args: z.infer, extra?: McpToolExtraLike) => { + // Retrieve must use the SAME principal the CCR store used at compression time: + // `String(apiKeyInfo.id)` (chatCore → getApiKeyMetadata(rawKey)). On MCP HTTP + // transports the raw key lives in httpAuthContext (not in extra.authInfo, since + // OmniRoute auth is API-key not OAuth-clientId) — resolve it to the same key id + // so the block is found. Without this the caller resolved to "anonymous" and the + // store-key never matched (#5649). Cross-tenant IDOR stays closed: a different + // key → different id → miss; no key → undefined → anonymous bucket only. + const apiKeyPrincipal = await resolveMcpCallerApiKeyId(); + if (apiKeyPrincipal) { + return handleCcrRetrieve(args, apiKeyPrincipal); + } + // Fallback (unchanged): OAuth clientId / session scope context, then anonymous. const { callerId } = resolveCallerScopeContext(extra, ["read:compression"]); return handleCcrRetrieve(args, callerId === "anonymous" ? undefined : callerId); }, }, + omniroute_rtk_discover: { + name: "omniroute_rtk_discover", + description: + "Mine the opt-in RTK raw-output sample store for recurring noise lines and return them " + + "as ranked candidates the operator can turn into strip/collapse filters. Read-only; " + + "suggestions only. Scope: read:compression.", + scopes: ["read:compression"], + inputSchema: rtkDiscoverInput, + handler: (args: z.infer) => handleRtkDiscover(args), + }, + omniroute_rtk_learn: { + name: "omniroute_rtk_learn", + description: + "Suggest an RTK filter draft for a specific command, learned from that command's captured " + + "outputs in the opt-in raw-output sample store. Read-only; returns a draft for the operator " + + "to review and save. Scope: read:compression.", + scopes: ["read:compression"], + inputSchema: rtkLearnInput, + handler: (args: z.infer) => handleRtkLearn(args), + }, }; diff --git a/open-sse/package.json b/open-sse/package.json index c301e08bc7..1b03c6aeaf 100644 --- a/open-sse/package.json +++ b/open-sse/package.json @@ -1,6 +1,6 @@ { "name": "@omniroute/open-sse", - "version": "3.8.42", + "version": "3.8.43", "description": "Express SSE sidecar for OmniRoute — handles streaming, protocol translation, and provider orchestration", "type": "module", "main": "index.js", diff --git a/open-sse/services/accountFallback.ts b/open-sse/services/accountFallback.ts index 4139845c2e..0efddc1762 100644 --- a/open-sse/services/accountFallback.ts +++ b/open-sse/services/accountFallback.ts @@ -1296,6 +1296,7 @@ export function checkFallbackError( * provider itself). Callers should apply connection cooldown only — do NOT record a provider * circuit-breaker failure when this flag is set. */ skipProviderBreaker?: boolean; + quotaResetHintMs?: number; } { // G-02: detect embedded service supervisor failures (X-Omni-Fallback-Hint: connection_cooldown). // These are NOT upstream AI provider failures — they are local supervisor state changes. @@ -1478,11 +1479,13 @@ export function checkFallbackError( // profile.useUpstreamRetryHints. const hintMs = getUpstreamRetryHintMs(); const SUBSCRIPTION_QUOTA_COOLDOWN_MS = 60 * 60 * 1000; // 1 hour + const bodyHint = parseRetryFromErrorText(errorStr); return { shouldFallback: true, cooldownMs: hintMs ?? SUBSCRIPTION_QUOTA_COOLDOWN_MS, reason: RateLimitReason.QUOTA_EXHAUSTED, usedUpstreamRetryHint: Boolean(hintMs), + quotaResetHintMs: bodyHint ?? undefined, }; } @@ -1492,7 +1495,11 @@ export function checkFallbackError( quotaResetHintMs && classifyErrorText(errorStr) === RateLimitReason.QUOTA_EXHAUSTED ) { - return buildRetryableFallback(RateLimitReason.QUOTA_EXHAUSTED); + const fallbackResult = buildRetryableFallback(RateLimitReason.QUOTA_EXHAUSTED); + return { + ...fallbackResult, + quotaResetHintMs, + }; } // #2929: A route-restriction 403 (e.g. Fireworks Fire Pass keys returning diff --git a/open-sse/services/antigravity429Engine.ts b/open-sse/services/antigravity429Engine.ts index 909ed8adff..a29b0dba2e 100644 --- a/open-sse/services/antigravity429Engine.ts +++ b/open-sse/services/antigravity429Engine.ts @@ -21,10 +21,7 @@ export type Category = "unknown" | "rate_limited" | "quota_exhausted" | "soft_rate_limit"; export type DecisionKind = - | "soft_retry" - | "instant_retry_same_auth" - | "short_cooldown_switch_auth" - | "full_quota_exhausted"; + "soft_retry" | "instant_retry_same_auth" | "short_cooldown_switch_auth" | "full_quota_exhausted"; export interface Decision { kind: DecisionKind; @@ -53,6 +50,8 @@ const CREDITS_EXHAUSTED_KEYWORDS = [ "minimumcreditamountforusage", "minimum credit amount for usage", "minimum credit", + "insufficient_g1_credits_balance", + "g1_credits", ]; const SHORT_COOLDOWN_MS = 5 * 60 * 1000; // 5 minutes diff --git a/open-sse/services/antigravityProjectBootstrap.ts b/open-sse/services/antigravityProjectBootstrap.ts index d9d4cdb58c..bd9fb91617 100644 --- a/open-sse/services/antigravityProjectBootstrap.ts +++ b/open-sse/services/antigravityProjectBootstrap.ts @@ -10,8 +10,7 @@ * attempt. Results are memoized per-token for the process lifetime to * avoid redundant round-trips. * - * Based on AntigravityService.loadCodeAssist() in - * src/lib/oauth/services/antigravity.ts and the CLIProxyAPI reference + * Based on the Antigravity loadCodeAssist flow and the CLIProxyAPI reference * implementation in internal/runtime/executor/antigravity_executor.go. */ diff --git a/open-sse/services/autoCombo/__tests__/autoCombo.test.ts b/open-sse/services/autoCombo/__tests__/autoCombo.test.ts index 6ee4b5997c..c78aa5a557 100644 --- a/open-sse/services/autoCombo/__tests__/autoCombo.test.ts +++ b/open-sse/services/autoCombo/__tests__/autoCombo.test.ts @@ -628,9 +628,12 @@ describe("Task Fitness DB Resolution Chain", () => { }); it("getTaskFitnessWithSource identifies fitness_table as source for known models", () => { - const result = getTaskFitnessWithSource("gpt-4o", "coding"); + const model = "claude-sonnet"; + const category = "coding"; + + const result = getTaskFitnessWithSource(model, category); expect(result.source).toBe("fitness_table"); - expect(result.score).toBe(0.9); + expect(result.score).toBe(0.95); }); it("case insensitivity: model names are lowercased before lookup", () => { diff --git a/open-sse/services/autoCombo/routerStrategy.ts b/open-sse/services/autoCombo/routerStrategy.ts index 04cb55fb1b..466d815607 100644 --- a/open-sse/services/autoCombo/routerStrategy.ts +++ b/open-sse/services/autoCombo/routerStrategy.ts @@ -100,32 +100,101 @@ class CostStrategyImpl implements RouterStrategy { // ── LatencyStrategy: prioritize low latency + reliability ─────────────────── +function positiveMetric(value: unknown): number | null { + const numericValue = Number(value); + return Number.isFinite(numericValue) && numericValue > 0 ? numericValue : null; +} + +function boundedRate(value: unknown): number { + const numericValue = Number(value); + return Number.isFinite(numericValue) && numericValue >= 0 ? Math.min(1, numericValue) : 0; +} + +function maxPositiveMetric( + candidates: ProviderCandidate[], + readMetric: (candidate: ProviderCandidate) => unknown, + fallback = 1 +): number { + return Math.max( + ...candidates.map((candidate) => positiveMetric(readMetric(candidate)) ?? 0), + fallback + ); +} + +function latencyMetricScore(value: number | null, maxValue: number): number { + if (value == null) return 0.5; + return inverseNormalized(value, maxValue); +} + +function throughputMetricScore(value: number | null, maxValue: number): number { + if (value == null) return 0.5; + return clamp01(value / Math.max(maxValue, 0.000_001)); +} + class LatencyStrategyImpl implements RouterStrategy { readonly name = "latency"; - readonly description = "Prioritizes lowest p95 latency with reliability weighting"; + readonly description = + "Prioritizes the fastest reliable provider-model pair using TTFT, TPS, E2E latency, health, fail rate, and stability"; select(pool: ProviderCandidate[], context: RoutingContext): RoutingDecision { const healthy = pool.filter((c) => c.circuitBreakerState !== "OPEN"); const candidates = healthy.length > 0 ? healthy : pool; - const sorted = [...candidates].sort((a, b) => { - const aPenalty = a.errorRate * 1000; - const bPenalty = b.errorRate * 1000; - return a.p95LatencyMs + aPenalty - (b.p95LatencyMs + bPenalty); - }); - const best = sorted[0]; + if (candidates.length === 0) throw new Error("[LatencyStrategy] No candidates available"); + + const maxP95 = maxPositiveMetric(candidates, (candidate) => candidate.p95LatencyMs); + const maxTtft = maxPositiveMetric( + candidates, + (candidate) => candidate.avgTtftMs ?? candidate.p95LatencyMs + ); + const maxE2E = maxPositiveMetric( + candidates, + (candidate) => candidate.avgE2ELatencyMs ?? candidate.p95LatencyMs + ); + const maxTps = maxPositiveMetric(candidates, (candidate) => candidate.avgTokensPerSecond); + const maxStdDev = maxPositiveMetric(candidates, (candidate) => candidate.latencyStdDev, 0.001); + + const scored = candidates + .map((candidate) => { + const p95 = positiveMetric(candidate.p95LatencyMs); + const ttft = positiveMetric(candidate.avgTtftMs) ?? p95; + const e2e = positiveMetric(candidate.avgE2ELatencyMs) ?? p95; + const tps = positiveMetric(candidate.avgTokensPerSecond); + const failureRate = boundedRate(candidate.failureRate ?? candidate.errorRate); + const healthScore = getHealthScore(candidate); + const p95Score = latencyMetricScore(p95, maxP95); + const ttftScore = latencyMetricScore(ttft, maxTtft); + const e2eScore = latencyMetricScore(e2e, maxE2E); + const throughputScore = throughputMetricScore(tps, maxTps); + const reliabilityScore = 1 - failureRate; + const stabilityScore = latencyMetricScore( + positiveMetric(candidate.latencyStdDev), + maxStdDev + ); + const rawScore = + ttftScore * 0.25 + + throughputScore * 0.2 + + e2eScore * 0.18 + + p95Score * 0.12 + + reliabilityScore * 0.15 + + healthScore * 0.05 + + stabilityScore * 0.05; + const reliabilityMultiplier = Math.max(0.05, reliabilityScore * reliabilityScore); + const score = rawScore * reliabilityMultiplier * Math.max(0.25, healthScore); + + return { candidate, score, ttft, e2e, tps, failureRate }; + }) + .sort((a, b) => b.score - a.score); + + const best = scored[0]; if (!best) throw new Error("[LatencyStrategy] No candidates available"); - const latencyScore = best.p95LatencyMs > 0 ? Math.max(0.001, 10_000 / best.p95LatencyMs) : 1; - const reliability = Math.max(0, 1 - best.errorRate); - const finalScore = latencyScore * 0.7 + reliability * 0.3; - return { - provider: best.provider, - model: best.model, + provider: best.candidate.provider, + model: best.candidate.model, strategy: this.name, - reason: `LatencyStrategy: p95=${best.p95LatencyMs}ms, errorRate=${(best.errorRate * 100).toFixed(2)}%`, + reason: `LatencyStrategy: ttft=${best.ttft ?? "n/a"}ms, tps=${best.tps ?? "n/a"}, e2e=${best.e2e ?? "n/a"}ms, p95=${best.candidate.p95LatencyMs}ms, failRate=${(best.failureRate * 100).toFixed(2)}%`, candidatesConsidered: candidates.length, - finalScore, + finalScore: best.score, }; } } diff --git a/open-sse/services/autoCombo/scoring.ts b/open-sse/services/autoCombo/scoring.ts index 5a9d4eed76..fccffa0191 100644 --- a/open-sse/services/autoCombo/scoring.ts +++ b/open-sse/services/autoCombo/scoring.ts @@ -61,8 +61,16 @@ export interface ProviderCandidate { circuitBreakerState: "CLOSED" | "HALF_OPEN" | "OPEN"; costPer1MTokens: number; p95LatencyMs: number; + /** Average time-to-first-token in ms, when stream telemetry is available. */ + avgTtftMs?: number; + /** Average end-to-end request latency in ms, when usage telemetry is available. */ + avgE2ELatencyMs?: number; + /** Average generation throughput in output tokens/sec, when token telemetry is available. */ + avgTokensPerSecond?: number; latencyStdDev: number; errorRate: number; + /** Optional provider/model observed failure rate. Falls back to errorRate. */ + failureRate?: number; /** T10: Optional account tier for priority boosting (Ultra > Pro > Free) */ accountTier?: "ultra" | "pro" | "standard" | "free"; /** T10: Optional quota reset interval in seconds (shorter = higher priority when same quota) */ diff --git a/open-sse/services/autoCombo/virtualFactory.ts b/open-sse/services/autoCombo/virtualFactory.ts index dd386b7775..4a959120cd 100644 --- a/open-sse/services/autoCombo/virtualFactory.ts +++ b/open-sse/services/autoCombo/virtualFactory.ts @@ -240,15 +240,18 @@ export async function createVirtualAutoCombo( const candidatePool: VirtualAutoComboCandidate[] = []; for (const conn of validConnections) { + // #5873: custom OpenAI-/Anthropic-compatible providers have dynamic connection + // IDs (`*-compatible-*`) that are never keys of the static registry. Do NOT drop + // them from `auto/` routing — only fall back to the registry's first model when + // the connection has no explicit defaultModel. const providerInfo = getProviderRegistry()[conn.provider]; - if (!providerInfo) continue; // Skip unknown providers let modelId: string | undefined = conn.defaultModel; - if (!modelId) { + if (!modelId && providerInfo) { const firstModel = providerInfo.models[0]; modelId = firstModel?.id; } - if (!modelId) continue; // Skip providers without a model + if (!modelId) continue; // Skip providers without a resolvable model // Skip models that the user has hidden in the dashboard const hiddenModels = hiddenModelsMap.get(conn.provider); diff --git a/open-sse/services/backgroundTaskDetector.ts b/open-sse/services/backgroundTaskDetector.ts index c767c4c7f7..20ac799c5d 100644 --- a/open-sse/services/backgroundTaskDetector.ts +++ b/open-sse/services/backgroundTaskDetector.ts @@ -65,12 +65,27 @@ const DEFAULT_DEGRADATION_MAP: Record = { // ── State ─────────────────────────────────────────────────────────────────── -let _config: DegradationConfig = { - enabled: false, // Disabled by default — user must opt in - degradationMap: { ...DEFAULT_DEGRADATION_MAP }, - detectionPatterns: [...DEFAULT_DETECTION_PATTERNS], - stats: { detected: 0, tokensSaved: 0 }, -}; +// Backed by globalThis so the singleton is shared across the SEPARATE webpack +// module graphs Next.js builds for `instrumentation.ts` (boot-time hydration via +// applyRuntimeSettings → setBackgroundDegradationConfig) and the app-route / +// open-sse executors (per-request reads in the chat handler). A module-local `let` +// is duplicated per graph, so the operator's opt-in (`enabled:true`) applied at boot +// never reaches the request path — the degradation silently never fires (the +// #5312-class module-graph bug). Mirrors systemPrompt.ts (#2470) and thinkingBudget.ts. +const GLOBAL_KEY = "__omniroute_backgroundDegradation_config__"; +const _store = globalThis as unknown as Record; + +function getConfig(): DegradationConfig { + if (!_store[GLOBAL_KEY]) { + _store[GLOBAL_KEY] = { + enabled: false, // Disabled by default — user must opt in + degradationMap: { ...DEFAULT_DEGRADATION_MAP }, + detectionPatterns: [...DEFAULT_DETECTION_PATTERNS], + stats: { detected: 0, tokensSaved: 0 }, + }; + } + return _store[GLOBAL_KEY]!; +} // ── Config Management ─────────────────────────────────────────────────────── @@ -78,10 +93,10 @@ let _config: DegradationConfig = { * Set the background degradation config (called from settings API or startup). */ export function setBackgroundDegradationConfig(config: Partial): void { - _config = { - ..._config, + _store[GLOBAL_KEY] = { + ...getConfig(), ...config, - stats: _config.stats, // preserve stats across config changes + stats: getConfig().stats, // preserve stats across config changes }; } @@ -90,10 +105,10 @@ export function setBackgroundDegradationConfig(config: Partial + const matched = getConfig().detectionPatterns.some((pattern) => systemContent.includes(pattern.toLowerCase()) ); @@ -224,9 +239,9 @@ export function isBackgroundTask( export function getDegradedModel(originalModel: string): string { if (!originalModel) return originalModel; - const degraded = _config.degradationMap[originalModel]; + const degraded = getConfig().degradationMap[originalModel]; if (degraded) { - _config.stats.detected++; + getConfig().stats.detected++; return degraded; } diff --git a/open-sse/services/batchProcessor.ts b/open-sse/services/batchProcessor.ts index 9cde8a0321..40818df4d4 100644 --- a/open-sse/services/batchProcessor.ts +++ b/open-sse/services/batchProcessor.ts @@ -1,14 +1,20 @@ import { v4 as uuidv4 } from "uuid"; -import type { BatchRecord } from "@/lib/localDb"; +import type { BatchItemCheckpoint, BatchRecord } from "@/lib/localDb"; import { + countBatchItemCheckpoints, createFile, deleteFile, + ensureBatchItemCheckpoints, getApiKeyById, getBatch, getFileContent, getPendingBatches, getTerminalBatches, + listBatchItemCheckpoints, listFiles, + markBatchItemError, + markBatchItemProcessing, + markBatchItemResult, updateBatch, } from "@/lib/localDb"; import { dispatch } from "@/lib/batches/dispatch"; @@ -67,28 +73,12 @@ export async function processPendingBatches(): Promise { const pending = getPendingBatches(); // Phase 1: Stale recovery — in_progress/finalizing batches not in activeBatches - // are from a previous session; reset them to validating so they get picked up fresh + // are from a previous session; reset checkpointed batches to validating so they + // can be completed without replaying already-dispatched items. for (const batch of pending) { if (batch.status === "in_progress" || batch.status === "finalizing") { if (!activeBatches.has(batch.id)) { - console.log(`[BATCH] Recovering stale batch ${batch.id} (${batch.status}) → validating`); - - if (batch.outputFileId) { - deleteFile(batch.outputFileId); - } - if (batch.errorFileId) { - deleteFile(batch.errorFileId); - } - - updateBatch(batch.id, { - status: "validating", - inProgressAt: null, - finalizingAt: null, - outputFileId: null, - errorFileId: null, - requestCountsCompleted: 0, - requestCountsFailed: 0, - }); + recoverStaleBatch(batch); } } } @@ -116,6 +106,49 @@ export async function processPendingBatches(): Promise { await cleanupExpiredBatches(); } +function recoverStaleBatch(batch: BatchRecord): void { + const checkpointCount = countBatchItemCheckpoints(batch.id); + const hasPotentialExternalEffects = + batch.requestCountsTotal > 0 || + batch.requestCountsCompleted > 0 || + batch.requestCountsFailed > 0 || + batch.status === "finalizing"; + + if (checkpointCount === 0 && hasPotentialExternalEffects) { + console.warn( + `[BATCH] Stale batch ${batch.id} has no item checkpoints; failing instead of replaying provider calls` + ); + failBatch( + batch.id, + "Cannot safely recover stale batch because item checkpoints are unavailable; create a new batch to retry intentionally." + ); + return; + } + + console.log(`[BATCH] Recovering stale batch ${batch.id} (${batch.status}) → validating`); + + if (batch.outputFileId) { + deleteFile(batch.outputFileId); + } + if (batch.errorFileId) { + deleteFile(batch.errorFileId); + } + + updateBatch(batch.id, { + status: "validating", + inProgressAt: null, + finalizingAt: null, + outputFileId: null, + errorFileId: null, + ...(checkpointCount === 0 + ? { + requestCountsCompleted: 0, + requestCountsFailed: 0, + } + : {}), + }); +} + function parseBatchWindowSeconds(window: string | null | undefined): number { if (!window) return DEFAULT_BATCH_WINDOW_SECONDS; const match = /^(\d+)([hdm])$/.exec(window); @@ -285,6 +318,8 @@ async function startBatch(batch: any): Promise { console.log(`[BATCH] Batch ${batch.id} contains (${total} items)`); + ensureBatchItemCheckpoints(batch.id, parsedItems.items); + updateBatch(batch.id, { status: "in_progress", inProgressAt: Math.floor(Date.now() / 1000), @@ -316,12 +351,21 @@ const HEADERS_CACHE_TTL_MS = 60_000; async function processBatchItems(batch: BatchRecord, items: BatchRequestItem[]): Promise { const state = createBatchState(batch); + const checkpoints = new Map( + listBatchItemCheckpoints(batch.id).map((checkpoint) => [checkpoint.lineNumber, checkpoint]) + ); const apiKey = await resolveApiKey(batch); for (const item of items) { if (isBatchCancelled(batch.id)) break; + const checkpoint = checkpoints.get(item.lineNumber); + if (checkpoint && applyRecoveredCheckpoint(batch.id, item, checkpoint, state)) { + maybePersistProgress(batch.id, state); + continue; + } + const cachedHeaders = prevHeaders && Date.now() - prevHeadersTimestamp < HEADERS_CACHE_TTL_MS ? prevHeaders : null; if (cachedHeaders) { @@ -331,6 +375,8 @@ async function processBatchItems(batch: BatchRecord, items: BatchRequestItem[]): } } + markBatchItemProcessing(batch.id, item); + try { const response = await processSingleItemWithRetry(item, apiKey); let responseBody: unknown; @@ -352,13 +398,17 @@ async function processBatchItems(batch: BatchRecord, items: BatchRequestItem[]): }, }; + markBatchItemResult(batch.id, item, wrapped); state.results.push(wrapped); applyItemResult(state, response.status, responseBody); prevHeaders = response.headers; prevHeadersTimestamp = Date.now(); } catch (exception) { // Track processing-level errors separately (items that failed to be processed) - state.errors.push({ custom_id: item.customId ?? null, error: String(exception) }); + const error = { custom_id: item.customId ?? null, error: String(exception) }; + markBatchItemError(batch.id, item, error); + state.errors.push(error); + state.failed++; prevHeaders = null; prevHeadersTimestamp = 0; } @@ -369,6 +419,43 @@ async function processBatchItems(batch: BatchRecord, items: BatchRequestItem[]): return finalizeBatch(batch.id, state.results, state.errors); } +function applyRecoveredCheckpoint( + batchId: string, + item: BatchRequestItem, + checkpoint: BatchItemCheckpoint, + state: ReturnType +): boolean { + if (checkpoint.status === "completed" && checkpoint.result) { + state.results.push(checkpoint.result); + applyItemResult( + state, + checkpoint.result.response?.status_code ?? 500, + checkpoint.result.response?.body + ); + return true; + } + + if (checkpoint.status === "errored" && checkpoint.error) { + state.errors.push(checkpoint.error); + state.failed++; + return true; + } + + if (checkpoint.status === "processing") { + const error = { + custom_id: item.customId ?? null, + error: + "Batch item was interrupted before its provider response was recorded; it was not replayed to avoid duplicate provider work.", + }; + markBatchItemError(batchId, item, error); + state.errors.push(error); + state.failed++; + return true; + } + + return false; +} + function isBatchCancelled(batchId: string): boolean { const current = getBatch(batchId); diff --git a/open-sse/services/claudeTlsClient.ts b/open-sse/services/claudeTlsClient.ts index 9e09e832ac..499c32e130 100644 --- a/open-sse/services/claudeTlsClient.ts +++ b/open-sse/services/claudeTlsClient.ts @@ -20,7 +20,7 @@ import { randomUUID } from "node:crypto"; let clientPromise: Promise | null = null; let exitHookInstalled = false; -const CLAUDE_PROFILE = "chrome_149"; // matches the Vivaldi/Chrome 149 UA we send +const CLAUDE_PROFILE = "chrome_146"; // closest supported wreq-js profile (chrome_149 absent in 2.3.1, #5591) const DEFAULT_TIMEOUT_MS = Number.parseInt(process.env.OMNIROUTE_CLAUDE_TLS_TIMEOUT_MS || "", 10) || 60_000; // Grace period added to the binding's wire-level timeout before our JS-level diff --git a/open-sse/services/combo.ts b/open-sse/services/combo.ts index 1cfdc5131b..c99c9bbd1d 100644 --- a/open-sse/services/combo.ts +++ b/open-sse/services/combo.ts @@ -1278,12 +1278,22 @@ export async function handleComboChat({ log.warn("COMBO", "Failed to retrieve Last Known Good Provider. This is non-fatal.", { err }); } + const autoCandidateResilienceSettings = + relayOptions?.bypassProviderQuotaPolicy === true + ? { + ...resilienceSettings, + quotaPreflight: { + ...resilienceSettings.quotaPreflight, + enabled: false, + }, + } + : resilienceSettings; const candidates = await buildAutoCandidates( eligibleTargets, combo.name, relayOptions?.sessionId, resetWindowConfig, - resilienceSettings + autoCandidateResilienceSettings ); const routableCandidates = candidates.filter( (candidate) => candidate.quotaCutoffBlocked !== true @@ -1881,8 +1891,11 @@ export async function handleComboChat({ }); // Deep clone the body to ensure context preservation and prevent mutations - // from affecting other targets in the combo - let attemptBody = JSON.parse(JSON.stringify(body)); + // from affecting other targets in the combo. structuredClone avoids the + // full intermediate JSON string that JSON.parse(JSON.stringify(...)) builds + // (a second multi-hundred-KB allocation per target on large agent payloads), + // halving the per-target transient heap on the hot path (#5152). + let attemptBody = structuredClone(body); // Proactive Context Compression for fallbacks (Zero-Latency optimization) if ( @@ -1968,7 +1981,12 @@ export async function handleComboChat({ undefined; const effectiveConnectionId = selectedConnectionId || target.connectionId || ""; - const quality = await validateResponseQuality(result, clientRequestedStream, log); + const quality = await validateResponseQuality( + result, + clientRequestedStream, + log, + config.responseValidation + ); if (!quality.valid) { log.warn( "COMBO", @@ -2330,6 +2348,7 @@ export async function handleComboChat({ log, tag: "COMBO", exhaustedLogLevel: "info", + structuredError, }); // #2101: Prevent infinite fallback loops with 400 Bad Request errors that are genuinely @@ -3012,7 +3031,12 @@ async function handleRoundRobinCombo({ // Success — validate response quality before returning if (result.ok) { - const quality = await validateResponseQuality(result, clientRequestedStream, log); + const quality = await validateResponseQuality( + result, + clientRequestedStream, + log, + config.responseValidation + ); if (!quality.valid) { log.warn( "COMBO-RR", @@ -3237,6 +3261,7 @@ async function handleRoundRobinCombo({ log, tag: "COMBO-RR", exhaustedLogLevel: "debug", + structuredError, }); // Transient errors → mark in semaphore so round-robin stops stampeding this target. diff --git a/open-sse/services/combo/responseValidation.ts b/open-sse/services/combo/responseValidation.ts new file mode 100644 index 0000000000..d54855ee4e --- /dev/null +++ b/open-sse/services/combo/responseValidation.ts @@ -0,0 +1,206 @@ +/** + * Feature 4985 — configurable response-body validation for combo routing. + * + * A combo can declare a `responseValidation` predicate. When an upstream returns 200 OK + * but the parsed body fails the predicate, `validateResponseQuality` reports it as + * invalid, which the combo orchestrator already treats exactly like an HTTP error + * (skip this target → fail over to the next). All checks are declarative and safe: + * substring matching (no regex / no ReDoS) and a bounded dot-path resolver (no eval). + */ + +export type JsonPathCondition = "exists" | "nonEmpty" | "equals" | "notEquals"; + +export interface JsonPathPredicate { + path: string; + condition: JsonPathCondition; + value?: string | number | boolean; +} + +export interface ResponseValidationConfig { + /** The assistant content must NOT contain any of these substrings. */ + forbiddenSubstrings?: string[]; + /** The assistant content must contain ALL of these substrings. */ + requiredSubstrings?: string[]; + /** The trimmed assistant content must be at least this many characters. */ + minContentLength?: number; + /** Structural/value checks against the parsed JSON body (shape validation). */ + jsonPathPredicates?: JsonPathPredicate[]; +} + +export interface ResponseValidationResult { + valid: boolean; + reason?: string; +} + +const MAX_REASON_SNIPPET = 60; + +function snippet(value: string): string { + return value.length > MAX_REASON_SNIPPET ? `${value.slice(0, MAX_REASON_SNIPPET)}…` : value; +} + +/** + * Parse a dot/bracket path (e.g. `choices[0].message.content`) into tokens with a + * single bounded left-to-right scan — no regex, no backtracking, no eval. + */ +export function parseJsonPath(path: string): Array { + const tokens: Array = []; + let buf = ""; + const flush = () => { + if (buf) { + tokens.push(buf); + buf = ""; + } + }; + for (let i = 0; i < path.length; i++) { + const ch = path[i]; + if (ch === ".") { + flush(); + } else if (ch === "[") { + flush(); + let inner = ""; + i++; + while (i < path.length && path[i] !== "]") { + inner += path[i]; + i++; + } + const trimmed = inner.trim(); + const n = Number(trimmed); + tokens.push(trimmed !== "" && Number.isInteger(n) ? n : trimmed); + } else { + buf += ch; + } + } + flush(); + return tokens; +} + +/** Resolve a dot-path against a parsed JSON value. Returns `undefined` if any hop misses. */ +export function resolveJsonPath(root: unknown, path: string): unknown { + let current: unknown = root; + for (const token of parseJsonPath(path)) { + if (current === null || current === undefined) return undefined; + if (typeof token === "number") { + if (!Array.isArray(current)) return undefined; + current = current[token]; + } else { + if (typeof current !== "object") return undefined; + current = (current as Record)[token]; + } + } + return current; +} + +function isNonEmpty(value: unknown): boolean { + if (value === null || value === undefined) return false; + if (typeof value === "string") return value.trim().length > 0; + if (Array.isArray(value)) return value.length > 0; + if (typeof value === "object") return Object.keys(value as object).length > 0; + return Boolean(value); +} + +function checkCondition(value: unknown, condition: JsonPathCondition, expected: unknown): boolean { + switch (condition) { + case "exists": + return value !== undefined && value !== null; + case "nonEmpty": + return isNonEmpty(value); + case "equals": + return value === expected; + case "notEquals": + return value !== expected; + } +} + +/** Best-effort extraction of the assistant's text content from a chat/Responses body. */ +export function extractContentText(json: unknown): string { + if (!json || typeof json !== "object") return ""; + const obj = json as Record; + + // Chat Completions: choices[].message.content (string or array of parts). + const choices = obj.choices; + if (Array.isArray(choices)) { + const parts: string[] = []; + for (const choice of choices) { + const message = (choice as Record)?.message as + | Record + | undefined; + const content = message?.content; + if (typeof content === "string") parts.push(content); + else if (Array.isArray(content)) { + for (const part of content) { + const text = (part as Record)?.text; + if (typeof text === "string") parts.push(text); + } + } + } + if (parts.length) return parts.join(""); + } + + // Responses API: output[].content[].text + const output = obj.output; + if (Array.isArray(output)) { + const parts: string[] = []; + for (const item of output) { + const content = (item as Record)?.content; + if (Array.isArray(content)) { + for (const part of content) { + const text = (part as Record)?.text; + if (typeof text === "string") parts.push(text); + } + } + } + if (parts.length) return parts.join(""); + } + + return ""; +} + +/** + * Evaluate the configured predicate against a parsed JSON response body. + * Returns `{ valid: true }` when there is no config or all checks pass. + */ +export function evaluateResponseValidation( + json: unknown, + config: ResponseValidationConfig | undefined | null +): ResponseValidationResult { + if (!config || typeof config !== "object") return { valid: true }; + + const content = extractContentText(json); + + for (const sub of config.forbiddenSubstrings ?? []) { + if (typeof sub === "string" && sub.length > 0 && content.includes(sub)) { + return { valid: false, reason: `response contains forbidden substring "${snippet(sub)}"` }; + } + } + + for (const sub of config.requiredSubstrings ?? []) { + if (typeof sub === "string" && sub.length > 0 && !content.includes(sub)) { + return { valid: false, reason: `response missing required substring "${snippet(sub)}"` }; + } + } + + if ( + typeof config.minContentLength === "number" && + Number.isFinite(config.minContentLength) && + config.minContentLength > 0 && + content.trim().length < config.minContentLength + ) { + return { + valid: false, + reason: `response content shorter than ${config.minContentLength} chars`, + }; + } + + for (const predicate of config.jsonPathPredicates ?? []) { + if (!predicate || typeof predicate.path !== "string" || !predicate.path) continue; + const resolved = resolveJsonPath(json, predicate.path); + if (!checkCondition(resolved, predicate.condition, predicate.value)) { + return { + valid: false, + reason: `jsonpath check failed: "${snippet(predicate.path)}" ${predicate.condition}`, + }; + } + } + + return { valid: true }; +} diff --git a/open-sse/services/combo/runtimeUnits.ts b/open-sse/services/combo/runtimeUnits.ts index bceadb2fa4..96951b81a2 100644 --- a/open-sse/services/combo/runtimeUnits.ts +++ b/open-sse/services/combo/runtimeUnits.ts @@ -3,6 +3,7 @@ import { errorResponse } from "../../utils/error.ts"; import { recordComboRequest } from "../comboMetrics.ts"; import { resolveDelayMs } from "./comboPredicates.ts"; import { validateResponseQuality } from "./validateQuality.ts"; +import type { ResponseValidationConfig } from "./responseValidation.ts"; import type { ComboCollectionLike, ComboLike, @@ -227,7 +228,12 @@ export async function executeRuntimeUnitCombo(args: { }); return { response, unit }; } - const quality = await validateResponseQuality(response, clientRequestedStream, args.log); + const quality = await validateResponseQuality( + response, + clientRequestedStream, + args.log, + args.config.responseValidation as ResponseValidationConfig | undefined + ); if (quality.valid) { recordComboRequest(args.combo.name, unit.modelStr, { success: true, diff --git a/open-sse/services/combo/targetExhaustion.ts b/open-sse/services/combo/targetExhaustion.ts index bf3d69dc57..01f73d58db 100644 --- a/open-sse/services/combo/targetExhaustion.ts +++ b/open-sse/services/combo/targetExhaustion.ts @@ -15,7 +15,11 @@ * The only standardization is the log MESSAGE wording (round-robin previously dropped the * "on remaining targets" suffix) — diagnostic text only, same #code + provider info. */ -import { classifyErrorText, hasPerModelQuota, isProviderExhaustedReason } from "../accountFallback.ts"; +import { + classifyErrorText, + hasPerModelQuota, + isProviderExhaustedReason, +} from "../accountFallback.ts"; import { RateLimitReason } from "../../config/constants.ts"; import { isProviderCircuitOpenResult } from "./comboPredicates.ts"; import type { ComboLogger, ResolvedComboTarget } from "./types.ts"; @@ -51,6 +55,8 @@ export type ApplyComboTargetExhaustionOptions = { log: ComboLogger; tag: string; exhaustedLogLevel: "info" | "debug"; + /** Structured error object from upstream response — preferred over raw errorText for classification */ + structuredError?: { code?: string; type?: string; message?: string }; }; /** @@ -73,6 +79,7 @@ export function applyComboTargetExhaustion( log, tag, exhaustedLogLevel, + structuredError, } = opts; const { exhaustedProviders, exhaustedConnections, transientRateLimitedProviders } = sets; const provider = target.provider; @@ -84,12 +91,15 @@ export function applyComboTargetExhaustion( Boolean(provider && provider !== "unknown") && !hasPerModelQuota(provider, rawModel) && (isProviderExhaustedReason(fallbackResult) || - classifyErrorText(errorText) === RateLimitReason.QUOTA_EXHAUSTED || + classifyErrorText(structuredError?.code || errorText) === RateLimitReason.QUOTA_EXHAUSTED || allAccountsRateLimited); if (providerExhausted) { exhaustedProviders.add(provider); const emit = exhaustedLogLevel === "debug" ? log.debug : log.info; - emit?.(tag, `Provider ${provider} quota exhausted — marking for skip on remaining targets (#1731)`); + emit?.( + tag, + `Provider ${provider} quota exhausted — marking for skip on remaining targets (#1731)` + ); } else { if (result.status === 429 && !isTokenLimitBreach && provider && provider !== "unknown") { transientRateLimitedProviders.add(provider); diff --git a/open-sse/services/combo/types.ts b/open-sse/services/combo/types.ts index df42aef4a3..ba75daf328 100644 --- a/open-sse/services/combo/types.ts +++ b/open-sse/services/combo/types.ts @@ -63,6 +63,7 @@ export type IsModelAvailable = ( export type ComboRelayOptions = { sessionId?: string | null; config?: Record | null; + bypassProviderQuotaPolicy?: boolean; [key: string]: unknown; }; diff --git a/open-sse/services/combo/validateQuality.ts b/open-sse/services/combo/validateQuality.ts index e571c34200..fe7b00313e 100644 --- a/open-sse/services/combo/validateQuality.ts +++ b/open-sse/services/combo/validateQuality.ts @@ -10,6 +10,10 @@ import { createSSEDataLineNormalizer, isKnownNonClaudeStreamPayload, } from "../../utils/streamHelpers.ts"; +import { + evaluateResponseValidation, + type ResponseValidationConfig, +} from "./responseValidation.ts"; import { getReasoningTokens } from "../../../src/lib/usage/tokenAccounting.ts"; import type { ComboRetryAfter } from "./types.ts"; @@ -57,7 +61,8 @@ function responsesApiOutputHasContent(output: unknown): boolean { export async function validateResponseQuality( response: Response, isStreaming: boolean, - log: { warn?: (...args: unknown[]) => void } + log: { warn?: (...args: unknown[]) => void }, + responseValidation?: ResponseValidationConfig | null ): Promise<{ valid: boolean; reason?: string; clonedResponse?: Response }> { // Issue #3685: For Claude SSE streaming responses, use a BOUNDED PEEK to // detect the empty-content-block pattern (content_filter stop_reason with @@ -292,6 +297,15 @@ export async function validateResponseQuality( return { valid: false, reason: "response is not valid JSON" }; } + // Feature 4985: apply the combo's configured response-body predicate. A failure here + // fails over to the next target via the same path as the built-in empty-content checks. + if (responseValidation) { + const verdict = evaluateResponseValidation(json, responseValidation); + if (!verdict.valid) { + return { valid: false, reason: verdict.reason }; + } + } + const choices = json?.choices; if (json?.object === "response") { if (!responsesApiOutputHasContent(json.output)) return { valid: false, reason: "empty_choices" }; diff --git a/open-sse/services/comboConfig.ts b/open-sse/services/comboConfig.ts index 1378283796..aa26cd4703 100644 --- a/open-sse/services/comboConfig.ts +++ b/open-sse/services/comboConfig.ts @@ -6,6 +6,7 @@ */ import { MAX_TIMER_TIMEOUT_MS } from "../../src/shared/utils/runtimeTimeouts.ts"; +import type { ResponseValidationConfig } from "./combo/responseValidation.ts"; /** * Maximum number of concurrent pre-screen checks (provider profile + availability) @@ -63,6 +64,9 @@ const DEFAULT_COMBO_CONFIG = { resetAwareTieBandPercent: 5, resetAwareExhaustionGuardPercent: 10, failoverBeforeRetry: true, + // Feature 4985: configurable response-body validation predicate (per-combo). When set, + // a 200 OK whose body fails the predicate fails over to the next target. + responseValidation: undefined as ResponseValidationConfig | undefined, maxSetRetries: 0, setRetryDelayMs: 2000, // Zero-latency optimizations are opt-in because some modes can race targets or diff --git a/open-sse/services/compression/cacheAwareConfig.ts b/open-sse/services/compression/cacheAwareConfig.ts index 5f0c01fb71..a6ebea6ae8 100644 --- a/open-sse/services/compression/cacheAwareConfig.ts +++ b/open-sse/services/compression/cacheAwareConfig.ts @@ -1,23 +1,60 @@ import type { CompressionConfig } from "./types.ts"; import type { CachingDetectionContext } from "./cachingAware.ts"; import { detectCachingContext, getCacheAwareStrategy } from "./cachingAware.ts"; +import { + normalizePreserveSystemPromptMode, + resolvePreserveSystemPrompt, +} from "./preserveSystemPromptMode.ts"; +import { + resolvePrefixFreezeConfig, + extractStablePrefixHash, + observePrefix, + isPrefixFrozen, +} from "./prefixFreeze.ts"; /** - * #3890: honor the cache-aware `skipSystemPrompt` decision that - * `getCacheAwareStrategy` already computes but `selectCompressionStrategy` - * cannot return. In a caching context the system prompt is part of the - * cacheable prefix, so compressing it breaks the upstream prompt cache. + * T08/H5 — augment the static cache signal with usage-observed prefix freeze. Opt-in + * (default off): when a system prompt has recurred `>= threshold` times, treat it as a stable + * cacheable prefix to preserve even for a provider the static check does not flag as caching. + * "Freeze" only *preserves* the prefix, so it never corrupts a payload. + */ +function observeAndCheckPrefixFreeze(body: Record): boolean { + const cfg = resolvePrefixFreezeConfig(); + if (!cfg.enabled) return false; + const hash = extractStablePrefixHash(body); + if (!hash) return false; + observePrefix(hash); + return isPrefixFrozen(hash, cfg.threshold); +} + +/** + * #3890/#3955 + T05/C5: materialize the engine-facing `preserveSystemPrompt` + * boolean from the authoritative `preserveSystemPromptMode` intent, using the + * cache-aware `skipSystemPrompt` signal that `getCacheAwareStrategy` already + * computes (a caching provider — or `cache_control` — means the system prompt is + * part of the cacheable prefix, so compressing it breaks the upstream cache). + * + * This generalizes the previous hard-coded "force `true` when a cache is present + * and the operator disabled preservation" into the three modes: + * - `always` → always `true`. + * - `whenNoCache` → `true` only when a cache is present (the legacy `false` + * behaviour — preserved exactly for back-compat). + * - `never` → always `false`, even when it breaks a prompt cache. */ export function resolveCacheAwareConfig( config: CompressionConfig, body?: Record, context?: CachingDetectionContext ): CompressionConfig { - if (!body) return config; - const ctx = detectCachingContext(body, context); - const cacheAware = getCacheAwareStrategy(config.defaultMode, ctx); - if (cacheAware.skipSystemPrompt && config.preserveSystemPrompt === false) { - return { ...config, preserveSystemPrompt: true }; - } - return config; + const mode = normalizePreserveSystemPromptMode(config); + // No request body → no cacheable prefix to detect; honor the mode at its no-cache baseline. + // The signal is the static caching heuristic OR (H5, opt-in) a usage-observed stable prefix. + const staticCache = body + ? getCacheAwareStrategy(config.defaultMode, detectCachingContext(body, context)) + .skipSystemPrompt + : false; + const hasCache = staticCache || (body ? observeAndCheckPrefixFreeze(body) : false); + const effective = resolvePreserveSystemPrompt(mode, { hasCache }); + if (effective === config.preserveSystemPrompt) return config; + return { ...config, preserveSystemPrompt: effective }; } diff --git a/open-sse/services/compression/engines/ccr/index.ts b/open-sse/services/compression/engines/ccr/index.ts index 1fe9bdb9a7..3308d201ed 100644 --- a/open-sse/services/compression/engines/ccr/index.ts +++ b/open-sse/services/compression/engines/ccr/index.ts @@ -54,6 +54,12 @@ const ENGINE_ID = "ccr"; const DEFAULT_MIN_CHARS = 600; /** Number of retrievals before a block is flagged "do-not-compress" for that principal. */ const RETRIEVAL_THRESHOLD = 3; +/** + * H8 — default retrieval ramp factor. Each prior retrieval (below the threshold) raises a block's + * effective `minChars` linearly, so hot content is compressed progressively less; `1` disables the + * ramp (only the >= threshold cliff remains — the legacy binary behavior). + */ +const RETRIEVAL_RAMP_FACTOR_DEFAULT = 2; /** * Maximum number of entries in each bounded store. * When inserting beyond this cap, the oldest entry (Map insertion order) is evicted. @@ -141,6 +147,35 @@ export function shouldSkipCompression(hash: string, principalId?: string): boole return (retrievalCounts.get(key) ?? 0) >= RETRIEVAL_THRESHOLD; } +/** + * H8 — retrieval-aware minimum block size (graduated feedback). A frequently-retrieved + * `(principal, hash)` block is compressed progressively less: each prior retrieval (below the + * threshold) raises the size bar linearly, and once the retrieval count reaches + * `RETRIEVAL_THRESHOLD` the block is never compressed (`Infinity` — this subsumes the previous + * binary `shouldSkipCompression` cliff). Pure function of the retrieval counter; `rampFactor <= 1` + * disables the ramp so only the `>= threshold` cliff remains (byte-identical to the legacy path). + */ +export function effectiveMinChars( + baseMinChars: number, + hash: string, + principalId: string | undefined, + rampFactor: number +): number { + const count = retrievalCounts.get(buildStoreKey(hash, principalId)) ?? 0; + if (count >= RETRIEVAL_THRESHOLD) return Number.POSITIVE_INFINITY; + if (count <= 0 || rampFactor <= 1) return baseMinChars; + // Linear ramp: count=1 → base·rampFactor; count=2 → base·(1 + 2·(rampFactor−1)); … + return Math.round(baseMinChars * (1 + (rampFactor - 1) * count)); +} + +/** Resolve the H8 ramp factor from the env (`COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR`), default 2. */ +export function resolveRetrievalRampFactor(env: NodeJS.ProcessEnv = process.env): number { + const raw = env.COMPRESSION_CCR_RETRIEVAL_RAMP_FACTOR; + if (raw === undefined) return RETRIEVAL_RAMP_FACTOR_DEFAULT; + const n = Number(raw); + return Number.isFinite(n) && n >= 1 ? n : RETRIEVAL_RAMP_FACTOR_DEFAULT; +} + /** * Reset the CCR store and retrieval counts (for testing). */ @@ -202,16 +237,20 @@ function buildMarker(hash: string, charCount: number): string { function maybeCcrReplace( text: string, minChars: number, - principalId?: string + principalId: string | undefined, + rampFactor: number ): { text: string; replaced: boolean; hash: string | null } { + // Base floor first (no hash cost for tiny blocks that could never compress anyway). if (text.length < minChars) { return { text, replaced: false, hash: null }; } const hash = hashContent(text); - // Skip if this (principal, hash) pair is flagged as do-not-compress - if (shouldSkipCompression(hash, principalId)) { + // H8: retrieved blocks demand an ever-larger size before compressing; at/above the retrieval + // threshold the effective minimum is Infinity, so the block is excluded (subsumes the former + // binary shouldSkipCompression cliff). + if (text.length < effectiveMinChars(minChars, hash, principalId, rampFactor)) { return { text, replaced: false, hash: null }; } @@ -232,7 +271,8 @@ function maybeCcrReplace( function processMessages( messages: MessageLike[], minChars: number, - principalId?: string + principalId: string | undefined, + rampFactor: number ): { messages: MessageLike[]; replacedCount: number } { let replacedCount = 0; @@ -240,7 +280,7 @@ function processMessages( if (msg.role === "system") return { ...msg }; if (typeof msg.content === "string") { - const { text, replaced } = maybeCcrReplace(msg.content, minChars, principalId); + const { text, replaced } = maybeCcrReplace(msg.content, minChars, principalId, rampFactor); if (replaced) { replacedCount++; return { ...msg, content: text }; @@ -252,7 +292,12 @@ function processMessages( let changed = false; const newContent = msg.content.map((part) => { if (part["type"] !== "text" || typeof part["text"] !== "string") return part; - const { text, replaced } = maybeCcrReplace(part["text"] as string, minChars, principalId); + const { text, replaced } = maybeCcrReplace( + part["text"] as string, + minChars, + principalId, + rampFactor + ); if (replaced) { changed = true; replacedCount++; @@ -290,6 +335,18 @@ const CCR_SCHEMA: EngineConfigField[] = [ min: 100, max: 1_000_000, }, + { + key: "retrievalRampFactor", + type: "number", + label: "Retrieval ramp factor (H8)", + description: + "How steeply frequently-retrieved blocks resist compression. Each prior retrieval raises " + + "the effective minimum block size linearly; 1 disables the ramp (binary skip at the " + + "threshold only).", + defaultValue: RETRIEVAL_RAMP_FACTOR_DEFAULT, + min: 1, + max: 100, + }, ]; function validateCcrConfig(config: Record): EngineValidationResult { @@ -303,6 +360,12 @@ function validateCcrConfig(config: Record): EngineValidationRes errors.push("minChars must be a positive number"); } } + if (config["retrievalRampFactor"] !== undefined) { + const v = config["retrievalRampFactor"]; + if (typeof v !== "number" || !Number.isFinite(v) || v < 1) { + errors.push("retrievalRampFactor must be a number >= 1"); + } + } return { valid: errors.length === 0, errors }; } @@ -346,6 +409,12 @@ export const ccrEngine: CompressionEngine = { ? (stepConfig["minChars"] as number) : DEFAULT_MIN_CHARS; + // H8: retrieval-aware ramp factor — stepConfig wins, else env (default 2). 1 = binary cliff only. + const rampFactor = + typeof stepConfig["retrievalRampFactor"] === "number" + ? (stepConfig["retrievalRampFactor"] as number) + : resolveRetrievalRampFactor(); + const messages = body["messages"]; if (!Array.isArray(messages) || messages.length === 0) { return { body, compressed: false, stats: null }; @@ -355,7 +424,8 @@ export const ccrEngine: CompressionEngine = { const { messages: newMessages, replacedCount } = processMessages( messages as MessageLike[], minChars, - options?.principalId + options?.principalId, + rampFactor ); if (replacedCount === 0) { diff --git a/open-sse/services/compression/engines/index.ts b/open-sse/services/compression/engines/index.ts index 4c9c9f2a2d..e58cc3d893 100644 --- a/open-sse/services/compression/engines/index.ts +++ b/open-sse/services/compression/engines/index.ts @@ -7,6 +7,8 @@ import { ccrEngine } from "./ccr/index.ts"; import { llmlinguaEngine } from "./llmlingua/index.ts"; import { ionizerEngine } from "./ionizer/index.ts"; import { relevanceEngine } from "./relevance/index.ts"; +import { llmCompressorEngine } from "./llm/index.ts"; +import { readLifecycleEngine } from "./readLifecycle/index.ts"; let registered = false; @@ -30,6 +32,8 @@ export function registerBuiltinCompressionEngines(): void { { id: "llmlingua", engine: llmlinguaEngine }, { id: "ionizer", engine: ionizerEngine }, { id: "relevance", engine: relevanceEngine }, + { id: "llm", engine: llmCompressorEngine }, + { id: "read-lifecycle", engine: readLifecycleEngine }, ]; for (const { id, engine } of engines) { diff --git a/open-sse/services/compression/engines/llm/index.ts b/open-sse/services/compression/engines/llm/index.ts new file mode 100644 index 0000000000..0ead7e5e75 --- /dev/null +++ b/open-sse/services/compression/engines/llm/index.ts @@ -0,0 +1,346 @@ +/** + * LLM-tier async compression engine (T05/C3 — opt-in, default-off). + * + * A generic "LLM compressor" tier: it can condense the prose of non-system messages via + * a pluggable LLM backend. It mirrors the `llmlingua` engine's contract exactly, but the + * backend is a full chat-completion model rather than a local ONNX classifier. + * + * ## Safe by construction (default-off) + * - The DEFAULT backend is a no-op (`text => text`), so out of the box the engine NEVER + * mutates the payload — it returns the body unchanged (`compressed:false`). + * - It is NOT part of the default stacked pipeline; an operator must add it explicitly. + * - `enabled` defaults to `false` in the config schema. + * A real LLM backend is wired via `setLlmCompressorBackend()` (the same injection seam the + * tests use). The real production model is intentionally left as a VPS-validated follow-up + * (Hard Rule #18), exactly as the `llmlingua` worker backend is gated. + * + * ## Fail-open everywhere (never throws, never corrupts) + * 1. Backend rejection/error per prose segment → segment kept as-is. + * 2. Any unexpected error in `applyAsync` → original body, `compressed:false`, no throw. + * + * ## Code-block protection (inviolable) + * Fenced code blocks (and other preserved constructs) are tombstoned via + * `extractPreservedBlocks` and re-stitched verbatim — the engine physically never passes + * code to the backend. System messages are never touched. + */ + +import { createCompressionStats, estimateCompressionTokens } from "../../stats.ts"; +import { extractPreservedBlocks } from "../../preservation.ts"; +import type { + CompressionEngine, + CompressionEngineApplyOptions, + EngineConfigField, + EngineValidationResult, +} from "../types.ts"; +import type { CompressionResult } from "../../types.ts"; + +// ─── backend abstraction ────────────────────────────────────────────────────── + +export interface LlmCompressorBackendOptions { + model?: string; + compressionRate?: number; +} + +/** + * A backend takes a prose text segment (+ optional config) and returns a compressed + * version. Any rejection/error MUST be caught by the caller; the engine fail-opens. + */ +export type LlmCompressorBackend = ( + text: string, + opts?: LlmCompressorBackendOptions +) => Promise; + +/** The default production backend: a no-op that returns the text unchanged (pass-through). */ +const noopBackend: LlmCompressorBackend = async (text) => text; + +/** Module-level injectable backend (null = use the no-op default). */ +let _backend: LlmCompressorBackend | null = null; + +/** Override the backend — for tests and for wiring a real LLM. `null` restores the no-op. */ +export function setLlmCompressorBackend(b: LlmCompressorBackend | null): void { + _backend = b; +} + +function resolveBackend(): LlmCompressorBackend { + return _backend ?? noopBackend; +} + +// ─── prose / code splitting (code is never sent to the backend) ───────────────── + +interface TextSegment { + kind: "prose" | "preserved"; + text: string; +} + +function splitProseAndPreserved(text: string): TextSegment[] { + const { text: withPlaceholders, blocks } = extractPreservedBlocks(text); + if (blocks.length === 0) return [{ kind: "prose", text }]; + + const placeholderToOriginal = new Map(blocks.map((b) => [b.placeholder, b.content])); + const escaped = blocks.map((b) => b.placeholder.replace(/[.*+?^${}()|[\]\\]/g, "\\$&")); + const splitRe = new RegExp(`(${escaped.join("|")})`, "g"); + + const segments: TextSegment[] = []; + for (const part of withPlaceholders.split(splitRe)) { + if (!part) continue; + const original = placeholderToOriginal.get(part); + segments.push( + original !== undefined ? { kind: "preserved", text: original } : { kind: "prose", text: part } + ); + } + return segments; +} + +// ─── message processing ───────────────────────────────────────────────────────── + +type MessageLike = { + role?: string; + content?: string | Array>; + [key: string]: unknown; +}; + +async function compressProseText( + text: string, + backend: LlmCompressorBackend, + opts?: LlmCompressorBackendOptions +): Promise<{ text: string; didCompress: boolean }> { + if (!text.trim()) return { text, didCompress: false }; + try { + const compressed = await backend(text, opts); + if (typeof compressed === "string" && compressed.length < text.length) { + return { text: compressed, didCompress: true }; + } + return { text, didCompress: false }; + } catch { + return { text, didCompress: false }; + } +} + +async function compressMessageText( + text: string, + backend: LlmCompressorBackend, + opts?: LlmCompressorBackendOptions +): Promise<{ text: string; didCompress: boolean }> { + const segments = splitProseAndPreserved(text); + let anyCompressed = false; + const parts: string[] = []; + for (const seg of segments) { + if (seg.kind === "preserved") { + parts.push(seg.text); + } else { + const { text: out, didCompress } = await compressProseText(seg.text, backend, opts); + parts.push(out); + if (didCompress) anyCompressed = true; + } + } + return { text: parts.join(""), didCompress: anyCompressed }; +} + +async function processMessages( + messages: MessageLike[], + backend: LlmCompressorBackend, + opts?: LlmCompressorBackendOptions +): Promise<{ messages: MessageLike[]; compressedCount: number }> { + let compressedCount = 0; + const result: MessageLike[] = []; + for (const msg of messages) { + if (msg.role === "system") { + result.push({ ...msg }); + continue; + } + try { + if (typeof msg.content === "string") { + const { text, didCompress } = await compressMessageText(msg.content, backend, opts); + if (didCompress) { + compressedCount++; + result.push({ ...msg, content: text }); + } else { + result.push({ ...msg }); + } + } else if (Array.isArray(msg.content)) { + let changed = false; + const newContent: Array> = []; + for (const part of msg.content) { + if (part["type"] === "text" && typeof part["text"] === "string") { + const { text, didCompress } = await compressMessageText( + part["text"] as string, + backend, + opts + ); + if (didCompress) { + changed = true; + compressedCount++; + newContent.push({ ...part, text }); + } else { + newContent.push(part); + } + } else { + newContent.push(part); + } + } + result.push(changed ? { ...msg, content: newContent } : { ...msg }); + } else { + result.push({ ...msg }); + } + } catch { + result.push({ ...msg }); + } + } + return { messages: result, compressedCount }; +} + +// ─── config schema ──────────────────────────────────────────────────────────── + +const LLM_COMPRESSOR_SCHEMA: EngineConfigField[] = [ + // Default OFF: this tier costs an extra model call and mutates the payload, so it is + // opt-in (Hard Rule #20 spirit) — never on by default. + { key: "enabled", type: "boolean", label: "Enabled", defaultValue: false }, + { key: "model", type: "string", label: "Compression model", defaultValue: "" }, + { + key: "minTokens", + type: "number", + label: "Min tokens (floor)", + defaultValue: 2000, + min: 0, + max: 100000, + }, + { + key: "compressionRate", + type: "number", + label: "Compression rate (keep ratio)", + defaultValue: 0.5, + min: 0.1, + max: 0.9, + }, +]; + +function validateLlmCompressorConfig(config: Record): EngineValidationResult { + const errors: string[] = []; + if (config["enabled"] !== undefined && typeof config["enabled"] !== "boolean") { + errors.push("enabled must be a boolean"); + } + if (config["model"] !== undefined && typeof config["model"] !== "string") { + errors.push("model must be a string"); + } + if (config["minTokens"] !== undefined) { + const v = config["minTokens"]; + if (typeof v !== "number" || Number.isNaN(v) || v < 0) errors.push("minTokens must be a number >= 0"); + } + if (config["compressionRate"] !== undefined) { + const v = config["compressionRate"]; + if (typeof v !== "number" || Number.isNaN(v) || v < 0.1 || v > 0.9) { + errors.push("compressionRate must be a number between 0.1 and 0.9"); + } + } + return { valid: errors.length === 0, errors }; +} + +// ─── engine export ────────────────────────────────────────────────────────────── + +const ENGINE_ID = "llm"; + +export const llmCompressorEngine: CompressionEngine = { + id: ENGINE_ID, + name: "LLM Compressor (opt-in)", + description: + "Opt-in LLM-tier compression: condenses the prose of non-system messages via a pluggable " + + "chat-completion backend. Default-off and a no-op until an operator both enables it and wires " + + "a real backend; fenced code blocks and system messages are never sent to the model. " + + "Fail-opens on any backend error.", + icon: "robot", + targets: ["messages"], + stackable: true, + // Runs after llmlingua (35) but before ultra (40); semantic LLM rewriting is most useful + // once cheaper structural/semantic passes have already reduced the prose. + stackPriority: 38, + metadata: { + id: ENGINE_ID, + name: "LLM Compressor (opt-in)", + description: + "Opt-in LLM-tier prose compression via a pluggable backend. Default-off / no-op; " + + "code blocks and system messages are protected; fail-open on backend error.", + inputScope: "messages", + targetLatencyMs: 1500, + supportsPreview: false, + stable: true, + }, + + /** Synchronous pass-through — the real work is async-only (`applyAsync`). */ + apply(body: Record): CompressionResult { + return { body, compressed: false, stats: null }; + }, + + async applyAsync( + body: Record, + options?: CompressionEngineApplyOptions + ): Promise { + const stepConfig = options?.stepConfig ?? {}; + // Opt-in: only runs when explicitly enabled. + if (stepConfig["enabled"] !== true) { + return { body, compressed: false, stats: null }; + } + + const messages = body["messages"]; + if (!Array.isArray(messages) || messages.length === 0) { + return { body, compressed: false, stats: null }; + } + + const minTokens = + typeof stepConfig["minTokens"] === "number" ? (stepConfig["minTokens"] as number) : 2000; + if (minTokens > 0) { + const nonSystemText = (messages as MessageLike[]) + .filter((m) => m.role !== "system") + .map((m) => (typeof m.content === "string" ? m.content : JSON.stringify(m.content ?? ""))) + .join("\n"); + if (estimateCompressionTokens(nonSystemText) < minTokens) { + return { body, compressed: false, stats: null }; + } + } + + const backendOpts: LlmCompressorBackendOptions = { + model: typeof stepConfig["model"] === "string" ? (stepConfig["model"] as string) : undefined, + compressionRate: + typeof stepConfig["compressionRate"] === "number" + ? (stepConfig["compressionRate"] as number) + : undefined, + }; + + try { + const backend = resolveBackend(); + const start = performance.now(); + const { messages: newMessages, compressedCount } = await processMessages( + messages as MessageLike[], + backend, + backendOpts + ); + if (compressedCount === 0) { + return { body, compressed: false, stats: null }; + } + const newBody: Record = { ...body, messages: newMessages }; + const durationMs = Math.round(performance.now() - start); + const stats = createCompressionStats( + body, + newBody, + "stacked", + [ENGINE_ID], + [`llm-compressed-${compressedCount}-messages`], + durationMs + ); + return { body: newBody, compressed: true, stats }; + } catch { + return { body, compressed: false, stats: null }; + } + }, + + compress(body: Record, config?: Record): CompressionResult { + return this.apply(body, { stepConfig: config ?? {} }); + }, + + getConfigSchema(): EngineConfigField[] { + return LLM_COMPRESSOR_SCHEMA; + }, + + validateConfig(config: Record): EngineValidationResult { + return validateLlmCompressorConfig(config); + }, +}; diff --git a/open-sse/services/compression/engines/readLifecycle/index.ts b/open-sse/services/compression/engines/readLifecycle/index.ts new file mode 100644 index 0000000000..391ff473a2 --- /dev/null +++ b/open-sse/services/compression/engines/readLifecycle/index.ts @@ -0,0 +1,299 @@ +/** + * read-lifecycle compression engine (T08/H7 — gaps v3.8.42; opt-in, default-off). + * + * Agentic conversations re-Read the same files repeatedly. An earlier Read becomes **stale** + * once the same path is Read again (superseded by a newer view) or modified by a later + * Write/Edit. This engine collapses those superseded Read tool-results to a short stub, keeping + * only the current (last, un-superseded) Read intact. + * + * Unlike `session-dedup` (content-addressed — collapses *identical* blocks) or `ccr` (reversible + * markers), this is **semantic + lossy**: it removes stale content the later Read supersedes, so + * it is **opt-in / default-off**. It is conservative: it matches only well-known Read/Write tool + * names, compares exact paths, and collapses a Read only when a strictly-later invocation touches + * the same path. Fail-open — any unexpected shape is left untouched. + * + * Supports both the Anthropic content-block shape (`tool_use` / `tool_result`) and the OpenAI + * shape (assistant `tool_calls` + `role:"tool"` messages), linked by call id. + */ + +import { createCompressionStats } from "../../stats.ts"; +import type { + CompressionEngine, + CompressionEngineApplyOptions, + EngineConfigField, + EngineValidationResult, +} from "../types.ts"; +import type { CompressionResult } from "../../types.ts"; + +const ENGINE_ID = "read-lifecycle"; + +/** Conservative tool-name classification (exact, lower-cased) — avoids false collapses. */ +const READ_NAMES = new Set(["read", "read_file", "readfile", "view", "view_file", "cat"]); +const WRITE_NAMES = new Set([ + "write", + "write_file", + "writefile", + "edit", + "edit_file", + "multiedit", + "str_replace", + "str_replace_editor", + "str_replace_based_edit_tool", + "apply_patch", + "create_file", + "update_file", +]); + +function classifyTool(name: unknown): "read" | "write" | null { + if (typeof name !== "string") return null; + const lc = name.trim().toLowerCase(); + if (READ_NAMES.has(lc)) return "read"; + if (WRITE_NAMES.has(lc)) return "write"; + return null; +} + +function isRecord(v: unknown): v is Record { + return v !== null && typeof v === "object" && !Array.isArray(v); +} + +/** Extract a file path from a tool input object across common field names. */ +function extractPath(input: unknown): string | null { + if (!isRecord(input)) return null; + for (const key of ["file_path", "path", "filePath", "filename", "file", "target_file"]) { + const v = input[key]; + if (typeof v === "string" && v.trim()) return v.trim(); + } + return null; +} + +function parseMaybeJson(value: unknown): unknown { + if (isRecord(value)) return value; + if (typeof value === "string") { + try { + return JSON.parse(value); + } catch { + return null; + } + } + return null; +} + +interface ToolInvocation { + callId: string; + kind: "read" | "write"; + path: string; + order: number; +} + +type MessageLike = { role?: string; content?: unknown; tool_calls?: unknown; [k: string]: unknown }; + +/** + * Collect Read/Write tool invocations in conversation order (Anthropic `tool_use` + + * OpenAI `tool_calls`), with a map of read call ids → path for stub rendering. + */ +export function extractInvocations(messages: MessageLike[]): { + invocations: ToolInvocation[]; + readPathByCallId: Map; +} { + const invocations: ToolInvocation[] = []; + const readPathByCallId = new Map(); + let order = 0; + + for (const msg of messages) { + // Anthropic: content array with tool_use blocks. + if (Array.isArray(msg?.content)) { + for (const block of msg.content) { + if (!isRecord(block) || block.type !== "tool_use") continue; + const kind = classifyTool(block.name); + const callId = typeof block.id === "string" ? block.id : null; + const path = extractPath(block.input); + if (kind && callId && path) { + invocations.push({ callId, kind, path, order: order++ }); + if (kind === "read") readPathByCallId.set(callId, path); + } + } + } + // OpenAI: assistant.tool_calls array. + if (Array.isArray(msg?.tool_calls)) { + for (const call of msg.tool_calls) { + if (!isRecord(call)) continue; + const fn = isRecord(call.function) ? call.function : null; + const kind = classifyTool(fn?.name); + const callId = typeof call.id === "string" ? call.id : null; + const path = extractPath(parseMaybeJson(fn?.arguments)); + if (kind && callId && path) { + invocations.push({ callId, kind, path, order: order++ }); + if (kind === "read") readPathByCallId.set(callId, path); + } + } + } + } + return { invocations, readPathByCallId }; +} + +/** + * A read is superseded when a strictly-later invocation (read OR write) touches the same path. + * Returns the set of Read call ids whose tool-results can be collapsed. + */ +export function findSupersededReadCallIds(invocations: ToolInvocation[]): Set { + const superseded = new Set(); + for (const inv of invocations) { + if (inv.kind !== "read") continue; + const hasLater = invocations.some((o) => o.path === inv.path && o.order > inv.order); + if (hasLater) superseded.add(inv.callId); + } + return superseded; +} + +function stubFor(path: string): string { + return `[read superseded — "${path}" was re-read or modified later in the conversation; the current content appears below]`; +} + +function replaceResultText(content: unknown, stub: string): unknown { + if (typeof content === "string") return stub; + if (Array.isArray(content)) { + return content.map((part) => + isRecord(part) && part.type === "text" && typeof part.text === "string" + ? { ...part, text: stub } + : part + ); + } + return stub; +} + +/** + * Collapse the tool-results of superseded reads to a stub (Anthropic `tool_result` blocks + + * OpenAI `role:"tool"` messages), keyed by call id. Returns the new messages + collapse count. + */ +export function collapseSupersededReads( + messages: MessageLike[], + superseded: Set, + readPathByCallId: Map +): { messages: MessageLike[]; collapsedCount: number } { + if (superseded.size === 0) return { messages, collapsedCount: 0 }; + let collapsedCount = 0; + + const out = messages.map((msg) => { + // OpenAI tool result message. + if ( + msg?.role === "tool" && + typeof msg.tool_call_id === "string" && + superseded.has(msg.tool_call_id) + ) { + collapsedCount++; + const path = readPathByCallId.get(msg.tool_call_id) ?? "file"; + return { ...msg, content: stubFor(path) }; + } + // Anthropic tool_result blocks inside a content array. + if (Array.isArray(msg?.content)) { + let changed = false; + const newContent = msg.content.map((block) => { + if ( + isRecord(block) && + block.type === "tool_result" && + typeof block.tool_use_id === "string" && + superseded.has(block.tool_use_id) + ) { + changed = true; + collapsedCount++; + const path = readPathByCallId.get(block.tool_use_id) ?? "file"; + return { ...block, content: replaceResultText(block.content, stubFor(path)) }; + } + return block; + }); + if (changed) return { ...msg, content: newContent }; + } + return msg; + }); + + return { messages: out, collapsedCount }; +} + +// ─── schema ────────────────────────────────────────────────────────────────── + +const SCHEMA: EngineConfigField[] = [ + { key: "enabled", type: "boolean", label: "Enabled", defaultValue: false }, +]; + +function validate(config: Record): EngineValidationResult { + const errors: string[] = []; + if (config["enabled"] !== undefined && typeof config["enabled"] !== "boolean") { + errors.push("enabled must be a boolean"); + } + return { valid: errors.length === 0, errors }; +} + +export const readLifecycleEngine: CompressionEngine = { + id: ENGINE_ID, + name: "Read Lifecycle (opt-in)", + description: + "Collapses stale/superseded file-Read tool results: when the same path is re-read or modified " + + "later in the conversation, earlier Reads are replaced with a short stub, keeping the current " + + "Read intact. Lossy + opt-in (default off). Supports Anthropic and OpenAI tool shapes.", + icon: "history", + targets: ["messages"], + stackable: true, + // Runs early (before content engines) so stale reads are gone before other passes work on them. + stackPriority: 5, + metadata: { + id: ENGINE_ID, + name: "Read Lifecycle (opt-in)", + description: + "Collapse superseded file-Read tool results (same path re-read/modified later). " + + "Lossy, opt-in, default off. Anthropic + OpenAI shapes.", + inputScope: "messages", + targetLatencyMs: 2, + supportsPreview: true, + stable: true, + }, + + apply(body: Record, options?: CompressionEngineApplyOptions): CompressionResult { + const stepConfig = options?.stepConfig ?? {}; + if (stepConfig["enabled"] !== true) { + return { body, compressed: false, stats: null }; + } + const messages = body["messages"]; + if (!Array.isArray(messages) || messages.length === 0) { + return { body, compressed: false, stats: null }; + } + try { + const start = performance.now(); + const { invocations, readPathByCallId } = extractInvocations(messages as MessageLike[]); + const superseded = findSupersededReadCallIds(invocations); + const { messages: newMessages, collapsedCount } = collapseSupersededReads( + messages as MessageLike[], + superseded, + readPathByCallId + ); + if (collapsedCount === 0) { + return { body, compressed: false, stats: null }; + } + const newBody: Record = { ...body, messages: newMessages }; + const durationMs = Math.round(performance.now() - start); + const stats = createCompressionStats( + body, + newBody, + "stacked", + [ENGINE_ID], + [`read-lifecycle-collapsed-${collapsedCount}`], + durationMs + ); + return { body: newBody, compressed: true, stats }; + } catch { + // Fail-open: any unexpected shape leaves the body untouched. + return { body, compressed: false, stats: null }; + } + }, + + compress(body: Record, config?: Record): CompressionResult { + return this.apply(body, { stepConfig: config ?? {} }); + }, + + getConfigSchema(): EngineConfigField[] { + return SCHEMA; + }, + + validateConfig(config: Record): EngineValidationResult { + return validate(config); + }, +}; diff --git a/open-sse/services/compression/fidelityGateStep.ts b/open-sse/services/compression/fidelityGateStep.ts index cdec453b8a..df74f84278 100644 --- a/open-sse/services/compression/fidelityGateStep.ts +++ b/open-sse/services/compression/fidelityGateStep.ts @@ -1,13 +1,15 @@ import { extractTextContent } from "./messageContent.ts"; import { checkFidelity, type FidelityGateConfig } from "./fidelityGate.ts"; import type { CompressionResult } from "./types.ts"; -import type { StackAccumulator } from "./strategySelector.ts"; +import type { StackAccumulator } from "./stackedStepCore.ts"; import { getCompressionEngine } from "./engines/registry.ts"; function bodyToText(body: Record): string { const messages = body.messages; if (!Array.isArray(messages)) return ""; - return messages.map((m) => extractTextContent((m as { content?: unknown }).content as never)).join("\n"); + return messages + .map((m) => extractTextContent((m as { content?: unknown }).content as never)) + .join("\n"); } /** diff --git a/open-sse/services/compression/pipelineEngineBreaker.ts b/open-sse/services/compression/pipelineEngineBreaker.ts new file mode 100644 index 0000000000..4f75fd448f --- /dev/null +++ b/open-sse/services/compression/pipelineEngineBreaker.ts @@ -0,0 +1,139 @@ +/** + * T02 — pipeline engine circuit-breaker (gaps v3.8.42; opt-in, default OFF). + * + * A lightweight, in-memory, per-engine breaker for the stacked compression pipeline. When an + * engine throws repeatedly ACROSS requests, the breaker opens for that engine id and the + * stacked loops skip it (keeping the body verbatim for that step — fail-open) until a cooldown + * elapses, then probe once (lazy half-open). Success closes it; a failed probe re-opens it. + * + * This is deliberately NOT the provider breaker (`src/shared/utils/circuitBreaker.ts`), which + * is provider-scoped and DB-persisted. This one is engine-scoped, process-local, and adds zero + * DB/IO on the hot path. It composes with — but is independent of — the TV1 per-request bail-out + * (which skips within a single request); the breaker adds cross-request memory. + * + * Default OFF: with `enabled:false` the stacked loops never consult or mutate this state, so + * behavior is byte-identical to the pre-breaker pipeline (a throwing engine still propagates + * unless TV1 bail-out is separately enabled). + */ + +export interface PipelineCircuitBreakerConfig { + /** Master switch. Default false — the pipeline never consults the breaker when off. */ + enabled: boolean; + /** Consecutive cross-request failures before an engine's breaker opens. Default 3. */ + failureThreshold: number; + /** How long the breaker stays open before a half-open probe (ms). Default 30_000. */ + cooldownMs: number; +} + +export const DEFAULT_PIPELINE_BREAKER: PipelineCircuitBreakerConfig = { + enabled: false, + failureThreshold: 3, + cooldownMs: 30_000, +}; + +interface EngineBreakerState { + failures: number; + /** Epoch ms until which the engine is OPEN; null = CLOSED. */ + openedUntil: number | null; +} + +const _state = new Map(); + +function get(engine: string): EngineBreakerState { + let s = _state.get(engine); + if (!s) { + s = { failures: 0, openedUntil: null }; + _state.set(engine, s); + } + return s; +} + +function toNonNegativeInt(raw: string | undefined, fallback: number): number { + if (raw === undefined) return fallback; + const n = Number(raw); + return Number.isFinite(n) && n >= 0 ? Math.floor(n) : fallback; +} + +/** + * Resolve the breaker config from a partial (per-combo config) with env fallback. Env keys: + * `COMPRESSION_PIPELINE_BREAKER_ENABLED|_THRESHOLD|_COOLDOWN_MS`. The partial wins over env. + */ +export function resolvePipelineBreakerConfig( + partial?: Partial, + env: NodeJS.ProcessEnv = process.env +): PipelineCircuitBreakerConfig { + const enabled = partial?.enabled ?? env.COMPRESSION_PIPELINE_BREAKER_ENABLED === "true"; + const failureThreshold = + partial?.failureThreshold ?? + toNonNegativeInt( + env.COMPRESSION_PIPELINE_BREAKER_THRESHOLD, + DEFAULT_PIPELINE_BREAKER.failureThreshold + ); + const cooldownMs = + partial?.cooldownMs ?? + toNonNegativeInt( + env.COMPRESSION_PIPELINE_BREAKER_COOLDOWN_MS, + DEFAULT_PIPELINE_BREAKER.cooldownMs + ); + return { + enabled: enabled === true, + failureThreshold: Math.max(1, failureThreshold), + cooldownMs: Math.max(0, cooldownMs), + }; +} + +/** + * Whether an engine may run now. CLOSED → true. OPEN within cooldown → false. OPEN past the + * cooldown → lazily transitions to half-open (one probe allowed; a probe failure re-opens it + * immediately because failures is left one short of the threshold). `now` is injectable for tests. + */ +export function canRunEngine( + engine: string, + config: PipelineCircuitBreakerConfig, + now: number = Date.now() +): boolean { + if (!config.enabled) return true; + const s = get(engine); + if (s.openedUntil == null) return true; // CLOSED (or already half-open) + if (now >= s.openedUntil) { + // Cooldown elapsed → half-open: allow a single probe, but keep failures one below the + // threshold so the very next failure re-opens the breaker without N more round-trips. + s.openedUntil = null; + s.failures = Math.max(0, config.failureThreshold - 1); + return true; + } + return false; // OPEN +} + +/** Record a failed engine run; opens the breaker once consecutive failures hit the threshold. */ +export function recordEngineFailure( + engine: string, + config: PipelineCircuitBreakerConfig, + now: number = Date.now() +): void { + if (!config.enabled) return; + const s = get(engine); + s.failures += 1; + if (s.failures >= config.failureThreshold) { + s.openedUntil = now + config.cooldownMs; + } +} + +/** Record a successful engine run; fully closes the breaker. */ +export function recordEngineSuccess(engine: string, config: PipelineCircuitBreakerConfig): void { + if (!config.enabled) return; + const s = get(engine); + s.failures = 0; + s.openedUntil = null; +} + +/** Inspect breaker state for telemetry/tests. */ +export function getEngineBreakerState(engine: string): { failures: number; open: boolean } { + const s = _state.get(engine); + return { failures: s?.failures ?? 0, open: s?.openedUntil != null }; +} + +/** Clear all breaker state (tests + operator reset). */ +export function resetPipelineEngineBreakers(): void { + _state.clear(); +} diff --git a/open-sse/services/compression/prefixFreeze.ts b/open-sse/services/compression/prefixFreeze.ts new file mode 100644 index 0000000000..a46595ea4d --- /dev/null +++ b/open-sse/services/compression/prefixFreeze.ts @@ -0,0 +1,124 @@ +/** + * T08/H5 — usage-observed prefix freeze (gaps v3.8.42; opt-in, default OFF). + * + * Augments the static cache-aware heuristic (`getCacheAwareStrategy`). Instead of only freezing the + * system prompt for providers the static check knows to cache, it OBSERVES which system prompts + * actually recur across requests and, once one has been seen `>= threshold` times, treats it as a + * stable cacheable prefix to preserve — even for a provider the static check does not recognize. + * + * Content-addressed by a hash of the system prompt (no principal needed): a "freeze" only + * *preserves* the prefix from compression, which never corrupts a payload and is safe to share + * across tenants. In-memory + bounded, zero DB/IO. Default OFF + * (`COMPRESSION_PREFIX_FREEZE_ENABLED`); when off the observer is never consulted (zero cost). + */ + +import crypto from "node:crypto"; + +export interface PrefixFreezeConfig { + /** Master switch. Default false — the resolver never observes/consults when off. */ + enabled: boolean; + /** Times a system prompt must be observed before it is treated as a frozen stable prefix. */ + threshold: number; +} + +export const DEFAULT_PREFIX_FREEZE: PrefixFreezeConfig = { enabled: false, threshold: 3 }; + +/** Upper bound on tracked prefixes (oldest evicted first), mirroring the CCR store cap. */ +export const MAX_PREFIX_ENTRIES = 5_000; + +const observations = new Map(); + +function boundedInc(hash: string): void { + if (!observations.has(hash) && observations.size >= MAX_PREFIX_ENTRIES) { + const oldest = observations.keys().next().value; + if (oldest !== undefined) observations.delete(oldest); + } + observations.set(hash, (observations.get(hash) ?? 0) + 1); +} + +function toPositiveInt(raw: string | undefined, fallback: number): number { + if (raw === undefined) return fallback; + const n = Number(raw); + return Number.isFinite(n) && n >= 1 ? Math.floor(n) : fallback; +} + +/** Resolve config from env (`COMPRESSION_PREFIX_FREEZE_ENABLED` / `_THRESHOLD`). Opt-in. */ +export function resolvePrefixFreezeConfig( + env: NodeJS.ProcessEnv = process.env +): PrefixFreezeConfig { + return { + enabled: env.COMPRESSION_PREFIX_FREEZE_ENABLED === "true", + threshold: toPositiveInt( + env.COMPRESSION_PREFIX_FREEZE_THRESHOLD, + DEFAULT_PREFIX_FREEZE.threshold + ), + }; +} + +// ─── prefix extraction ────────────────────────────────────────────────────────── + +function isRecord(v: unknown): v is Record { + return v !== null && typeof v === "object" && !Array.isArray(v); +} + +/** + * Pull plain text out of the system-prompt shapes: `string`, `Array<…>`, `{text}` (OpenAI/Claude + * text parts), and `{parts: [...]}` (Gemini `systemInstruction`). + */ +function collectText(value: unknown, out: string[]): void { + if (typeof value === "string") { + if (value.trim()) out.push(value); + return; + } + if (Array.isArray(value)) { + for (const item of value) collectText(item, out); + return; + } + if (isRecord(value)) { + if (typeof value.text === "string" && value.text.trim()) out.push(value.text); + if (value.parts !== undefined) collectText(value.parts, out); // Gemini systemInstruction + } +} + +/** + * Compute a stable 24-hex hash of the request's system prompt across the common body shapes + * (OpenAI `messages[].role==="system"`, Claude `system`, Gemini `systemInstruction`). Returns + * `null` when there is no system prompt to freeze. + */ +export function extractStablePrefixHash(body: unknown): string | null { + if (!isRecord(body)) return null; + const parts: string[] = []; + collectText(body.system, parts); + collectText(body.systemInstruction, parts); + collectText(body.system_instruction, parts); + const messages = body.messages; + if (Array.isArray(messages)) { + for (const msg of messages) { + if (isRecord(msg) && msg.role === "system") collectText(msg.content, parts); + } + } + if (parts.length === 0) return null; + return crypto.createHash("sha256").update(parts.join("\n")).digest("hex").slice(0, 24); +} + +// ─── observation + query ────────────────────────────────────────────────────────── + +/** Record one observation of a system-prompt hash (call once per request when enabled). */ +export function observePrefix(hash: string): void { + boundedInc(hash); +} + +/** True once a prefix has been observed `>= threshold` times (a stable, freezable prefix). */ +export function isPrefixFrozen(hash: string, threshold: number): boolean { + return (observations.get(hash) ?? 0) >= threshold; +} + +/** Current observation count for a prefix hash (telemetry/tests). */ +export function getPrefixObservations(hash: string): number { + return observations.get(hash) ?? 0; +} + +/** Clear all observations (tests + operator reset). */ +export function resetPrefixFreeze(): void { + observations.clear(); +} diff --git a/open-sse/services/compression/preserveSystemPromptMode.ts b/open-sse/services/compression/preserveSystemPromptMode.ts new file mode 100644 index 0000000000..39528d442a --- /dev/null +++ b/open-sse/services/compression/preserveSystemPromptMode.ts @@ -0,0 +1,72 @@ +import type { CompressionConfig, PreserveSystemPromptMode } from "./types.ts"; + +/** + * T05/C5 — system-prompt preservation mode. + * + * The engine-facing field is the boolean `CompressionConfig.preserveSystemPrompt` + * (truthy = skip/preserve the system prompt). Its authoritative *intent* is the + * `preserveSystemPromptMode` enum, resolved to that boolean at the cache-aware layer + * (`resolveCacheAwareConfig`), which already runs upstream in `chatCore` with the + * caching context. This module is the single source of the enum<->boolean mapping so + * the legacy boolean and the new enum can never drift. + * + * Mode semantics: + * - `always` → preserve the system prompt unconditionally. + * - `whenNoCache` → preserve it only when there is a cache to protect (provider caches + * the prefix or `cache_control` is present); compress it otherwise. + * This is exactly what the legacy `preserveSystemPrompt: false` already + * did via the #3890/#3955 cache guard. + * - `never` → compress the system prompt even when it breaks a prompt cache. + */ +export const PRESERVE_SYSTEM_PROMPT_MODES: readonly PreserveSystemPromptMode[] = [ + "always", + "whenNoCache", + "never", +]; + +export function isPreserveSystemPromptMode(value: unknown): value is PreserveSystemPromptMode { + return ( + typeof value === "string" && + (PRESERVE_SYSTEM_PROMPT_MODES as readonly string[]).includes(value) + ); +} + +/** + * Back-compat shim: derive the authoritative mode from a config. An explicit + * `preserveSystemPromptMode` always wins; otherwise the legacy boolean is mapped + * 1:1 to the behaviour it already had (`false → whenNoCache`, anything else → `always`). + */ +export function normalizePreserveSystemPromptMode( + config: Pick +): PreserveSystemPromptMode { + if (isPreserveSystemPromptMode(config.preserveSystemPromptMode)) { + return config.preserveSystemPromptMode; + } + return config.preserveSystemPrompt === false ? "whenNoCache" : "always"; +} + +/** + * Resolve a mode to the effective engine-facing boolean given whether a cacheable + * prefix is present. `true` = preserve (skip), `false` = compress the system prompt. + */ +export function resolvePreserveSystemPrompt( + mode: PreserveSystemPromptMode, + { hasCache }: { hasCache: boolean } +): boolean { + switch (mode) { + case "always": + return true; + case "never": + return false; + case "whenNoCache": + return hasCache; + } +} + +/** + * The no-cache projection of a mode — the stored/effective boolean used as a sane + * default for any reader that runs before the cache-aware layer materializes it. + */ +export function modeToBaselineBoolean(mode: PreserveSystemPromptMode): boolean { + return resolvePreserveSystemPrompt(mode, { hasCache: false }); +} diff --git a/open-sse/services/compression/stackedStepCore.ts b/open-sse/services/compression/stackedStepCore.ts new file mode 100644 index 0000000000..f82413d994 --- /dev/null +++ b/open-sse/services/compression/stackedStepCore.ts @@ -0,0 +1,87 @@ +/** + * Core telemetry + step-decision machinery for the stacked compression pipeline, extracted + * from `strategySelector.ts` so that god-file stays bounded. This module is a leaf: it depends + * only on the shared compression types, so it never participates in a cycle. + * + * It holds the per-run accumulator (`StackAccumulator` + `createStackAccumulator`), the TV1 + * bail-out config + advance decision (`BailoutConfig` + `decideStep`), and the per-step + * telemetry fold (`mergeStackStep`). The sync/async stacked loops in `strategySelector.ts` + * consume these. + */ + +import type { CompressionResult, CompressionStats } from "./types.ts"; + +/** + * TV1 — Opt-in bail-out configuration for the stacked pipeline. + * When enabled: a step that throws is silently skipped (verbatim kept); a step whose gain is + * below `minGainPercent` is also skipped. DEFAULT = disabled (byte-identical to pre-TV1). + */ +export interface BailoutConfig { + enabled: boolean; + /** Minimum savings percent required to advance currentBody. Default: 10. */ + minGainPercent?: number; +} + +/** Accumulates per-step telemetry across a stacked run (shared sync/async). */ +export interface StackAccumulator { + techniques: Set; + rules: Set; + breakdown: NonNullable; + rtkRawOutputPointers: NonNullable; + validationWarnings: Set; + validationErrors: Set; + fallbackApplied: boolean; +} + +export function createStackAccumulator(): StackAccumulator { + return { + techniques: new Set(), + rules: new Set(), + breakdown: [], + rtkRawOutputPointers: [], + validationWarnings: new Set(), + validationErrors: new Set(), + fallbackApplied: false, + }; +} + +/** + * TV1 — Pure helper that decides whether a completed step should advance `currentBody`. Called + * only when bail-out is ENABLED; the loops bypass it on the default-off path (zero cost). Returns + * `{ advance: true }` to accept the step, or `{ advance: false }` to skip it (verbatim kept). + */ +export function decideStep( + result: CompressionResult, + bailout: BailoutConfig +): { advance: boolean } { + if (!result.compressed) return { advance: false }; + // Clamp: a negative minGainPercent would mean "always advance" (invalid state). + const minGain = Math.max(0, bailout.minGainPercent ?? 10); + const gain = result.stats?.savingsPercent ?? 0; + if (gain < minGain) return { advance: false }; + return { advance: true }; +} + +/** Folds one engine result into the accumulator (telemetry + breakdown entry). */ +export function mergeStackStep( + acc: StackAccumulator, + engineId: string, + result: CompressionResult +): void { + if (!result.stats) return; + result.stats.techniquesUsed.forEach((technique) => acc.techniques.add(technique)); + result.stats.rulesApplied?.forEach((rule) => acc.rules.add(rule)); + result.stats.rtkRawOutputPointers?.forEach((pointer) => acc.rtkRawOutputPointers.push(pointer)); + result.stats.validationWarnings?.forEach((warning) => acc.validationWarnings.add(warning)); + result.stats.validationErrors?.forEach((error) => acc.validationErrors.add(error)); + acc.fallbackApplied = acc.fallbackApplied || result.stats.fallbackApplied === true; + acc.breakdown.push({ + engine: engineId, + originalTokens: result.stats.originalTokens, + compressedTokens: result.stats.compressedTokens, + savingsPercent: result.stats.savingsPercent, + techniquesUsed: result.stats.techniquesUsed, + ...(result.stats.rulesApplied ? { rulesApplied: result.stats.rulesApplied } : {}), + ...(result.stats.durationMs !== undefined ? { durationMs: result.stats.durationMs } : {}), + }); +} diff --git a/open-sse/services/compression/strategySelector.ts b/open-sse/services/compression/strategySelector.ts index baa180599d..0a7adbd0e8 100644 --- a/open-sse/services/compression/strategySelector.ts +++ b/open-sse/services/compression/strategySelector.ts @@ -3,7 +3,6 @@ import type { CompressionMode, CompressionPipelineStep, CompressionResult, - CompressionStats, } from "./types.ts"; import { applyHardBudget } from "./hardBudget.ts"; import { type FidelityGateConfig } from "./fidelityGate.ts"; @@ -15,6 +14,20 @@ import { compressAggressive } from "./aggressive.ts"; import { ultraCompress, ultraCompressHeuristic } from "./ultra.ts"; import { createCompressionStats } from "./stats.ts"; import { guardPipelineInflation } from "./pipelineGuards.ts"; +import { + resolvePipelineBreakerConfig, + canRunEngine, + recordEngineFailure, + recordEngineSuccess, + type PipelineCircuitBreakerConfig, +} from "./pipelineEngineBreaker.ts"; +import { + type BailoutConfig, + type StackAccumulator, + createStackAccumulator, + decideStep, + mergeStackStep, +} from "./stackedStepCore.ts"; import { registerBuiltinCompressionEngines } from "./engines/index.ts"; import { getCompressionEngine, getEngineEntry } from "./engines/registry.ts"; import { applyRtkCompression } from "./engines/rtk/index.ts"; @@ -603,18 +616,6 @@ function normalizePipelineStep(step: CompressionPipelineStep | string): Compress return { engine: "caveman" }; } -/** - * TV1 — Opt-in bail-out configuration for the stacked pipeline. - * When enabled: a step that throws is silently skipped (verbatim kept); - * a step whose gain is below minGainPercent is also skipped. - * DEFAULT = disabled — behaviour is byte-identical to pre-TV1 when absent. - */ -interface BailoutConfig { - enabled: boolean; - /** Minimum savings percent required to advance currentBody. Default: 10. */ - minGainPercent?: number; -} - /** Per-engine progress emitted mid-pipeline by the stacked loops (F3.3 live streaming). */ export interface StackedCompressionStep { stepIndex: number; @@ -634,6 +635,8 @@ interface StackOptions { compressionComboId?: string | null; /** TV1 bail-out discipline (opt-in, default disabled). */ bailout?: BailoutConfig; + /** T02 per-engine circuit-breaker (opt-in, default disabled). Falls back to config + env. */ + circuitBreaker?: Partial; /** Opt-in per-step fidelity gate (default disabled). */ fidelityGate?: FidelityGateConfig; /** Risk-gate mask/restore wrapper (opt-in, default off). Read via resolveRiskGate. */ @@ -666,29 +669,6 @@ function reportEngineStep( }); } -/** Accumulates per-step telemetry across a stacked run (shared sync/async). */ -export interface StackAccumulator { - techniques: Set; - rules: Set; - breakdown: NonNullable; - rtkRawOutputPointers: NonNullable; - validationWarnings: Set; - validationErrors: Set; - fallbackApplied: boolean; -} - -function createStackAccumulator(): StackAccumulator { - return { - techniques: new Set(), - rules: new Set(), - breakdown: [], - rtkRawOutputPointers: [], - validationWarnings: new Set(), - validationErrors: new Set(), - fallbackApplied: false, - }; -} - function resolveStackSteps( pipeline?: Array ): CompressionPipelineStep[] { @@ -715,43 +695,6 @@ function buildStepOptions( }; } -/** - * TV1 — Pure helper that decides whether a completed step should advance - * `currentBody`. Called only when bailout is ENABLED; the sync/async loops - * bypass this entirely on the default-off path (zero cost, zero behaviour change). - * - * Returns `{ advance: true }` when the step should be accepted, or - * `{ advance: false }` when it should be skipped (verbatim kept). - */ -function decideStep(result: CompressionResult, bailout: BailoutConfig): { advance: boolean } { - if (!result.compressed) return { advance: false }; - // Clamp: a negative minGainPercent would mean "always advance" (invalid state). - const minGain = Math.max(0, bailout.minGainPercent ?? 10); - const gain = result.stats?.savingsPercent ?? 0; - if (gain < minGain) return { advance: false }; - return { advance: true }; -} - -/** Folds one engine result into the accumulator (telemetry + breakdown entry). */ -function mergeStackStep(acc: StackAccumulator, engineId: string, result: CompressionResult): void { - if (!result.stats) return; - result.stats.techniquesUsed.forEach((technique) => acc.techniques.add(technique)); - result.stats.rulesApplied?.forEach((rule) => acc.rules.add(rule)); - result.stats.rtkRawOutputPointers?.forEach((pointer) => acc.rtkRawOutputPointers.push(pointer)); - result.stats.validationWarnings?.forEach((warning) => acc.validationWarnings.add(warning)); - result.stats.validationErrors?.forEach((error) => acc.validationErrors.add(error)); - acc.fallbackApplied = acc.fallbackApplied || result.stats.fallbackApplied === true; - acc.breakdown.push({ - engine: engineId, - originalTokens: result.stats.originalTokens, - compressedTokens: result.stats.compressedTokens, - savingsPercent: result.stats.savingsPercent, - techniquesUsed: result.stats.techniquesUsed, - ...(result.stats.rulesApplied ? { rulesApplied: result.stats.rulesApplied } : {}), - ...(result.stats.durationMs !== undefined ? { durationMs: result.stats.durationMs } : {}), - }); -} - function finalizeStackedResult( originalBody: Record, currentBody: Record, @@ -815,6 +758,52 @@ function finalizeStackedResult( return { body: currentBody, compressed, stats }; } +// ── Shared per-step helpers (used by the sync + async stacked loops; keep them in lockstep) ── + +interface StepCommitCtx { + bailout?: BailoutConfig; + breakerOn: boolean; + breaker: PipelineCircuitBreakerConfig; + fidelityGate?: FidelityGateConfig; +} + +/** Failure path: record the breaker failure (when on) + keep the verbatim body, surfacing it in telemetry. */ +function recordStepFailure( + acc: StackAccumulator, + engineId: string, + err: unknown, + ctx: StepCommitCtx +): void { + if (ctx.breakerOn) recordEngineFailure(engineId, ctx.breaker); + acc.validationErrors.add( + `${engineId}: bailed out — ${err instanceof Error ? err.message : String(err)}` + ); + acc.fallbackApplied = true; +} + +/** + * Success path: record the breaker success (when on), merge telemetry, and decide whether to + * advance `currentBody`. Advance rule: TV1 bail-out uses min-gain (`decideStep`); otherwise the + * legacy `result.compressed`. Returns the (possibly unchanged) body + whether it advanced. + */ +function commitStepResult( + acc: StackAccumulator, + step: CompressionPipelineStep, + result: CompressionResult, + currentBody: Record, + ctx: StepCommitCtx +): { body: Record; advanced: boolean } { + if (ctx.breakerOn) recordEngineSuccess(step.engine, ctx.breaker); + mergeStackStep(acc, step.engine, result); + const advance = ctx.bailout?.enabled + ? decideStep(result, ctx.bailout).advance + : result.compressed; + if (advance && gateAdvance(result, currentBody, ctx.fidelityGate, acc, step.engine)) { + return { body: result.body, advanced: true }; + } + return { body: currentBody, advanced: false }; +} + export function applyStackedCompression( body: Record, pipeline?: Array, @@ -839,6 +828,10 @@ function runStackedCompression( const start = performance.now(); const bailout = options?.bailout; + const breaker = resolvePipelineBreakerConfig( + options?.circuitBreaker ?? options?.config?.pipelineCircuitBreaker + ); + const breakerOn = breaker.enabled; const fidelityGate = options?.fidelityGate ?? options?.config?.fidelityGate; const onStep = options?.onEngineStep; const totalSteps = steps.length; @@ -850,39 +843,31 @@ function runStackedCompression( // Respect the registry enabled flag: a step naming a disabled engine is skipped, so an // operator can turn an engine off (setEngineEnabled) without editing every pipeline. if (getEngineEntry(step.engine)?.enabled === false) continue; + // T02: when the per-engine breaker is OPEN, skip this step (verbatim body kept — fail-open). + if (breakerOn && !canRunEngine(step.engine, breaker)) { + acc.validationWarnings.add(`${step.engine}: skipped (pipeline circuit-breaker open)`); + continue; + } - // TV1: when bail-out is ENABLED, wrap apply() and apply skip rules. - // When DISABLED (default), the code path below is identical to pre-TV1. - if (bailout?.enabled) { - let result: CompressionResult; + // TV1 bail-out (per-request) OR T02 breaker (cross-request) wrap the call so a throwing engine + // is caught + recorded; when neither is on, a throw propagates (byte-identical to legacy). + const ctx = { bailout, breakerOn, breaker, fidelityGate }; + let result: CompressionResult; + if (bailout?.enabled || breakerOn) { try { result = engine.apply(currentBody, buildStepOptions(step, options)); } catch (err) { - // Failure bail-out: keep the verbatim body for this step, but RECORD the - // failure so a crashing engine is visible in telemetry (not silently gone). - acc.validationErrors.add( - `${step.engine}: bailed out — ${err instanceof Error ? err.message : String(err)}` - ); - acc.fallbackApplied = true; + recordStepFailure(acc, step.engine, err, ctx); continue; } - mergeStackStep(acc, step.engine, result); - if ( - decideStep(result, bailout).advance && - gateAdvance(result, currentBody, fidelityGate, acc, step.engine) - ) { - currentBody = result.body; - compressed = true; - } } else { - const result = engine.apply(currentBody, buildStepOptions(step, options)); - mergeStackStep(acc, step.engine, result); - if (result.compressed && gateAdvance(result, currentBody, fidelityGate, acc, step.engine)) { - currentBody = result.body; - compressed = true; - } - reportEngineStep(onStep, stepIdx++, totalSteps, step.engine, result); + result = engine.apply(currentBody, buildStepOptions(step, options)); } + const committed = commitStepResult(acc, step, result, currentBody, ctx); + currentBody = committed.body; + if (committed.advanced) compressed = true; + // The pre-existing bail-out path did not stream per-step; everything else does. + if (!bailout?.enabled) reportEngineStep(onStep, stepIdx++, totalSteps, step.engine, result); } // Hard-budget post-pass (#17): runs after all engines, before finalize. @@ -943,6 +928,10 @@ async function runStackedCompressionAsync( const start = performance.now(); const bailout = options?.bailout; + const breaker = resolvePipelineBreakerConfig( + options?.circuitBreaker ?? options?.config?.pipelineCircuitBreaker + ); + const breakerOn = breaker.enabled; const fidelityGate = options?.fidelityGate ?? options?.config?.fidelityGate; const onStep = options?.onEngineStep; const totalSteps = steps.length; @@ -953,43 +942,34 @@ async function runStackedCompressionAsync( if (!engine) continue; // Respect the registry enabled flag (same as the sync loop) — keep both in lockstep. if (getEngineEntry(step.engine)?.enabled === false) continue; + // T02: skip an engine whose breaker is OPEN (verbatim body kept — fail-open). Lockstep w/ sync. + if (breakerOn && !canRunEngine(step.engine, breaker)) { + acc.validationWarnings.add(`${step.engine}: skipped (pipeline circuit-breaker open)`); + continue; + } const stepOptions = buildStepOptions(step, options); - // TV1: same bail-out discipline as the sync loop (opt-in, default off). - if (bailout?.enabled) { - let result: CompressionResult; + // TV1 bail-out (per-request) OR T02 breaker (cross-request) wrap the call (lockstep w/ sync). + const ctx = { bailout, breakerOn, breaker, fidelityGate }; + let result: CompressionResult; + if (bailout?.enabled || breakerOn) { try { result = engine.applyAsync ? await engine.applyAsync(currentBody, stepOptions) : engine.apply(currentBody, stepOptions); } catch (err) { - // Failure bail-out: keep the verbatim body, but RECORD the failure so a - // crashing engine is visible in telemetry (not silently gone). - acc.validationErrors.add( - `${step.engine}: bailed out — ${err instanceof Error ? err.message : String(err)}` - ); - acc.fallbackApplied = true; + recordStepFailure(acc, step.engine, err, ctx); continue; } - mergeStackStep(acc, step.engine, result); - if ( - decideStep(result, bailout).advance && - gateAdvance(result, currentBody, fidelityGate, acc, step.engine) - ) { - currentBody = result.body; - compressed = true; - } } else { - const result = engine.applyAsync + result = engine.applyAsync ? await engine.applyAsync(currentBody, stepOptions) : engine.apply(currentBody, stepOptions); - mergeStackStep(acc, step.engine, result); - if (result.compressed && gateAdvance(result, currentBody, fidelityGate, acc, step.engine)) { - currentBody = result.body; - compressed = true; - } - reportEngineStep(onStep, stepIdx++, totalSteps, step.engine, result); } + const committed = commitStepResult(acc, step, result, currentBody, ctx); + currentBody = committed.body; + if (committed.advanced) compressed = true; + if (!bailout?.enabled) reportEngineStep(onStep, stepIdx++, totalSteps, step.engine, result); } // Hard-budget post-pass (#17): runs after all engines, before finalize. diff --git a/open-sse/services/compression/types.ts b/open-sse/services/compression/types.ts index 755d9256f1..5df5f266c4 100644 --- a/open-sse/services/compression/types.ts +++ b/open-sse/services/compression/types.ts @@ -13,6 +13,7 @@ import { ENGINE_IDS } from "./engineCatalog.ts"; import type { ContextBudgetConfig } from "./adaptiveCompression/types.ts"; import type { FidelityGateConfig } from "./fidelityGate.ts"; import type { RiskGateConfig } from "./riskGate/riskGate.ts"; +import type { PipelineCircuitBreakerConfig } from "./pipelineEngineBreaker.ts"; import type { RiskGateStats } from "./riskGate/riskGateStep.ts"; import type { QuantumLockConfig, QuantumLockStats } from "./quantumLock/quantumPatterns.ts"; @@ -24,13 +25,7 @@ import type { QuantumLockConfig, QuantumLockStats } from "./quantumLock/quantumP export { ENGINE_IDS }; export type CompressionMode = - | "off" - | "lite" - | "standard" - | "aggressive" - | "ultra" - | "rtk" - | "stacked"; + "off" | "lite" | "standard" | "aggressive" | "ultra" | "rtk" | "stacked"; export type CavemanIntensity = "lite" | "full" | "ultra"; export type RtkIntensity = "minimal" | "standard" | "aggressive"; export type RtkRawOutputRetention = "never" | "failures" | "always"; @@ -145,13 +140,34 @@ export interface EngineToggle { level?: string; } +/** T05/C5 — system-prompt preservation intent (see `CompressionConfig.preserveSystemPromptMode`). */ +export type PreserveSystemPromptMode = "always" | "whenNoCache" | "never"; + export interface CompressionConfig { enabled: boolean; defaultMode: CompressionMode; autoTriggerMode?: CompressionMode; autoTriggerTokens: number; cacheMinutes: number; + /** + * Effective, engine-facing boolean: when truthy the system prompt is skipped + * (preserved, not compressed). Kept as the materialized value all engines read. + * Its authoritative *intent* is `preserveSystemPromptMode` (T05/C5); this boolean + * is the no-cache projection of that mode, refined up to `true` by + * `resolveCacheAwareConfig` when a cacheable prefix is detected. + */ preserveSystemPrompt: boolean; + /** + * T05/C5 — authoritative system-prompt preservation intent: + * - `always`: never compress the system prompt. + * - `whenNoCache`: compress it only when there is no cache to protect + * (preserve when the provider caches or `cache_control` is present). This is the + * behaviour the legacy `preserveSystemPrompt: false` already had via the cache guard. + * - `never`: always compress the system prompt, even when it breaks a prompt cache. + * Optional/back-compat: absent → derived from the legacy boolean + * (`false → whenNoCache`, otherwise `always`). + */ + preserveSystemPromptMode?: PreserveSystemPromptMode; mcpDescriptionCompressionEnabled?: boolean; comboOverrides: Record; compressionComboId?: string | null; @@ -162,6 +178,8 @@ export interface CompressionConfig { fidelityGate?: FidelityGateConfig; /** Opt-in risk-gate pre-pass: shields sensitive spans from compression (default disabled). */ riskGate?: RiskGateConfig; + /** T02 — opt-in per-engine circuit-breaker for the stacked pipeline (default disabled). */ + pipelineCircuitBreaker?: PipelineCircuitBreakerConfig; cavemanConfig?: CavemanConfig; cavemanOutputMode?: CavemanOutputModeConfig; /** Phase 4A: selected output styles (supersedes cavemanOutputMode via a back-compat shim). */ @@ -289,6 +307,7 @@ export const DEFAULT_COMPRESSION_CONFIG: CompressionConfig = { autoTriggerTokens: 0, cacheMinutes: 5, preserveSystemPrompt: true, + preserveSystemPromptMode: "always", mcpDescriptionCompressionEnabled: true, comboOverrides: {}, compressionComboId: null, diff --git a/open-sse/services/grokTlsClient.ts b/open-sse/services/grokTlsClient.ts index cdc888f896..bd24b37a87 100644 --- a/open-sse/services/grokTlsClient.ts +++ b/open-sse/services/grokTlsClient.ts @@ -25,7 +25,7 @@ import { randomUUID } from "node:crypto"; let clientPromise: Promise | null = null; let exitHookInstalled = false; -const GROK_PROFILE = "chrome_149"; // closest Chrome profile to the UA we send +const GROK_PROFILE = "chrome_146"; // closest supported wreq-js profile (chrome_149 absent in 2.3.1, #5591) const DEFAULT_TIMEOUT_MS = Number.parseInt(process.env.OMNIROUTE_GROK_TLS_TIMEOUT_MS || "", 10) || 60_000; // Grace period added to the binding's wire-level timeout before our JS-level diff --git a/open-sse/services/model.ts b/open-sse/services/model.ts index 3150750a56..54cc6a3e2c 100644 --- a/open-sse/services/model.ts +++ b/open-sse/services/model.ts @@ -127,7 +127,13 @@ const CODEX_PREFERRED_UNPREFIXED_MODELS = new Set([ "gpt-5.5-medium", "gpt-5.5-low", ]); -const CODEX_PREFERRED_UNPREFIXED_MODEL_ALIASES = new Map([["gpt-5.5", "gpt-5.5-medium"]]); +// Intentionally empty: an unprefixed codex-preferred model keeps its BARE id when +// inferred to codex. #2877 established that baking a `-medium` effort suffix silently +// overrides a client `reasoning.effort` (the Codex executor reads the suffix as an +// explicit modelEffort). This map was dormant while bare `gpt-5.5` hit the OpenAI +// short-circuit; #5887 makes the codex block reachable for bare `gpt-5.5`, so the +// `gpt-5.5 → gpt-5.5-medium` entry is removed to preserve #2877's bare-id contract. +const CODEX_PREFERRED_UNPREFIXED_MODEL_ALIASES = new Map([]); export const CODEX_NATIVE_UNPREFIXED_MODELS = new Set(["codex-auto-review"]); interface ProviderConnectionLike { @@ -511,17 +517,11 @@ async function resolveModelByProviderInference(modelId: string, extendedContext: getPreferClaudeCodeForUnprefixedClaudeModels(), ]); - // Preserve historical behavior: OpenAI stays default when model exists there. - // Connection availability must not make unprefixed OpenAI models resolve to a - // different provider; callers can still force Codex with an explicit prefix. - if (providers.includes("openai")) { - return { - provider: "openai", - model: modelId, - extendedContext, - }; - } - + // Codex-only setups must keep auto-routing codex-preferred unprefixed models + // (e.g. `gpt-5.5`) to codex even after those ids were added to the OpenAI + // static catalog (#5887). This block is guarded by `!activeProviders.has("openai")`, + // so it must run BEFORE the OpenAI short-circuit below; users WITH an active + // OpenAI connection still fall through to the OpenAI default. if ( activeProviders?.has("codex") && !activeProviders.has("openai") && @@ -535,6 +535,17 @@ async function resolveModelByProviderInference(modelId: string, extendedContext: }; } + // Preserve historical behavior: OpenAI stays default when model exists there. + // Connection availability must not make unprefixed OpenAI models resolve to a + // different provider; callers can still force Codex with an explicit prefix. + if (providers.includes("openai")) { + return { + provider: "openai", + model: modelId, + extendedContext, + }; + } + // Fallback for newly released OpenAI-family model IDs that may not be in the local catalog yet. if (/^gpt-/i.test(modelId) || /^o1/i.test(modelId) || /^o3/i.test(modelId)) { return { diff --git a/open-sse/services/modelDeprecation.ts b/open-sse/services/modelDeprecation.ts index b38b76c36f..cf965abf34 100644 --- a/open-sse/services/modelDeprecation.ts +++ b/open-sse/services/modelDeprecation.ts @@ -60,20 +60,40 @@ const BUILT_IN_ALIASES: Record = { }; // ── Custom Aliases (persisted via Settings API) ───────────────────────────── -let _customAliases: Record = {}; +// +// Backed by globalThis so the singleton store is shared across the SEPARATE webpack +// module graphs Next.js builds for `instrumentation.ts` (boot-time hydration via +// applyRuntimeSettings → setCustomAliases) and the app-route `GET /api/settings/model-aliases`. +// A plain module-level `let` is DUPLICATED per graph, so startup hydration lands on the +// instrumentation graph's copy while the API route reads an empty copy — the exact +// symptom #5777 patched at the route layer. Migrating the store to globalThis fixes the +// root cause (both instances read/write one store), mirroring the #5312 pattern already +// applied to thinkingBudget.ts and backgroundTaskDetector.ts (and systemPrompt.ts #2470). +const CUSTOM_ALIASES_GLOBAL_KEY = "__omniroute_customAliases__"; +const _aliasStore = globalThis as unknown as Record< + string, + Record | undefined +>; + +function customAliases(): Record { + if (!_aliasStore[CUSTOM_ALIASES_GLOBAL_KEY]) { + _aliasStore[CUSTOM_ALIASES_GLOBAL_KEY] = {}; + } + return _aliasStore[CUSTOM_ALIASES_GLOBAL_KEY]!; +} /** * Set custom aliases (called from settings API or startup). */ export function setCustomAliases(aliases: Record): void { - _customAliases = { ...aliases }; + _aliasStore[CUSTOM_ALIASES_GLOBAL_KEY] = { ...aliases }; } /** * Get current custom aliases. */ export function getCustomAliases(): Record { - return { ..._customAliases }; + return { ...customAliases() }; } /** @@ -81,7 +101,7 @@ export function getCustomAliases(): Record { * Custom aliases take precedence over built-in. */ export function getAllAliases(): Record { - return { ...BUILT_IN_ALIASES, ..._customAliases }; + return { ...BUILT_IN_ALIASES, ...customAliases() }; } /** @@ -95,7 +115,8 @@ export function resolveModelAlias(modelId: string): string { if (!modelId) return modelId; // Check custom aliases first (higher priority) - if (_customAliases[modelId]) return _customAliases[modelId]; + const custom = customAliases(); + if (custom[modelId]) return custom[modelId]; // Then check built-in if (BUILT_IN_ALIASES[modelId]) return BUILT_IN_ALIASES[modelId]; @@ -129,15 +150,16 @@ export function isDeprecated(modelId: string): boolean { * Add a custom alias. */ export function addCustomAlias(from: string, to: string): void { - _customAliases[from] = to; + customAliases()[from] = to; } /** * Remove a custom alias. */ export function removeCustomAlias(from: string): boolean { - if (_customAliases[from]) { - delete _customAliases[from]; + const custom = customAliases(); + if (custom[from]) { + delete custom[from]; return true; } return false; diff --git a/open-sse/services/rateLimitManager.ts b/open-sse/services/rateLimitManager.ts index 6adfe42c0b..2a667028af 100644 --- a/open-sse/services/rateLimitManager.ts +++ b/open-sse/services/rateLimitManager.ts @@ -18,6 +18,12 @@ import { resolveResilienceSettings, type RequestQueueSettings, } from "../../src/lib/resilience/settings"; +import { + STANDARD_HEADERS, + ANTHROPIC_HEADERS, + parseResetTime, + toPlainHeaders, +} from "./rateLimitManager/headers"; interface LearnedLimitEntry { provider: string; @@ -598,101 +604,6 @@ export async function withRateLimit(provider, connectionId, model, fn, signal = } } -// ─── Header Parsing ────────────────────────────────────────────────────────── - -/** - * Standard headers used by most providers (OpenAI, Fireworks, etc.) - */ -const STANDARD_HEADERS = { - limit: "x-ratelimit-limit-requests", - remaining: "x-ratelimit-remaining-requests", - reset: "x-ratelimit-reset-requests", - limitTokens: "x-ratelimit-limit-tokens", - remainingTokens: "x-ratelimit-remaining-tokens", - resetTokens: "x-ratelimit-reset-tokens", - retryAfter: "retry-after", - overLimit: "x-ratelimit-over-limit", -}; - -/** - * Anthropic uses custom headers - */ -const ANTHROPIC_HEADERS = { - limit: "anthropic-ratelimit-requests-limit", - remaining: "anthropic-ratelimit-requests-remaining", - reset: "anthropic-ratelimit-requests-reset", - limitTokens: "anthropic-ratelimit-input-tokens-limit", - remainingTokens: "anthropic-ratelimit-input-tokens-remaining", - resetTokens: "anthropic-ratelimit-input-tokens-reset", - retryAfter: "retry-after", -}; - -/** - * Parse a reset time string into milliseconds. - * Formats: "1s", "1m", "1h", "1ms", "60", ISO date, Unix timestamp - */ -function parseResetTime(value) { - if (!value) return null; - - // Duration strings: "1s", "500ms", "1m30s" - const durationMatch = value.match(/^(?:(\d+)h)?(?:(\d+)m(?!s))?(?:(\d+)s)?(?:(\d+)ms)?$/); - if (durationMatch) { - const [, h, m, s, ms] = durationMatch; - return ( - (parseInt(h || 0) * 3600 + parseInt(m || 0) * 60 + parseInt(s || 0)) * 1000 + - parseInt(ms || 0) - ); - } - - // Pure number: assume seconds - const num = parseFloat(value); - if (!isNaN(num) && num > 0) { - // If it looks like a Unix timestamp (> year 2025) - if (num > 1700000000) { - return Math.max(0, num * 1000 - Date.now()); - } - return num * 1000; - } - - // ISO date string - try { - const date = new Date(value); - if (!isNaN(date.getTime())) { - return Math.max(0, date.getTime() - Date.now()); - } - } catch {} - - return null; -} - -function toPlainHeaders(headers: unknown): Record { - if (!headers) return {}; - const plain: Record = {}; - const obj = headers as Record; - if (typeof obj.forEach === "function") { - try { - (obj.forEach as (cb: (v: string, k: string) => void) => void)((v: string, k: string) => { - plain[k.toLowerCase()] = v; - }); - return plain; - } catch {} - } - if (typeof obj.entries === "function") { - try { - for (const [k, v] of (obj.entries as () => Iterable<[string, string]>)()) { - plain[k.toLowerCase()] = v; - } - return plain; - } catch {} - } - try { - for (const [k, v] of Object.entries(obj)) { - plain[k.toLowerCase()] = v == null ? "" : String(v); - } - } catch {} - return plain; -} - /** * Update rate limiter based on API response headers. * Called after every successful or failed response from a provider. diff --git a/open-sse/services/rateLimitManager/headers.ts b/open-sse/services/rateLimitManager/headers.ts new file mode 100644 index 0000000000..fac554cc68 --- /dev/null +++ b/open-sse/services/rateLimitManager/headers.ts @@ -0,0 +1,94 @@ +// ─── Header Parsing ────────────────────────────────────────────────────────── + +/** + * Standard headers used by most providers (OpenAI, Fireworks, etc.) + */ +export const STANDARD_HEADERS = { + limit: "x-ratelimit-limit-requests", + remaining: "x-ratelimit-remaining-requests", + reset: "x-ratelimit-reset-requests", + limitTokens: "x-ratelimit-limit-tokens", + remainingTokens: "x-ratelimit-remaining-tokens", + resetTokens: "x-ratelimit-reset-tokens", + retryAfter: "retry-after", + overLimit: "x-ratelimit-over-limit", +}; + +/** + * Anthropic uses custom headers + */ +export const ANTHROPIC_HEADERS = { + limit: "anthropic-ratelimit-requests-limit", + remaining: "anthropic-ratelimit-requests-remaining", + reset: "anthropic-ratelimit-requests-reset", + limitTokens: "anthropic-ratelimit-input-tokens-limit", + remainingTokens: "anthropic-ratelimit-input-tokens-remaining", + resetTokens: "anthropic-ratelimit-input-tokens-reset", + retryAfter: "retry-after", +}; + +/** + * Parse a reset time string into milliseconds. + * Formats: "1s", "1m", "1h", "1ms", "60", ISO date, Unix timestamp + */ +export function parseResetTime(value) { + if (!value) return null; + + // Duration strings: "1s", "500ms", "1m30s" + const durationMatch = value.match(/^(?:(\d+)h)?(?:(\d+)m(?!s))?(?:(\d+)s)?(?:(\d+)ms)?$/); + if (durationMatch) { + const [, h, m, s, ms] = durationMatch; + return ( + (parseInt(h || 0) * 3600 + parseInt(m || 0) * 60 + parseInt(s || 0)) * 1000 + + parseInt(ms || 0) + ); + } + + // Pure number: assume seconds + const num = parseFloat(value); + if (!isNaN(num) && num > 0) { + // If it looks like a Unix timestamp (> year 2025) + if (num > 1700000000) { + return Math.max(0, num * 1000 - Date.now()); + } + return num * 1000; + } + + // ISO date string + try { + const date = new Date(value); + if (!isNaN(date.getTime())) { + return Math.max(0, date.getTime() - Date.now()); + } + } catch {} + + return null; +} + +export function toPlainHeaders(headers: unknown): Record { + if (!headers) return {}; + const plain: Record = {}; + const obj = headers as Record; + if (typeof obj.forEach === "function") { + try { + (obj.forEach as (cb: (v: string, k: string) => void) => void)((v: string, k: string) => { + plain[k.toLowerCase()] = v; + }); + return plain; + } catch {} + } + if (typeof obj.entries === "function") { + try { + for (const [k, v] of (obj.entries as () => Iterable<[string, string]>)()) { + plain[k.toLowerCase()] = v; + } + return plain; + } catch {} + } + try { + for (const [k, v] of Object.entries(obj)) { + plain[k.toLowerCase()] = v == null ? "" : String(v); + } + } catch {} + return plain; +} diff --git a/open-sse/services/roleNormalizer.ts b/open-sse/services/roleNormalizer.ts index 7f63d2dd2f..b7190a67a6 100644 --- a/open-sse/services/roleNormalizer.ts +++ b/open-sse/services/roleNormalizer.ts @@ -49,14 +49,30 @@ function defaultPreserveDeveloperForProvider(provider: string): boolean { /** * Models that are known to reject the `system` role regardless of provider. - * Uses prefix matching (e.g., "glm-" matches "glm-4.7", "glm-4.5", etc.) + * Uses prefix matching (e.g., "ernie-" matches "ernie-4.0"). */ const MODELS_WITHOUT_SYSTEM_ROLE = [ - "glm-", // ZhipuAI GLM models (prefix: glm-5.1, glm-4.7, etc.) - "glm", // Exact match for model id "glm" (e.g., Pollinations) "ernie-", // Baidu ERNIE models ]; +/** + * ZhipuAI GLM rejects the `system` role EXCEPT generation > 5.0: per z.ai docs, + * GLM 5.1 / 5.2 (and newer) accept it, so their system prompt must NOT be folded + * into the first user turn (#5610). Everything else GLM — bare "glm" (Pollinations), + * the 4.x family, and the 5.0 generation — still needs the fold. The version is read + * from "glm-." or the Fireworks "glm-5p1" point alias. + */ +function isGlmWithoutSystemRole(modelLower: string): boolean { + if (!modelLower.startsWith("glm")) return false; + const match = modelLower.match(/glm-?(\d+)(?:[.p](\d+))?/); + if (match) { + const major = Number(match[1]); + const minor = match[2] ? Number(match[2]) : 0; + if (major > 5 || (major === 5 && minor >= 1)) return false; + } + return true; +} + const PROVIDER_SCOPED_MODELS_WITHOUT_SYSTEM_ROLE: Record = { // ZenMux exposes Z.AI GLM through OpenAI-compatible model ids such as // "z-ai/glm-5.2". Z.AI rejects compressed histories that start with a @@ -106,6 +122,8 @@ function supportsSystemRole(provider: string, model: string): boolean { if (pattern.test(modelLower)) return false; } + if (isGlmWithoutSystemRole(modelLower)) return false; + for (const prefix of MODELS_WITHOUT_SYSTEM_ROLE) { if (modelLower.startsWith(prefix)) return false; } diff --git a/open-sse/services/systemTransforms.ts b/open-sse/services/systemTransforms.ts index a647a6c967..09eb210887 100644 --- a/open-sse/services/systemTransforms.ts +++ b/open-sse/services/systemTransforms.ts @@ -449,7 +449,23 @@ function resolveProviderConfig( // Runtime singleton. // ──────────────────────────────────────────────────────────────────────────── -let _systemTransformsConfig: SystemTransformsConfig = DEFAULT_SYSTEM_TRANSFORMS_CONFIG; +// Backed by globalThis so the config is shared across the SEPARATE webpack module +// graphs Next.js builds for `instrumentation.ts` (boot-time hydration via +// applyRuntimeSettings → setSystemTransformsConfig) and the app-route / open-sse +// executors (per-request reads in base.ts / claudeCodeCompatible.ts). A module-local +// `let` is duplicated per graph, so a Settings-UI override applied at boot never +// reaches the request path (the #5312-class module-graph bug; the protective +// compiled default still runs, but operator customizations were silently dropped). +// Mirrors systemPrompt.ts (#2470) and thinkingBudget.ts (#5312). +const GLOBAL_KEY = "__omniroute_systemTransforms_config__"; +const _store = globalThis as unknown as Record; + +function getStore(): SystemTransformsConfig { + if (!_store[GLOBAL_KEY]) { + _store[GLOBAL_KEY] = DEFAULT_SYSTEM_TRANSFORMS_CONFIG; + } + return _store[GLOBAL_KEY]!; +} /** * Replace the active system-transforms config. Called from @@ -460,14 +476,14 @@ let _systemTransformsConfig: SystemTransformsConfig = DEFAULT_SYSTEM_TRANSFORMS_ */ export function setSystemTransformsConfig(input: unknown): void { if (!input || typeof input !== "object") { - _systemTransformsConfig = DEFAULT_SYSTEM_TRANSFORMS_CONFIG; + _store[GLOBAL_KEY] = DEFAULT_SYSTEM_TRANSFORMS_CONFIG; return; } const candidate = input as Record; // Legacy shape: { enabled, pipeline } → migrate to per-provider map. if ("pipeline" in candidate && Array.isArray(candidate.pipeline)) { - _systemTransformsConfig = { + _store[GLOBAL_KEY] = { providers: { ...DEFAULT_SYSTEM_TRANSFORMS_CONFIG.providers, [PROVIDER_CC_BRIDGE]: { @@ -500,17 +516,17 @@ export function setSystemTransformsConfig(input: unknown): void { next.providers[providerId] = providerDefault; } } - _systemTransformsConfig = next; + _store[GLOBAL_KEY] = next; return; } - _systemTransformsConfig = DEFAULT_SYSTEM_TRANSFORMS_CONFIG; + _store[GLOBAL_KEY] = DEFAULT_SYSTEM_TRANSFORMS_CONFIG; } export function getSystemTransformsConfig(): SystemTransformsConfig { - return _systemTransformsConfig; + return getStore(); } export function resetSystemTransformsConfig(): void { - _systemTransformsConfig = DEFAULT_SYSTEM_TRANSFORMS_CONFIG; + _store[GLOBAL_KEY] = DEFAULT_SYSTEM_TRANSFORMS_CONFIG; } diff --git a/open-sse/services/thinkingBudget.ts b/open-sse/services/thinkingBudget.ts index a16f3b432d..812c00aa35 100644 --- a/open-sse/services/thinkingBudget.ts +++ b/open-sse/services/thinkingBudget.ts @@ -56,8 +56,24 @@ export const DEFAULT_THINKING_CONFIG = { effortLevel: "medium", } satisfies ThinkingBudgetConfig; -// In-memory config (loaded from DB on startup, or default) -let _config: ThinkingBudgetConfig = { ...DEFAULT_THINKING_CONFIG }; +// In-memory config (loaded from DB on startup, or default). +// +// Backed by globalThis so the singleton is shared across the SEPARATE webpack +// module graphs Next.js builds for `instrumentation.ts` (boot-time hydration via +// hydrateThinkingBudgetConfig) and the app-route / open-sse executors (per-request +// reads in base.ts). A plain module-level `let` is DUPLICATED per graph, so the +// boot hydration would land on the instrumentation graph's copy and never reach +// base.ts — exactly the #5312 fix-A break proven on the VPS. Mirrors the same +// globalThis pattern systemPrompt.ts already uses for the Global System Prompt (#2470). +const GLOBAL_KEY = "__omniroute_thinkingBudget_config__"; +const _store = globalThis as unknown as Record; + +function getConfig(): ThinkingBudgetConfig { + if (!_store[GLOBAL_KEY]) { + _store[GLOBAL_KEY] = { ...DEFAULT_THINKING_CONFIG }; + } + return _store[GLOBAL_KEY]!; +} function toRecord(value: unknown): JsonRecord { return value && typeof value === "object" && !Array.isArray(value) ? (value as JsonRecord) : {}; @@ -72,14 +88,14 @@ function getStringField(record: JsonRecord, key: string): string { * Set the thinking budget config (called from settings API or startup) */ export function setThinkingBudgetConfig(config: Partial) { - _config = { ...DEFAULT_THINKING_CONFIG, ...config }; + _store[GLOBAL_KEY] = { ...DEFAULT_THINKING_CONFIG, ...config }; } /** * Get current thinking budget config */ export function getThinkingBudgetConfig() { - return { ..._config }; + return { ...getConfig() }; } /** @@ -195,7 +211,7 @@ export function applyThinkingBudget( body: unknown, config: Partial | null = null ) { - const cfg = config || _config; + const cfg = config || getConfig(); if (!body || typeof body !== "object") return body; // Early exit: strip ALL reasoning/thinking params for models that don't support them. diff --git a/open-sse/services/tokenExtractionConfig.ts b/open-sse/services/tokenExtractionConfig.ts index f82804e2e0..7d141bacc1 100644 --- a/open-sse/services/tokenExtractionConfig.ts +++ b/open-sse/services/tokenExtractionConfig.ts @@ -193,14 +193,13 @@ const RAW_CONFIGS: TokenExtractionConfig[] = [ config( "kimi-web", "Kimi (Moonshot)", - "https://kimi.moonshot.cn/", - "https://kimi.moonshot.cn", + "https://www.kimi.com/", + "https://www.kimi.com", [ - { type: "cookie", name: "kimi_token", domain: ".kimi.moonshot.cn" }, - { type: "localStorage", key: "kimi_token" }, + { type: "cookie", name: "kimi-auth", domain: ".kimi.com" }, ], - "Log in to Kimi at kimi.moonshot.cn via phone/WeChat. The session token will be extracted.", - { cookieDomain: ".kimi.moonshot.cn" } + "Log in to Kimi at www.kimi.com (international). The kimi-auth JWT cookie will be extracted.", + { cookieDomain: ".kimi.com" } ), // ── Blackbox Web ────────────────────────────────────────── diff --git a/open-sse/services/toolLimitDetector.ts b/open-sse/services/toolLimitDetector.ts index 035003004a..48969fcb12 100644 --- a/open-sse/services/toolLimitDetector.ts +++ b/open-sse/services/toolLimitDetector.ts @@ -4,6 +4,10 @@ const DETECTED_LIMITS = new Map(); const TTL_MS = 24 * 60 * 60 * 1000; const DEFAULT_LIMIT = MAX_TOOLS_LIMIT; +const PROVIDER_TOOL_LIMITS: Record = { + "grok-cli": 200, +}; + const _detectedLimitsSweep = setInterval(() => { const now = Date.now(); for (const [key, entry] of DETECTED_LIMITS) { @@ -15,6 +19,10 @@ if (typeof _detectedLimitsSweep === "object" && "unref" in _detectedLimitsSweep) } export function getEffectiveToolLimit(provider: string): number { + const proactiveLimit = PROVIDER_TOOL_LIMITS[provider]; + if (proactiveLimit !== undefined) { + return proactiveLimit; + } const cached = DETECTED_LIMITS.get(provider); if (cached && Date.now() - cached.timestamp < TTL_MS) { return cached.limit; @@ -33,6 +41,7 @@ const TOOL_LIMIT_PATTERNS = [ /'tools':\s*maximum\s+number\s+of\s+items\s+is\s+(\d+)/i, /Maximum\s+number\s+of\s+tools\s+(?:allowed\s+)?(?:is\s+)?(\d+)/i, /Too\s+many\s+tools\.?\s*(?:Maximum\s+)?(\d+)/i, + /\d+\s+tools\s+have\s+been\s+provided\s+but\s+(?:the\s+)?maximum\s+is\s+(\d+)/i, /tool.*limit.*(\d+)/i, /tools.*exceeded.*(\d+)/i, ]; diff --git a/open-sse/services/usage.ts b/open-sse/services/usage.ts index 3aae5f02c6..a65d21daf8 100644 --- a/open-sse/services/usage.ts +++ b/open-sse/services/usage.ts @@ -2,18 +2,13 @@ * Usage Fetcher - Get usage data from provider APIs */ -import { buildCodexUsageQuotas } from "./codexUsageQuotas.ts"; -import { getGlmQuotaUrl } from "../config/glmProvider.ts"; import { getGitHubCopilotInternalUserHeaders } from "../config/providerHeaderProfiles.ts"; -import { safePercentage } from "@/shared/utils/formatting"; import { getDbInstance } from "@/lib/db/core"; import { fetchBailianQuota, type BailianTripleWindowQuota } from "./bailianQuotaFetcher.ts"; import { fetchDeepseekQuota, type DeepseekQuota } from "./deepseekQuotaFetcher.ts"; import { fetchOpencodeQuota, type OpencodeTripleWindowQuota } from "./opencodeQuotaFetcher.ts"; import { getOllamaCloudUsage, getOpenCodeGoUsage } from "./opencodeOllamaUsage.ts"; import { getCodeBuddyCnUsage } from "./usage/codebuddy-cn.ts"; -import { CLAUDE_CODE_VERSION, fetchClaudeBootstrap } from "../executors/claudeIdentity.ts"; -import { isClaudeOauthUsageCoolingDown, markClaudeOauthUsage429 } from "./claudeUsageCooldown.ts"; import { extractCodeAssistOnboardTierId, extractCodeAssistSubscriptionTier, @@ -55,48 +50,21 @@ import { mapCodeAssistTierIdToLabel, mapSubscriptionTierStringToPlanLabel, } from "./usage/antigravity.ts"; +import { getCursorUsage } from "./usage/cursor.ts"; +import { getKimiUsage } from "./usage/kimi.ts"; +import { getCodexUsage } from "./usage/codex.ts"; +import { getClaudeUsage, getClaudePlanLabel } from "./usage/claude.ts"; +import { getKiroUsage, buildKiroUsageResult, discoverKiroProfileArn } from "./usage/kiro.ts"; +// Re-exported para os testes kiro-* (importam de services/usage). +export { buildKiroUsageResult, discoverKiroProfileArn } from "./usage/kiro.ts"; // Quota / usage upstream URLs (overridable for testing or relays). const CROF_USAGE_URL = process.env.OMNIROUTE_CROF_USAGE_URL ?? "https://crof.ai/usage_api/"; -const CODEWHISPERER_BASE_URL = - process.env.OMNIROUTE_CODEWHISPERER_BASE_URL ?? "https://codewhisperer.us-east-1.amazonaws.com"; - -// Codex (OpenAI) API config -const CODEX_CONFIG = { - usageUrl: "https://chatgpt.com/backend-api/wham/usage", -}; - -// Claude API config -const CLAUDE_CONFIG = { - oauthUsageUrl: "https://api.anthropic.com/api/oauth/usage", - usageUrl: "https://api.anthropic.com/v1/organizations/{org_id}/usage", - settingsUrl: "https://api.anthropic.com/v1/settings", - apiVersion: "2023-06-01", -}; - -// Kimi Coding API config -const KIMI_CONFIG = { - baseUrl: "https://api.kimi.com/coding/v1", - usageUrl: "https://api.kimi.com/coding/v1/usages", - apiVersion: "2023-06-01", -}; const NANOGPT_CONFIG = { usageUrl: "https://nano-gpt.com/api/subscription/v1/usage", }; -// Cursor dashboard usage API config -// The endpoint that powers https://cursor.com/dashboard/spending. Validates the WorkOS -// session via the WorkosCursorSessionToken cookie (format: `${userId}::${jwt}`) and -// rejects requests without a matching Origin/Referer (Invalid origin for state-changing request). -const CURSOR_USAGE_CONFIG = { - usageUrl: "https://cursor.com/api/dashboard/get-current-period-usage", - origin: "https://cursor.com", - referer: "https://cursor.com/dashboard/spending", - userAgent: - "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/149.0.0.0 Safari/537.36", -}; - type JsonRecord = Record; type UsageProviderConnection = JsonRecord & { id?: string; @@ -114,57 +82,6 @@ function shouldDisplayGitHubQuota(quota: UsageQuota | null): quota is UsageQuota return quota.total > 0 || quota.remainingPercentage !== undefined; } -function isKiroOverageEnabled(data: JsonRecord): boolean { - const overageConfiguration = toRecord(data.overageConfiguration); - const overageStatus = String(overageConfiguration.overageStatus || "") - .trim() - .toUpperCase(); - - return ( - overageStatus === "ENABLED" || - data.overageEnabled === true || - overageConfiguration.overageEnabled === true - ); -} - -function buildKiroQuota( - used: number, - total: number, - resetAt: string | null, - overageEnabled: boolean -): UsageQuota { - const remaining = total - used; - - if (!overageEnabled) { - return { used, total, remaining, resetAt, unlimited: false }; - } - - return { - used, - total, - remaining, - remainingPercentage: 100, - resetAt, - unlimited: true, - }; -} - -function getClaudePlanLabel(...candidates: Array): string | null { - for (const candidate of candidates) { - if (typeof candidate !== "string") continue; - const trimmed = candidate.trim(); - if ( - !trimmed || - trimmed.toLowerCase() === "claude code" || - trimmed.toLowerCase() === "unknown" - ) { - continue; - } - return trimmed; - } - return null; -} - // CrofAI surfaces a tiny endpoint with two signals: // GET https://crof.ai/usage_api/ → { usable_requests: number|null, credits: number } // `usable_requests` is the daily request bucket on a subscription plan; `null` @@ -545,143 +462,6 @@ async function getNanoGptUsage(apiKey: string) { } } -/** - * Decode the `sub` claim of a Cursor JWT (the WorkOS user id). - * Returns null if the token is not a parseable JWT. - */ -function decodeCursorJwtSub(token: string): string | null { - if (!token || typeof token !== "string") return null; - const parts = token.split("."); - if (parts.length !== 3) return null; - try { - let payload = parts[1].replace(/-/g, "+").replace(/_/g, "/"); - while (payload.length % 4 !== 0) payload += "="; - const decoded = JSON.parse(Buffer.from(payload, "base64").toString("utf8")); - const sub = decoded?.sub; - return typeof sub === "string" && sub.length > 0 ? sub : null; - } catch { - return null; - } -} - -/** - * Cursor Pro Plan Usage - * Fetches current-billing-cycle spend from the cursor.com dashboard API and exposes three - * windows that mirror the cursor.com/dashboard/spending UI: Total / Auto + Composer / API. - */ -async function getCursorUsage(accessToken: string, providerSpecificData?: unknown) { - if (!accessToken) { - return { message: "Cursor access token missing. Re-import the connection from Cursor IDE." }; - } - - const storedUserId = (() => { - const raw = toRecord(providerSpecificData).userId; - return typeof raw === "string" && raw.length > 0 ? raw : null; - })(); - const userId = storedUserId || decodeCursorJwtSub(accessToken); - - if (!userId) { - return { - message: "Cursor token missing user id. Re-import the connection from Cursor IDE.", - }; - } - - try { - const response = await fetch(CURSOR_USAGE_CONFIG.usageUrl, { - method: "POST", - redirect: "manual", - headers: { - Cookie: `WorkosCursorSessionToken=${userId}::${accessToken}`, - Origin: CURSOR_USAGE_CONFIG.origin, - Referer: CURSOR_USAGE_CONFIG.referer, - "Content-Type": "application/json", - Accept: "application/json", - "User-Agent": CURSOR_USAGE_CONFIG.userAgent, - }, - body: "{}", - }); - - // 3xx redirect to WorkOS authkit means the session cookie was rejected. - if (response.status >= 300 && response.status < 400) { - return { - plan: "Cursor", - message: "Cursor session expired. Re-import the token from Cursor IDE.", - }; - } - - if (!response.ok) { - const errorText = (await response.text()).slice(0, 200); - if (response.status === 401 || response.status === 403) { - return { - plan: "Cursor", - message: "Cursor session unauthorized. Re-import the token from Cursor IDE.", - }; - } - return { - plan: "Cursor", - message: `Cursor usage endpoint error (${response.status}): ${errorText}`, - }; - } - - const data = toRecord(await response.json()); - const planUsage = toRecord(data.planUsage); - - if (Object.keys(planUsage).length === 0) { - return { - plan: "Cursor", - message: "Cursor connected. No active plan usage returned.", - }; - } - - const limitCents = Math.max(0, toNumber(planUsage.limit, 0)); - const totalSpendCents = Math.max(0, toNumber(planUsage.totalSpend, 0)); - const autoPercentUsed = clampPercentage(toNumber(planUsage.autoPercentUsed, 0)); - const apiPercentUsed = clampPercentage(toNumber(planUsage.apiPercentUsed, 0)); - const totalPercentUsed = clampPercentage(toNumber(planUsage.totalPercentUsed, 0)); - - // billingCycleEnd is a numeric-string in ms; coerce so parseResetTime sees a number. - const billingCycleEndMs = toNumber(data.billingCycleEnd, 0); - const resetAt = billingCycleEndMs > 0 ? parseResetTime(billingCycleEndMs) : null; - - // Convert cents → dollars rounded to 2 decimal places. - const toDollars = (cents: number) => Math.round(cents) / 100; - - const limitDollars = toDollars(limitCents); - const buildWindow = (percentUsed: number, usedCentsOverride?: number): UsageQuota => { - const usedCents = - typeof usedCentsOverride === "number" - ? usedCentsOverride - : Math.round((limitCents * percentUsed) / 100); - const used = toDollars(Math.min(usedCents, limitCents)); - const remaining = toDollars(Math.max(limitCents - Math.min(usedCents, limitCents), 0)); - return { - used, - total: limitDollars, - remaining, - remainingPercentage: clampPercentage(100 - percentUsed), - resetAt, - unlimited: false, - }; - }; - - const quotas: Record = { - Total: buildWindow(totalPercentUsed, totalSpendCents), - "Auto + Composer": buildWindow(autoPercentUsed), - API: buildWindow(apiPercentUsed), - }; - - return { - plan: "Cursor Pro", - quotas, - }; - } catch (error) { - return { - plan: "Cursor", - message: `Cursor connected. Unable to fetch usage: ${(error as Error).message}`, - }; - } -} - /** * Single source of truth for which providers have a `getUsageForProvider` * implementation. Consumers like `genericQuotaFetcher.ts` reference this so @@ -1011,421 +791,6 @@ function inferGitHubPlanName(data: JsonRecord, premiumQuota: UsageQuota | null): return "GitHub Copilot"; } -/** - * Claude Usage - Try to fetch from Anthropic API - */ -async function getClaudeUsage(accessToken?: string) { - if (!accessToken) { - return { message: "Claude connected. Access token not available.", bootstrap: null }; - } - - // Refresh bootstrap in parallel; best-effort, failure non-fatal. - const bootstrapPromise = fetchClaudeBootstrap(accessToken).catch(() => null); - // Skip OAuth usage call while this token is cooling down from a recent 429 - // (chat with the same token still works — only the quota endpoint is throttled). - if (isClaudeOauthUsageCoolingDown(accessToken)) { - const legacy = await getClaudeUsageLegacy(accessToken); - return { ...legacy, bootstrap: await bootstrapPromise }; - } - try { - // Real CLI uses axios here, not Stainless — UA is `claude-code/` - // (not `claude-cli/...`) and the shape is simpler than /v1/messages. - const ctrl = new AbortController(); - const timer = setTimeout(() => ctrl.abort(), 10_000); - let oauthResponse; - try { - oauthResponse = await fetch(CLAUDE_CONFIG.oauthUsageUrl, { - method: "GET", - headers: { - Accept: "application/json, text/plain, */*", - "Accept-Encoding": "gzip, compress, deflate, br", - Authorization: `Bearer ${accessToken}`, - "Content-Type": "application/json", - "User-Agent": `claude-code/${CLAUDE_CODE_VERSION}`, - "anthropic-beta": "oauth-2025-04-20", - }, - signal: ctrl.signal, - }); - } finally { - clearTimeout(timer); - } - - if (oauthResponse.ok) { - const data = toRecord(await oauthResponse.json()); - const quotas: Record = {}; - - // utilization = percentage USED (e.g., 90 means 90% used, 10% remaining) - // Confirmed via user report #299: Claude.ai shows 87% used = OmniRoute must show 13% remaining. - const hasUtilization = (window: JsonRecord) => - window && typeof window === "object" && safePercentage(window.utilization) !== undefined; - - const createQuotaObject = (window: JsonRecord) => { - const used = safePercentage(window.utilization) as number; // utilization = % used - const remaining = Math.max(0, 100 - used); - return { - used, - total: 100, - remaining, - resetAt: parseResetTime(window.resets_at), - remainingPercentage: remaining, - unlimited: false, - }; - }; - - const fiveHour = toRecord(data.five_hour); - if (hasUtilization(fiveHour)) { - quotas["session (5h)"] = createQuotaObject(fiveHour); - } - - const sevenDay = toRecord(data.seven_day); - if (hasUtilization(sevenDay)) { - quotas["weekly (7d)"] = createQuotaObject(sevenDay); - } - - // Map Anthropic's internal codenames (e.g., omelette → Designer) for display. - const MODEL_DISPLAY_NAMES: Record = { - omelette: "designer", - }; - for (const [key, value] of Object.entries(data)) { - const valueRecord = toRecord(value); - if (key.startsWith("seven_day_") && key !== "seven_day" && hasUtilization(valueRecord)) { - const codename = key.replace("seven_day_", ""); - const modelName = MODEL_DISPLAY_NAMES[codename] || codename; - quotas[`weekly ${modelName} (7d)`] = createQuotaObject(valueRecord); - } - } - - const bootstrap = await bootstrapPromise; - const plan = - getClaudePlanLabel( - typeof data.tier === "string" ? data.tier : null, - typeof data.plan === "string" ? data.plan : null, - typeof data.subscription_type === "string" ? data.subscription_type : null, - bootstrap?.organization_rate_limit_tier - ) ?? undefined; - - return { - ...(plan ? { plan } : {}), - quotas, - extraUsage: data.extra_usage ?? null, - bootstrap, - }; - } - - // Cool down OAuth usage polling after a 429 (quota endpoint only — chat is unaffected). - if (oauthResponse.status === 429) { - markClaudeOauthUsage429(accessToken); - } - - // Fallback: OAuth endpoint returned non-OK, try legacy settings/org endpoint - console.warn( - `[Claude Usage] OAuth endpoint returned ${oauthResponse.status}, falling back to legacy` - ); - const legacy = await getClaudeUsageLegacy(accessToken); - return { ...legacy, bootstrap: await bootstrapPromise }; - } catch (error) { - return { - message: `Claude connected. Unable to fetch usage: ${(error as Error).message}`, - bootstrap: await bootstrapPromise, - }; - } -} - -/** - * Legacy Claude usage fetcher for API key / org admin users. - * Uses /v1/settings + /v1/organizations/{org_id}/usage endpoints. - */ -async function getClaudeUsageLegacy(accessToken?: string) { - try { - const settingsResponse = await fetch(CLAUDE_CONFIG.settingsUrl, { - method: "GET", - headers: { - Authorization: `Bearer ${accessToken}`, - "anthropic-version": CLAUDE_CONFIG.apiVersion, - }, - }); - - if (settingsResponse.ok) { - const settings = toRecord(await settingsResponse.json()); - - const organizationId = - typeof settings.organization_id === "string" ? settings.organization_id : ""; - if (organizationId) { - const usageResponse = await fetch( - CLAUDE_CONFIG.usageUrl.replace("{org_id}", organizationId), - { - method: "GET", - headers: { - Authorization: `Bearer ${accessToken}`, - "anthropic-version": CLAUDE_CONFIG.apiVersion, - }, - } - ); - - if (usageResponse.ok) { - const usage = await usageResponse.json(); - return { - plan: settings.plan || "Unknown", - organization: settings.organization_name, - quotas: usage, - }; - } - } - - return { - plan: settings.plan || "Unknown", - organization: settings.organization_name, - message: "Claude connected. Usage details require admin access.", - }; - } - - return { message: "Claude connected. Usage API requires admin permissions." }; - } catch (error) { - return { message: `Claude connected. Unable to fetch usage: ${(error as Error).message}` }; - } -} - -/** - * Codex (OpenAI) Usage - Fetch from ChatGPT backend API - * IMPORTANT: Uses persisted workspaceId from OAuth to ensure correct workspace binding. - * No fallback to other workspaces - strict binding to user's selected workspace. - */ -async function getCodexUsage( - accessToken?: string, - providerSpecificData: Record = {} -) { - try { - // Use persisted workspace ID from OAuth - NO FALLBACK - const accountId = - typeof providerSpecificData.workspaceId === "string" - ? providerSpecificData.workspaceId - : null; - - const headers: Record = { - Authorization: `Bearer ${accessToken}`, - "Content-Type": "application/json", - Accept: "application/json", - }; - if (accountId) { - headers["chatgpt-account-id"] = accountId; - } - - const response = await fetch(CODEX_CONFIG.usageUrl, { - method: "GET", - headers, - }); - - if (!response.ok) { - if (response.status === 401 || response.status === 403) { - return { - message: `Codex token expired or access denied. Please re-authenticate the connection.`, - }; - } - throw new Error(`Codex API error: ${response.status}`); - } - - const data = await response.json(); - - const { rateLimit, quotas } = buildCodexUsageQuotas(data); - - return { - plan: String(getFieldValue(data, "plan_type", "planType") || "unknown"), - limitReached: Boolean(getFieldValue(rateLimit, "limit_reached", "limitReached")), - quotas, - }; - } catch (error) { - return { message: `Failed to fetch Codex usage: ${(error as Error).message}` }; - } -} - -/** - * Build the Kiro usage result from a GetUsageLimits response. When the account returns no - * usage breakdown (some AWS IAM / Builder ID accounts don't expose per-resource quota via - * GetUsageLimits), return an informative message instead of empty `quotas:{}` — otherwise the - * dashboard renders a blank quota card with no explanation (#3506). Exported for testing. - */ -export function buildKiroUsageResult( - data: JsonRecord -): { plan: string; quotas: Record } | { message: string } { - const usageList = Array.isArray(data.usageBreakdownList) ? data.usageBreakdownList : []; - const quotaInfo: Record = {}; - const resetAt = parseResetTime(data.nextDateReset || data.resetDate); - const overageEnabled = isKiroOverageEnabled(data); - - usageList.forEach((breakdownValue: unknown) => { - const breakdown = toRecord(breakdownValue); - const resourceType = - typeof breakdown.resourceType === "string" ? breakdown.resourceType.toLowerCase() : "unknown"; - const used = toNumber(breakdown.currentUsageWithPrecision, 0); - const total = toNumber(breakdown.usageLimitWithPrecision, 0); - - quotaInfo[resourceType] = buildKiroQuota(used, total, resetAt, overageEnabled); - - const freeTrialInfo = toRecord(breakdown.freeTrialInfo); - if (Object.keys(freeTrialInfo).length > 0) { - const freeUsed = toNumber(freeTrialInfo.currentUsageWithPrecision, 0); - const freeTotal = toNumber(freeTrialInfo.usageLimitWithPrecision, 0); - quotaInfo[`${resourceType}_freetrial`] = buildKiroQuota( - freeUsed, - freeTotal, - resetAt, - overageEnabled - ); - } - }); - - if (Object.keys(quotaInfo).length === 0) { - return { - message: - "Kiro connected, but the account returned no usage breakdown. Some AWS IAM / Builder ID accounts don't expose per-resource quota via GetUsageLimits.", - }; - } - - return { - plan: String(toRecord(data.subscriptionInfo).subscriptionTitle || "").trim() || "Kiro", - quotas: quotaInfo, - }; -} - -/** - * Discover a Kiro/CodeWhisperer profile ARN for an account that didn't persist one (common for - * AWS IAM Identity Center logins and kiro-cli imports). Calls ListAvailableProfiles on the - * region-matched endpoint and prefers a profile whose ARN is in the same region. Returns - * undefined when no profile is available (e.g. the org/token has no Kiro entitlement). - * Exported for testing. - */ -export async function discoverKiroProfileArn( - accessToken: string, - usageBaseUrl: string, - region: string -): Promise { - try { - const response = await fetch(usageBaseUrl, { - method: "POST", - headers: { - Authorization: `Bearer ${accessToken}`, - "Content-Type": "application/x-amz-json-1.0", - "x-amz-target": "AmazonCodeWhispererService.ListAvailableProfiles", - Accept: "application/json", - }, - body: JSON.stringify({ maxResults: 10 }), - // Don't let a hung profile lookup block the usage/quota refresh indefinitely. - signal: AbortSignal.timeout(10000), - }); - if (!response.ok) return undefined; - - const data = toRecord(await response.json()); - const profiles = Array.isArray(data.profiles) ? data.profiles : []; - const normalizedRegion = region.toLowerCase(); - const matched = - profiles.find((profile: unknown) => { - const arn = toRecord(profile).arn; - return typeof arn === "string" && arn.toLowerCase().includes(`:${normalizedRegion}:`); - }) || profiles[0]; - const arn = toRecord(matched).arn; - return typeof arn === "string" && arn.length > 0 ? arn : undefined; - } catch { - return undefined; - } -} - -/** - * Kiro (AWS CodeWhisperer) Usage - */ -async function getKiroUsage(accessToken?: string, providerSpecificData?: JsonRecord) { - try { - let profileArn = - typeof providerSpecificData?.profileArn === "string" - ? providerSpecificData.profileArn - : undefined; - - // Enterprise IAM Identity Center accounts are region-bound: the profileArn, token and - // endpoint must all match the region. Derive the region from the stored region (preferred) - // or the profileArn, then route to the regional Amazon Q endpoint (us-east-1 keeps the - // legacy codewhisperer host; codewhisperer.{region} does not resolve for other regions). - const regionFromArn = profileArn - ? profileArn.toLowerCase().match(/^arn:aws:codewhisperer:([a-z0-9-]+):/)?.[1] - : undefined; - const region = - (typeof providerSpecificData?.region === "string" && - providerSpecificData.region.trim().toLowerCase()) || - regionFromArn || - "us-east-1"; - const usageBaseUrl = - region === "us-east-1" ? CODEWHISPERER_BASE_URL : `https://q.${region}.amazonaws.com`; - - // IAM Identity Center logins and kiro-cli imports frequently don't persist a profileArn, which - // previously caused the quota card to show nothing ("0 used"). Discover it on demand from - // ListAvailableProfiles (region-matched) so usage still resolves for those accounts. - if (!profileArn && accessToken) { - profileArn = await discoverKiroProfileArn(accessToken, usageBaseUrl, region); - } - - if (!profileArn) { - return { message: "Kiro connected. Profile ARN not available for quota tracking." }; - } - - // Kiro uses AWS CodeWhisperer GetUsageLimits API - const payload = { - origin: "AI_EDITOR", - profileArn: profileArn, - resourceType: "AGENTIC_REQUEST", - }; - - const response = await fetch(usageBaseUrl, { - method: "POST", - headers: { - Authorization: `Bearer ${accessToken}`, - "Content-Type": "application/x-amz-json-1.0", - "x-amz-target": "AmazonCodeWhispererService.GetUsageLimits", - Accept: "application/json", - }, - body: JSON.stringify(payload), - }); - - if (!response.ok) { - // Social-auth Kiro accounts (added via /api/oauth/kiro/social-exchange with provider - // Google or GitHub) use a different token format that AWS CodeWhisperer's GetUsageLimits - // routinely rejects with 401/403, even when /messages still works. Surface a clear - // "auth expired, chat may still work" message instead of a generic upstream-error blob - // so the quota card matches what users with legacy social-auth accounts already see. - // Inspired by https://github.com/decolua/9router/pull/620. - if ( - (response.status === 401 || response.status === 403) && - isSocialAuthKiroAccount(providerSpecificData) - ) { - return { - message: "Kiro quota API authentication expired. Chat may still work.", - quotas: {}, - }; - } - const errorText = await response.text(); - throw new Error(`Kiro API error (${response.status}): ${errorText}`); - } - - const data = toRecord(await response.json()); - return buildKiroUsageResult(data); - } catch (error) { - throw new Error(`Failed to fetch Kiro usage: ${error.message}`); - } -} - -/** - * Was this Kiro connection added via the Google/GitHub social-auth device flow - * (POST /api/oauth/kiro/social-exchange)? That route persists - * `{ authMethod: "imported", provider: "Google" | "Github" }` on the connection. - * Builder-ID / IDC / kiro-cli imports use different markers and should keep the - * existing throw-on-failure behavior. - */ -function isSocialAuthKiroAccount(providerSpecificData?: JsonRecord): boolean { - if (!providerSpecificData || providerSpecificData.authMethod !== "imported") return false; - const provider = - typeof providerSpecificData.provider === "string" - ? providerSpecificData.provider.toLowerCase() - : ""; - return provider === "google" || provider === "github"; -} - /** * Vertex AI — SELF-TRACKED spend. * @@ -1473,193 +838,6 @@ async function getVertexUsage(connectionId: string, provider: string) { } } -/** - * Map Kimi membership level to display name - * LEVEL_BASIC = Moderato, LEVEL_INTERMEDIATE = Allegretto, - * LEVEL_ADVANCED = Allegro, LEVEL_STANDARD = Vivace - */ -function getKimiPlanName(level: unknown): string { - if (!level) return ""; - const normalizedLevel = String(level); - - const levelMap = { - LEVEL_BASIC: "Moderato", - LEVEL_INTERMEDIATE: "Allegretto", - LEVEL_ADVANCED: "Allegro", - LEVEL_STANDARD: "Vivace", - }; - - return ( - levelMap[normalizedLevel as keyof typeof levelMap] || - normalizedLevel.replace("LEVEL_", "").toLowerCase() - ); -} - -/** - * Kimi Coding Usage - Fetch quota from Kimi API - * Uses the official /v1/usages endpoint with custom X-Msh-* headers - */ -async function getKimiUsage(accessToken?: string, apiKey?: string) { - // Generate device info for headers (same as OAuth flow) - const deviceId = "kimi-usage-" + Date.now(); - const platform = "omniroute"; - const version = "2.1.2"; - const deviceModel = - typeof process !== "undefined" ? `${process.platform} ${process.arch}` : "unknown"; - - // API key auth takes precedence — Kimi's /usages endpoint accepts the same - // API key used for /messages (verified live: responds with - // authentication.method = METHOD_API_KEY). OAuth flow falls through to the - // Bearer + device-headers shape used by Kimi Coding OAuth. - const useApiKey = typeof apiKey === "string" && apiKey.length > 0; - - const authHeaders: Record = useApiKey - ? { "x-api-key": apiKey as string } - : { - Authorization: `Bearer ${accessToken}`, - "X-Msh-Platform": platform, - "X-Msh-Version": version, - "X-Msh-Device-Model": deviceModel, - "X-Msh-Device-Id": deviceId, - }; - - try { - const response = await fetch(KIMI_CONFIG.usageUrl, { - method: "GET", - headers: { - ...authHeaders, - "Content-Type": "application/json", - }, - }); - - const responseText = await response.text(); - - if (!response.ok) { - return { - plan: "Kimi Coding", - message: `Kimi Coding connected. API Error ${response.status}: ${responseText.slice(0, 100)}`, - }; - } - - let data; - try { - data = JSON.parse(responseText); - } catch { - return { - plan: "Kimi Coding", - message: "Kimi Coding connected. Invalid JSON response from API.", - }; - } - - const quotas: Record = {}; - const dataObj = toRecord(data); - - // Parse Kimi usage response format - // Format: { user: {...}, usage: { limit: "100", used: "92", remaining: "8", resetTime: "..." }, limits: [...] } - const usageObj = toRecord(dataObj.usage); - - // Check for Kimi's actual usage fields (strings, not numbers) - const usageLimit = toNumber(usageObj.limit || usageObj.Limit, 0); - const usageUsed = toNumber(usageObj.used || usageObj.Used, 0); - const usageRemaining = toNumber(usageObj.remaining || usageObj.Remaining, 0); - const usageResetTime = - usageObj.resetTime || usageObj.ResetTime || usageObj.reset_at || usageObj.resetAt; - - if (usageLimit > 0) { - const percentRemaining = usageLimit > 0 ? (usageRemaining / usageLimit) * 100 : 0; - - quotas["Weekly"] = { - used: usageUsed, - total: usageLimit, - remaining: usageRemaining, - remainingPercentage: percentRemaining, - resetAt: parseResetTime(usageResetTime), - unlimited: false, - }; - } - - // Also parse limits array for rate limits - const limitsArray = Array.isArray(dataObj.limits) ? dataObj.limits : []; - for (let i = 0; i < limitsArray.length; i++) { - const limitItem = toRecord(limitsArray[i]); - const window = toRecord(limitItem.window); - const detail = toRecord(limitItem.detail); - - const limit = toNumber(detail.limit || detail.Limit, 0); - const remaining = toNumber(detail.remaining || detail.Remaining, 0); - const resetTime = detail.resetTime || detail.reset_at || detail.resetAt; - - if (limit > 0) { - quotas["Ratelimit"] = { - used: limit - remaining, - total: limit, - remaining, - remainingPercentage: limit > 0 ? (remaining / limit) * 100 : 0, - resetAt: parseResetTime(resetTime), - unlimited: false, - }; - } - } - - // Check for quota windows (Claude-like format with utilization) as fallback - const hasUtilization = (window: JsonRecord) => - window && typeof window === "object" && safePercentage(window.utilization) !== undefined; - - const createQuotaObject = (window: JsonRecord) => { - const remaining = safePercentage(window.utilization) as number; - const used = 100 - remaining; - return { - used, - total: 100, - remaining, - resetAt: parseResetTime(window.resets_at), - remainingPercentage: remaining, - unlimited: false, - }; - }; - - if (hasUtilization(toRecord(dataObj.five_hour))) { - quotas["session (5h)"] = createQuotaObject(toRecord(dataObj.five_hour)); - } - - if (hasUtilization(toRecord(dataObj.seven_day))) { - quotas["weekly (7d)"] = createQuotaObject(toRecord(dataObj.seven_day)); - } - - // Check for model-specific quotas - for (const [key, value] of Object.entries(dataObj)) { - const valueRecord = toRecord(value); - if (key.startsWith("seven_day_") && key !== "seven_day" && hasUtilization(valueRecord)) { - const modelName = key.replace("seven_day_", ""); - quotas[`weekly ${modelName} (7d)`] = createQuotaObject(valueRecord); - } - } - - if (Object.keys(quotas).length > 0) { - const userRecord = toRecord(dataObj.user); - const membershipLevel = toRecord(userRecord.membership).level; - const planName = getKimiPlanName(membershipLevel); - return { - plan: planName || "Kimi Coding", - quotas, - }; - } - - // No quota data in response - const userRecord = toRecord(dataObj.user); - const membershipLevel = toRecord(userRecord.membership).level; - const planName = getKimiPlanName(membershipLevel); - return { - plan: planName || "Kimi Coding", - message: "Kimi Coding connected. Usage tracked per request.", - }; - } catch (error) { - return { - message: `Kimi Coding connected. Unable to fetch usage: ${(error as Error).message}`, - }; - } -} - /** * Qwen Usage */ diff --git a/open-sse/services/usage/claude.ts b/open-sse/services/usage/claude.ts new file mode 100644 index 0000000000..e205b473ea --- /dev/null +++ b/open-sse/services/usage/claude.ts @@ -0,0 +1,216 @@ +/** + * usage/claude.ts — Claude (Anthropic OAuth + legacy org) usage fetcher + plan-label helper. + * + * Extracted from services/usage.ts (god-file decomposition): the Claude family — the API + * config, the plan-label picker, the OAuth usage fetcher (getClaudeUsage) with its legacy + * settings/org fallback (getClaudeUsageLegacy). Depends only on the sibling scalar/quota + * leaves + Claude identity/cooldown helpers + safePercentage — no host coupling — so it + * lives as a co-located provider leaf. usage.ts imports getClaudeUsage (dispatcher) + + * getClaudePlanLabel (__testing). Behavior-preserving move. + */ + +import { safePercentage } from "@/shared/utils/formatting"; +import { CLAUDE_CODE_VERSION, fetchClaudeBootstrap } from "../../executors/claudeIdentity.ts"; +import { isClaudeOauthUsageCoolingDown, markClaudeOauthUsage429 } from "../claudeUsageCooldown.ts"; +import { toRecord } from "./scalars.ts"; +import { type UsageQuota, parseResetTime } from "./quota.ts"; + +type JsonRecord = Record; + +// Claude API config +const CLAUDE_CONFIG = { + oauthUsageUrl: "https://api.anthropic.com/api/oauth/usage", + usageUrl: "https://api.anthropic.com/v1/organizations/{org_id}/usage", + settingsUrl: "https://api.anthropic.com/v1/settings", + apiVersion: "2023-06-01", +}; + +export function getClaudePlanLabel(...candidates: Array): string | null { + for (const candidate of candidates) { + if (typeof candidate !== "string") continue; + const trimmed = candidate.trim(); + if ( + !trimmed || + trimmed.toLowerCase() === "claude code" || + trimmed.toLowerCase() === "unknown" + ) { + continue; + } + return trimmed; + } + return null; +} + +/** + * Claude Usage - Try to fetch from Anthropic API + */ +export async function getClaudeUsage(accessToken?: string) { + if (!accessToken) { + return { message: "Claude connected. Access token not available.", bootstrap: null }; + } + + // Refresh bootstrap in parallel; best-effort, failure non-fatal. + const bootstrapPromise = fetchClaudeBootstrap(accessToken).catch(() => null); + // Skip OAuth usage call while this token is cooling down from a recent 429 + // (chat with the same token still works — only the quota endpoint is throttled). + if (isClaudeOauthUsageCoolingDown(accessToken)) { + const legacy = await getClaudeUsageLegacy(accessToken); + return { ...legacy, bootstrap: await bootstrapPromise }; + } + try { + // Real CLI uses axios here, not Stainless — UA is `claude-code/` + // (not `claude-cli/...`) and the shape is simpler than /v1/messages. + const ctrl = new AbortController(); + const timer = setTimeout(() => ctrl.abort(), 10_000); + let oauthResponse; + try { + oauthResponse = await fetch(CLAUDE_CONFIG.oauthUsageUrl, { + method: "GET", + headers: { + Accept: "application/json, text/plain, */*", + "Accept-Encoding": "gzip, compress, deflate, br", + Authorization: `Bearer ${accessToken}`, + "Content-Type": "application/json", + "User-Agent": `claude-code/${CLAUDE_CODE_VERSION}`, + "anthropic-beta": "oauth-2025-04-20", + }, + signal: ctrl.signal, + }); + } finally { + clearTimeout(timer); + } + + if (oauthResponse.ok) { + const data = toRecord(await oauthResponse.json()); + const quotas: Record = {}; + + // utilization = percentage USED (e.g., 90 means 90% used, 10% remaining) + // Confirmed via user report #299: Claude.ai shows 87% used = OmniRoute must show 13% remaining. + const hasUtilization = (window: JsonRecord) => + window && typeof window === "object" && safePercentage(window.utilization) !== undefined; + + const createQuotaObject = (window: JsonRecord) => { + const used = safePercentage(window.utilization) as number; // utilization = % used + const remaining = Math.max(0, 100 - used); + return { + used, + total: 100, + remaining, + resetAt: parseResetTime(window.resets_at), + remainingPercentage: remaining, + unlimited: false, + }; + }; + + const fiveHour = toRecord(data.five_hour); + if (hasUtilization(fiveHour)) { + quotas["session (5h)"] = createQuotaObject(fiveHour); + } + + const sevenDay = toRecord(data.seven_day); + if (hasUtilization(sevenDay)) { + quotas["weekly (7d)"] = createQuotaObject(sevenDay); + } + + // Map Anthropic's internal codenames (e.g., omelette → Designer) for display. + const MODEL_DISPLAY_NAMES: Record = { + omelette: "designer", + }; + for (const [key, value] of Object.entries(data)) { + const valueRecord = toRecord(value); + if (key.startsWith("seven_day_") && key !== "seven_day" && hasUtilization(valueRecord)) { + const codename = key.replace("seven_day_", ""); + const modelName = MODEL_DISPLAY_NAMES[codename] || codename; + quotas[`weekly ${modelName} (7d)`] = createQuotaObject(valueRecord); + } + } + + const bootstrap = await bootstrapPromise; + const plan = + getClaudePlanLabel( + typeof data.tier === "string" ? data.tier : null, + typeof data.plan === "string" ? data.plan : null, + typeof data.subscription_type === "string" ? data.subscription_type : null, + bootstrap?.organization_rate_limit_tier + ) ?? undefined; + + return { + ...(plan ? { plan } : {}), + quotas, + extraUsage: data.extra_usage ?? null, + bootstrap, + }; + } + + // Cool down OAuth usage polling after a 429 (quota endpoint only — chat is unaffected). + if (oauthResponse.status === 429) { + markClaudeOauthUsage429(accessToken); + } + + // Fallback: OAuth endpoint returned non-OK, try legacy settings/org endpoint + console.warn( + `[Claude Usage] OAuth endpoint returned ${oauthResponse.status}, falling back to legacy` + ); + const legacy = await getClaudeUsageLegacy(accessToken); + return { ...legacy, bootstrap: await bootstrapPromise }; + } catch (error) { + return { + message: `Claude connected. Unable to fetch usage: ${(error as Error).message}`, + bootstrap: await bootstrapPromise, + }; + } +} + +/** + * Legacy Claude usage fetcher for API key / org admin users. + * Uses /v1/settings + /v1/organizations/{org_id}/usage endpoints. + */ +async function getClaudeUsageLegacy(accessToken?: string) { + try { + const settingsResponse = await fetch(CLAUDE_CONFIG.settingsUrl, { + method: "GET", + headers: { + Authorization: `Bearer ${accessToken}`, + "anthropic-version": CLAUDE_CONFIG.apiVersion, + }, + }); + + if (settingsResponse.ok) { + const settings = toRecord(await settingsResponse.json()); + + const organizationId = + typeof settings.organization_id === "string" ? settings.organization_id : ""; + if (organizationId) { + const usageResponse = await fetch( + CLAUDE_CONFIG.usageUrl.replace("{org_id}", organizationId), + { + method: "GET", + headers: { + Authorization: `Bearer ${accessToken}`, + "anthropic-version": CLAUDE_CONFIG.apiVersion, + }, + } + ); + + if (usageResponse.ok) { + const usage = await usageResponse.json(); + return { + plan: settings.plan || "Unknown", + organization: settings.organization_name, + quotas: usage, + }; + } + } + + return { + plan: settings.plan || "Unknown", + organization: settings.organization_name, + message: "Claude connected. Usage details require admin access.", + }; + } + + return { message: "Claude connected. Usage API requires admin permissions." }; + } catch (error) { + return { message: `Claude connected. Unable to fetch usage: ${(error as Error).message}` }; + } +} diff --git a/open-sse/services/usage/codex.ts b/open-sse/services/usage/codex.ts new file mode 100644 index 0000000000..b1d0261f03 --- /dev/null +++ b/open-sse/services/usage/codex.ts @@ -0,0 +1,70 @@ +/** + * usage/codex.ts — Codex (OpenAI / ChatGPT backend) usage fetcher. + * + * Extracted from services/usage.ts (god-file decomposition): the Codex family — the ChatGPT + * backend usage-API config and the getCodexUsage fetcher that reads the persisted workspace + * binding and shapes quotas via buildCodexUsageQuotas. Depends only on the scalar leaf + + * codexUsageQuotas — no host coupling — so it lives as a co-located provider leaf. usage.ts + * imports getCodexUsage (dispatcher). Behavior-preserving move. + */ + +import { buildCodexUsageQuotas } from "../codexUsageQuotas.ts"; +import { getFieldValue } from "./scalars.ts"; + +// Codex (OpenAI) API config +const CODEX_CONFIG = { + usageUrl: "https://chatgpt.com/backend-api/wham/usage", +}; + +/** + * Codex (OpenAI) Usage - Fetch from ChatGPT backend API + * IMPORTANT: Uses persisted workspaceId from OAuth to ensure correct workspace binding. + * No fallback to other workspaces - strict binding to user's selected workspace. + */ +export async function getCodexUsage( + accessToken?: string, + providerSpecificData: Record = {} +) { + try { + // Use persisted workspace ID from OAuth - NO FALLBACK + const accountId = + typeof providerSpecificData.workspaceId === "string" + ? providerSpecificData.workspaceId + : null; + + const headers: Record = { + Authorization: `Bearer ${accessToken}`, + "Content-Type": "application/json", + Accept: "application/json", + }; + if (accountId) { + headers["chatgpt-account-id"] = accountId; + } + + const response = await fetch(CODEX_CONFIG.usageUrl, { + method: "GET", + headers, + }); + + if (!response.ok) { + if (response.status === 401 || response.status === 403) { + return { + message: `Codex token expired or access denied. Please re-authenticate the connection.`, + }; + } + throw new Error(`Codex API error: ${response.status}`); + } + + const data = await response.json(); + + const { rateLimit, quotas } = buildCodexUsageQuotas(data); + + return { + plan: String(getFieldValue(data, "plan_type", "planType") || "unknown"), + limitReached: Boolean(getFieldValue(rateLimit, "limit_reached", "limitReached")), + quotas, + }; + } catch (error) { + return { message: `Failed to fetch Codex usage: ${(error as Error).message}` }; + } +} diff --git a/open-sse/services/usage/cursor.ts b/open-sse/services/usage/cursor.ts new file mode 100644 index 0000000000..67acc09a6a --- /dev/null +++ b/open-sse/services/usage/cursor.ts @@ -0,0 +1,161 @@ +/** + * usage/cursor.ts — Cursor (Pro) usage fetcher + JWT/config helpers. + * + * Extracted from services/usage.ts (god-file decomposition): the Cursor family — the + * dashboard usage-API config, the WorkOS JWT `sub` decoder, and the getCursorUsage fetcher + * that probes the cursor.com/dashboard/spending endpoint. Depends only on the sibling + * scalar/quota leaves — no host coupling — so it lives as a co-located provider leaf. + * usage.ts imports getCursorUsage (dispatcher). Behavior-preserving move. + */ + +import { toRecord, toNumber, clampPercentage } from "./scalars.ts"; +import { type UsageQuota, parseResetTime } from "./quota.ts"; + +// Cursor dashboard usage API config +// The endpoint that powers https://cursor.com/dashboard/spending. Validates the WorkOS +// session via the WorkosCursorSessionToken cookie (format: `${userId}::${jwt}`) and +// rejects requests without a matching Origin/Referer (Invalid origin for state-changing request). +const CURSOR_USAGE_CONFIG = { + usageUrl: "https://cursor.com/api/dashboard/get-current-period-usage", + origin: "https://cursor.com", + referer: "https://cursor.com/dashboard/spending", + userAgent: + "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/149.0.0.0 Safari/537.36", +}; + +/** + * Decode the `sub` claim of a Cursor JWT (the WorkOS user id). + * Returns null if the token is not a parseable JWT. + */ +function decodeCursorJwtSub(token: string): string | null { + if (!token || typeof token !== "string") return null; + const parts = token.split("."); + if (parts.length !== 3) return null; + try { + let payload = parts[1].replace(/-/g, "+").replace(/_/g, "/"); + while (payload.length % 4 !== 0) payload += "="; + const decoded = JSON.parse(Buffer.from(payload, "base64").toString("utf8")); + const sub = decoded?.sub; + return typeof sub === "string" && sub.length > 0 ? sub : null; + } catch { + return null; + } +} + +/** + * Cursor Pro Plan Usage + * Fetches current-billing-cycle spend from the cursor.com dashboard API and exposes three + * windows that mirror the cursor.com/dashboard/spending UI: Total / Auto + Composer / API. + */ +export async function getCursorUsage(accessToken: string, providerSpecificData?: unknown) { + if (!accessToken) { + return { message: "Cursor access token missing. Re-import the connection from Cursor IDE." }; + } + + const storedUserId = (() => { + const raw = toRecord(providerSpecificData).userId; + return typeof raw === "string" && raw.length > 0 ? raw : null; + })(); + const userId = storedUserId || decodeCursorJwtSub(accessToken); + + if (!userId) { + return { + message: "Cursor token missing user id. Re-import the connection from Cursor IDE.", + }; + } + + try { + const response = await fetch(CURSOR_USAGE_CONFIG.usageUrl, { + method: "POST", + redirect: "manual", + headers: { + Cookie: `WorkosCursorSessionToken=${userId}::${accessToken}`, + Origin: CURSOR_USAGE_CONFIG.origin, + Referer: CURSOR_USAGE_CONFIG.referer, + "Content-Type": "application/json", + Accept: "application/json", + "User-Agent": CURSOR_USAGE_CONFIG.userAgent, + }, + body: "{}", + }); + + // 3xx redirect to WorkOS authkit means the session cookie was rejected. + if (response.status >= 300 && response.status < 400) { + return { + plan: "Cursor", + message: "Cursor session expired. Re-import the token from Cursor IDE.", + }; + } + + if (!response.ok) { + const errorText = (await response.text()).slice(0, 200); + if (response.status === 401 || response.status === 403) { + return { + plan: "Cursor", + message: "Cursor session unauthorized. Re-import the token from Cursor IDE.", + }; + } + return { + plan: "Cursor", + message: `Cursor usage endpoint error (${response.status}): ${errorText}`, + }; + } + + const data = toRecord(await response.json()); + const planUsage = toRecord(data.planUsage); + + if (Object.keys(planUsage).length === 0) { + return { + plan: "Cursor", + message: "Cursor connected. No active plan usage returned.", + }; + } + + const limitCents = Math.max(0, toNumber(planUsage.limit, 0)); + const totalSpendCents = Math.max(0, toNumber(planUsage.totalSpend, 0)); + const autoPercentUsed = clampPercentage(toNumber(planUsage.autoPercentUsed, 0)); + const apiPercentUsed = clampPercentage(toNumber(planUsage.apiPercentUsed, 0)); + const totalPercentUsed = clampPercentage(toNumber(planUsage.totalPercentUsed, 0)); + + // billingCycleEnd is a numeric-string in ms; coerce so parseResetTime sees a number. + const billingCycleEndMs = toNumber(data.billingCycleEnd, 0); + const resetAt = billingCycleEndMs > 0 ? parseResetTime(billingCycleEndMs) : null; + + // Convert cents → dollars rounded to 2 decimal places. + const toDollars = (cents: number) => Math.round(cents) / 100; + + const limitDollars = toDollars(limitCents); + const buildWindow = (percentUsed: number, usedCentsOverride?: number): UsageQuota => { + const usedCents = + typeof usedCentsOverride === "number" + ? usedCentsOverride + : Math.round((limitCents * percentUsed) / 100); + const used = toDollars(Math.min(usedCents, limitCents)); + const remaining = toDollars(Math.max(limitCents - Math.min(usedCents, limitCents), 0)); + return { + used, + total: limitDollars, + remaining, + remainingPercentage: clampPercentage(100 - percentUsed), + resetAt, + unlimited: false, + }; + }; + + const quotas: Record = { + Total: buildWindow(totalPercentUsed, totalSpendCents), + "Auto + Composer": buildWindow(autoPercentUsed), + API: buildWindow(apiPercentUsed), + }; + + return { + plan: "Cursor Pro", + quotas, + }; + } catch (error) { + return { + plan: "Cursor", + message: `Cursor connected. Unable to fetch usage: ${(error as Error).message}`, + }; + } +} diff --git a/open-sse/services/usage/kimi.ts b/open-sse/services/usage/kimi.ts new file mode 100644 index 0000000000..f5e0657475 --- /dev/null +++ b/open-sse/services/usage/kimi.ts @@ -0,0 +1,209 @@ +/** + * usage/kimi.ts — Kimi Coding (kimi-coding / kimi-coding-apikey) usage fetcher + helpers. + * + * Extracted from services/usage.ts (god-file decomposition): the Kimi family — the coding + * API config, membership-level → display-name mapping, and the getKimiUsage fetcher that + * probes the official /v1/usages endpoint. Depends only on the sibling scalar/quota leaves + * plus safePercentage — no host coupling — so it lives as a co-located provider leaf. + * usage.ts imports getKimiUsage (dispatcher). Behavior-preserving move. + */ + +import { safePercentage } from "@/shared/utils/formatting"; +import { toRecord, toNumber } from "./scalars.ts"; +import { type UsageQuota, parseResetTime } from "./quota.ts"; + +type JsonRecord = Record; + +// Kimi Coding API config +const KIMI_CONFIG = { + baseUrl: "https://api.kimi.com/coding/v1", + usageUrl: "https://api.kimi.com/coding/v1/usages", + apiVersion: "2023-06-01", +}; + +/** + * Map Kimi membership level to display name + * LEVEL_BASIC = Moderato, LEVEL_INTERMEDIATE = Allegretto, + * LEVEL_ADVANCED = Allegro, LEVEL_STANDARD = Vivace + */ +function getKimiPlanName(level: unknown): string { + if (!level) return ""; + const normalizedLevel = String(level); + + const levelMap = { + LEVEL_BASIC: "Moderato", + LEVEL_INTERMEDIATE: "Allegretto", + LEVEL_ADVANCED: "Allegro", + LEVEL_STANDARD: "Vivace", + }; + + return ( + levelMap[normalizedLevel as keyof typeof levelMap] || + normalizedLevel.replace("LEVEL_", "").toLowerCase() + ); +} + +/** + * Kimi Coding Usage - Fetch quota from Kimi API + * Uses the official /v1/usages endpoint with custom X-Msh-* headers + */ +export async function getKimiUsage(accessToken?: string, apiKey?: string) { + // Generate device info for headers (same as OAuth flow) + const deviceId = "kimi-usage-" + Date.now(); + const platform = "omniroute"; + const version = "2.1.2"; + const deviceModel = + typeof process !== "undefined" ? `${process.platform} ${process.arch}` : "unknown"; + + // API key auth takes precedence — Kimi's /usages endpoint accepts the same + // API key used for /messages (verified live: responds with + // authentication.method = METHOD_API_KEY). OAuth flow falls through to the + // Bearer + device-headers shape used by Kimi Coding OAuth. + const useApiKey = typeof apiKey === "string" && apiKey.length > 0; + + const authHeaders: Record = useApiKey + ? { "x-api-key": apiKey as string } + : { + Authorization: `Bearer ${accessToken}`, + "X-Msh-Platform": platform, + "X-Msh-Version": version, + "X-Msh-Device-Model": deviceModel, + "X-Msh-Device-Id": deviceId, + }; + + try { + const response = await fetch(KIMI_CONFIG.usageUrl, { + method: "GET", + headers: { + ...authHeaders, + "Content-Type": "application/json", + }, + }); + + const responseText = await response.text(); + + if (!response.ok) { + return { + plan: "Kimi Coding", + message: `Kimi Coding connected. API Error ${response.status}: ${responseText.slice(0, 100)}`, + }; + } + + let data; + try { + data = JSON.parse(responseText); + } catch { + return { + plan: "Kimi Coding", + message: "Kimi Coding connected. Invalid JSON response from API.", + }; + } + + const quotas: Record = {}; + const dataObj = toRecord(data); + + // Parse Kimi usage response format + // Format: { user: {...}, usage: { limit: "100", used: "92", remaining: "8", resetTime: "..." }, limits: [...] } + const usageObj = toRecord(dataObj.usage); + + // Check for Kimi's actual usage fields (strings, not numbers) + const usageLimit = toNumber(usageObj.limit || usageObj.Limit, 0); + const usageUsed = toNumber(usageObj.used || usageObj.Used, 0); + const usageRemaining = toNumber(usageObj.remaining || usageObj.Remaining, 0); + const usageResetTime = + usageObj.resetTime || usageObj.ResetTime || usageObj.reset_at || usageObj.resetAt; + + if (usageLimit > 0) { + const percentRemaining = usageLimit > 0 ? (usageRemaining / usageLimit) * 100 : 0; + + quotas["Weekly"] = { + used: usageUsed, + total: usageLimit, + remaining: usageRemaining, + remainingPercentage: percentRemaining, + resetAt: parseResetTime(usageResetTime), + unlimited: false, + }; + } + + // Also parse limits array for rate limits + const limitsArray = Array.isArray(dataObj.limits) ? dataObj.limits : []; + for (let i = 0; i < limitsArray.length; i++) { + const limitItem = toRecord(limitsArray[i]); + const window = toRecord(limitItem.window); + const detail = toRecord(limitItem.detail); + + const limit = toNumber(detail.limit || detail.Limit, 0); + const remaining = toNumber(detail.remaining || detail.Remaining, 0); + const resetTime = detail.resetTime || detail.reset_at || detail.resetAt; + + if (limit > 0) { + quotas["Ratelimit"] = { + used: limit - remaining, + total: limit, + remaining, + remainingPercentage: limit > 0 ? (remaining / limit) * 100 : 0, + resetAt: parseResetTime(resetTime), + unlimited: false, + }; + } + } + + // Check for quota windows (Claude-like format with utilization) as fallback + const hasUtilization = (window: JsonRecord) => + window && typeof window === "object" && safePercentage(window.utilization) !== undefined; + + const createQuotaObject = (window: JsonRecord) => { + const remaining = safePercentage(window.utilization) as number; + const used = 100 - remaining; + return { + used, + total: 100, + remaining, + resetAt: parseResetTime(window.resets_at), + remainingPercentage: remaining, + unlimited: false, + }; + }; + + if (hasUtilization(toRecord(dataObj.five_hour))) { + quotas["session (5h)"] = createQuotaObject(toRecord(dataObj.five_hour)); + } + + if (hasUtilization(toRecord(dataObj.seven_day))) { + quotas["weekly (7d)"] = createQuotaObject(toRecord(dataObj.seven_day)); + } + + // Check for model-specific quotas + for (const [key, value] of Object.entries(dataObj)) { + const valueRecord = toRecord(value); + if (key.startsWith("seven_day_") && key !== "seven_day" && hasUtilization(valueRecord)) { + const modelName = key.replace("seven_day_", ""); + quotas[`weekly ${modelName} (7d)`] = createQuotaObject(valueRecord); + } + } + + if (Object.keys(quotas).length > 0) { + const userRecord = toRecord(dataObj.user); + const membershipLevel = toRecord(userRecord.membership).level; + const planName = getKimiPlanName(membershipLevel); + return { + plan: planName || "Kimi Coding", + quotas, + }; + } + + // No quota data in response + const userRecord = toRecord(dataObj.user); + const membershipLevel = toRecord(userRecord.membership).level; + const planName = getKimiPlanName(membershipLevel); + return { + plan: planName || "Kimi Coding", + message: "Kimi Coding connected. Usage tracked per request.", + }; + } catch (error) { + return { + message: `Kimi Coding connected. Unable to fetch usage: ${(error as Error).message}`, + }; + } +} diff --git a/open-sse/services/usage/kiro.ts b/open-sse/services/usage/kiro.ts new file mode 100644 index 0000000000..9b85579103 --- /dev/null +++ b/open-sse/services/usage/kiro.ts @@ -0,0 +1,243 @@ +/** + * usage/kiro.ts — Kiro / Amazon Q (AWS CodeWhisperer) usage fetcher + quota helpers. + * + * Extracted from services/usage.ts (god-file decomposition): the Kiro family — overage + * detection, per-resource quota assembly (buildKiroQuota / buildKiroUsageResult), region-aware + * profile-ARN discovery, the social-auth account marker, and the getKiroUsage fetcher that + * calls GetUsageLimits on the region-matched CodeWhisperer endpoint. Depends only on the + * sibling scalar/quota leaves — no host coupling — so it lives as a co-located provider leaf. + * usage.ts imports getKiroUsage (dispatcher) + re-exports buildKiroUsageResult / + * discoverKiroProfileArn (external kiro tests import them from services/usage) and pulls + * getKiroUsage into __testing. Behavior-preserving move. + */ + +import { toRecord, toNumber } from "./scalars.ts"; +import { type UsageQuota, parseResetTime } from "./quota.ts"; + +type JsonRecord = Record; + +const CODEWHISPERER_BASE_URL = + process.env.OMNIROUTE_CODEWHISPERER_BASE_URL ?? "https://codewhisperer.us-east-1.amazonaws.com"; + +function isKiroOverageEnabled(data: JsonRecord): boolean { + const overageConfiguration = toRecord(data.overageConfiguration); + const overageStatus = String(overageConfiguration.overageStatus || "") + .trim() + .toUpperCase(); + + return ( + overageStatus === "ENABLED" || + data.overageEnabled === true || + overageConfiguration.overageEnabled === true + ); +} + +function buildKiroQuota( + used: number, + total: number, + resetAt: string | null, + overageEnabled: boolean +): UsageQuota { + const remaining = total - used; + + if (!overageEnabled) { + return { used, total, remaining, resetAt, unlimited: false }; + } + + return { + used, + total, + remaining, + remainingPercentage: 100, + resetAt, + unlimited: true, + }; +} + +/** + * Build the Kiro usage result from a GetUsageLimits response. When the account returns no + * usage breakdown (some AWS IAM / Builder ID accounts don't expose per-resource quota via + * GetUsageLimits), return an informative message instead of empty `quotas:{}` — otherwise the + * dashboard renders a blank quota card with no explanation (#3506). Exported for testing. + */ +export function buildKiroUsageResult( + data: JsonRecord +): { plan: string; quotas: Record } | { message: string } { + const usageList = Array.isArray(data.usageBreakdownList) ? data.usageBreakdownList : []; + const quotaInfo: Record = {}; + const resetAt = parseResetTime(data.nextDateReset || data.resetDate); + const overageEnabled = isKiroOverageEnabled(data); + + usageList.forEach((breakdownValue: unknown) => { + const breakdown = toRecord(breakdownValue); + const resourceType = + typeof breakdown.resourceType === "string" ? breakdown.resourceType.toLowerCase() : "unknown"; + const used = toNumber(breakdown.currentUsageWithPrecision, 0); + const total = toNumber(breakdown.usageLimitWithPrecision, 0); + + quotaInfo[resourceType] = buildKiroQuota(used, total, resetAt, overageEnabled); + + const freeTrialInfo = toRecord(breakdown.freeTrialInfo); + if (Object.keys(freeTrialInfo).length > 0) { + const freeUsed = toNumber(freeTrialInfo.currentUsageWithPrecision, 0); + const freeTotal = toNumber(freeTrialInfo.usageLimitWithPrecision, 0); + quotaInfo[`${resourceType}_freetrial`] = buildKiroQuota( + freeUsed, + freeTotal, + resetAt, + overageEnabled + ); + } + }); + + if (Object.keys(quotaInfo).length === 0) { + return { + message: + "Kiro connected, but the account returned no usage breakdown. Some AWS IAM / Builder ID accounts don't expose per-resource quota via GetUsageLimits.", + }; + } + + return { + plan: String(toRecord(data.subscriptionInfo).subscriptionTitle || "").trim() || "Kiro", + quotas: quotaInfo, + }; +} + +/** + * Discover a Kiro/CodeWhisperer profile ARN for an account that didn't persist one (common for + * AWS IAM Identity Center logins and kiro-cli imports). Calls ListAvailableProfiles on the + * region-matched endpoint and prefers a profile whose ARN is in the same region. Returns + * undefined when no profile is available (e.g. the org/token has no Kiro entitlement). + * Exported for testing. + */ +export async function discoverKiroProfileArn( + accessToken: string, + usageBaseUrl: string, + region: string +): Promise { + try { + const response = await fetch(usageBaseUrl, { + method: "POST", + headers: { + Authorization: `Bearer ${accessToken}`, + "Content-Type": "application/x-amz-json-1.0", + "x-amz-target": "AmazonCodeWhispererService.ListAvailableProfiles", + Accept: "application/json", + }, + body: JSON.stringify({ maxResults: 10 }), + // Don't let a hung profile lookup block the usage/quota refresh indefinitely. + signal: AbortSignal.timeout(10000), + }); + if (!response.ok) return undefined; + + const data = toRecord(await response.json()); + const profiles = Array.isArray(data.profiles) ? data.profiles : []; + const normalizedRegion = region.toLowerCase(); + const matched = + profiles.find((profile: unknown) => { + const arn = toRecord(profile).arn; + return typeof arn === "string" && arn.toLowerCase().includes(`:${normalizedRegion}:`); + }) || profiles[0]; + const arn = toRecord(matched).arn; + return typeof arn === "string" && arn.length > 0 ? arn : undefined; + } catch { + return undefined; + } +} + +/** + * Kiro (AWS CodeWhisperer) Usage + */ +export async function getKiroUsage(accessToken?: string, providerSpecificData?: JsonRecord) { + try { + let profileArn = + typeof providerSpecificData?.profileArn === "string" + ? providerSpecificData.profileArn + : undefined; + + // Enterprise IAM Identity Center accounts are region-bound: the profileArn, token and + // endpoint must all match the region. Derive the region from the stored region (preferred) + // or the profileArn, then route to the regional Amazon Q endpoint (us-east-1 keeps the + // legacy codewhisperer host; codewhisperer.{region} does not resolve for other regions). + const regionFromArn = profileArn + ? profileArn.toLowerCase().match(/^arn:aws:codewhisperer:([a-z0-9-]+):/)?.[1] + : undefined; + const region = + (typeof providerSpecificData?.region === "string" && + providerSpecificData.region.trim().toLowerCase()) || + regionFromArn || + "us-east-1"; + const usageBaseUrl = + region === "us-east-1" ? CODEWHISPERER_BASE_URL : `https://q.${region}.amazonaws.com`; + + // IAM Identity Center logins and kiro-cli imports frequently don't persist a profileArn, which + // previously caused the quota card to show nothing ("0 used"). Discover it on demand from + // ListAvailableProfiles (region-matched) so usage still resolves for those accounts. + if (!profileArn && accessToken) { + profileArn = await discoverKiroProfileArn(accessToken, usageBaseUrl, region); + } + + if (!profileArn) { + return { message: "Kiro connected. Profile ARN not available for quota tracking." }; + } + + // Kiro uses AWS CodeWhisperer GetUsageLimits API + const payload = { + origin: "AI_EDITOR", + profileArn: profileArn, + resourceType: "AGENTIC_REQUEST", + }; + + const response = await fetch(usageBaseUrl, { + method: "POST", + headers: { + Authorization: `Bearer ${accessToken}`, + "Content-Type": "application/x-amz-json-1.0", + "x-amz-target": "AmazonCodeWhispererService.GetUsageLimits", + Accept: "application/json", + }, + body: JSON.stringify(payload), + }); + + if (!response.ok) { + // Social-auth Kiro accounts (added via /api/oauth/kiro/social-exchange with provider + // Google or GitHub) use a different token format that AWS CodeWhisperer's GetUsageLimits + // routinely rejects with 401/403, even when /messages still works. Surface a clear + // "auth expired, chat may still work" message instead of a generic upstream-error blob + // so the quota card matches what users with legacy social-auth accounts already see. + // Inspired by https://github.com/decolua/9router/pull/620. + if ( + (response.status === 401 || response.status === 403) && + isSocialAuthKiroAccount(providerSpecificData) + ) { + return { + message: "Kiro quota API authentication expired. Chat may still work.", + quotas: {}, + }; + } + const errorText = await response.text(); + throw new Error(`Kiro API error (${response.status}): ${errorText}`); + } + + const data = toRecord(await response.json()); + return buildKiroUsageResult(data); + } catch (error) { + throw new Error(`Failed to fetch Kiro usage: ${error.message}`); + } +} + +/** + * Was this Kiro connection added via the Google/GitHub social-auth device flow + * (POST /api/oauth/kiro/social-exchange)? That route persists + * `{ authMethod: "imported", provider: "Google" | "Github" }` on the connection. + * Builder-ID / IDC / kiro-cli imports use different markers and should keep the + * existing throw-on-failure behavior. + */ +function isSocialAuthKiroAccount(providerSpecificData?: JsonRecord): boolean { + if (!providerSpecificData || providerSpecificData.authMethod !== "imported") return false; + const provider = + typeof providerSpecificData.provider === "string" + ? providerSpecificData.provider.toLowerCase() + : ""; + return provider === "google" || provider === "github"; +} diff --git a/open-sse/translator/helpers/toolCallHelper.ts b/open-sse/translator/helpers/toolCallHelper.ts index 01f866fac7..caf69dd184 100644 --- a/open-sse/translator/helpers/toolCallHelper.ts +++ b/open-sse/translator/helpers/toolCallHelper.ts @@ -169,3 +169,56 @@ export function fixMissingToolResponses(body) { body.messages = newMessages; return body; } + +// Strip tool-result carriers whose id has no matching tool call anywhere in the +// conversation. Mirrors fixMissingToolResponses on the result side: that helper +// ensures every call has a result; this one ensures every result has a call. +// Client-side history truncation/compaction can drop the assistant turn that +// issued a tool call while leaving its stale tool result behind, which strict +// upstream APIs reject before model execution. Handles both OpenAI-format +// role:"tool" messages and Claude-format tool_result content blocks. Drops a +// user message entirely if stripping empties its content array. Returns the +// same body reference when nothing needs to change (no-op fast path). +export function stripOrphanedToolResults(body) { + if (!body.messages || !Array.isArray(body.messages)) return body; + + const knownCallIds = new Set(); + for (const msg of body.messages) { + for (const id of getToolCallIds(msg)) { + knownCallIds.add(id); + } + } + + let changed = false; + const filteredMessages = []; + + for (const msg of body.messages) { + if (msg.role === "tool" && msg.tool_call_id) { + if (knownCallIds.has(msg.tool_call_id)) { + filteredMessages.push(msg); + } else { + changed = true; + } + continue; + } + + if (Array.isArray(msg.content)) { + const cleanedContent = msg.content.filter((block) => { + if (block?.type !== "tool_result") return true; + return typeof block.tool_use_id === "string" && knownCallIds.has(block.tool_use_id); + }); + + if (cleanedContent.length !== msg.content.length) { + changed = true; + if (cleanedContent.length === 0) continue; + msg.content = cleanedContent; + } + } + + filteredMessages.push(msg); + } + + if (!changed) return body; + body.messages = filteredMessages; + return body; +} diff --git a/open-sse/translator/index.ts b/open-sse/translator/index.ts index 6933741489..12406078ec 100644 --- a/open-sse/translator/index.ts +++ b/open-sse/translator/index.ts @@ -1,5 +1,9 @@ import { FORMATS } from "./formats.ts"; -import { ensureToolCallIds, fixMissingToolResponses } from "./helpers/toolCallHelper.ts"; +import { + ensureToolCallIds, + fixMissingToolResponses, + stripOrphanedToolResults, +} from "./helpers/toolCallHelper.ts"; import { NON_ANTHROPIC_THINKING_PLACEHOLDER, prepareClaudeRequest, @@ -178,6 +182,9 @@ export function translateRequest( // Fix missing tool responses (insert empty tool_result if needed) fixMissingToolResponses(result); + // Strip orphaned tool results (tool_result/role:tool with no matching tool_call) + stripOrphanedToolResults(result); + // Normalize roles: developer→system unless preserved, system→user for incompatible models. // This handles (1) sourceFormat openai with messages containing developer → non-openai target // or preserveDeveloperRole=false, and (2) all other paths where result.messages already exists. @@ -340,6 +347,7 @@ export function translateRequest( // Ensure unique tool_call ids on final payload (translators may have introduced duplicates) ensureToolCallIds(result, { use9CharId }); fixMissingToolResponses(result); + stripOrphanedToolResults(result); if (result.tools) { result.tools = coerceToolSchemas(result.tools); diff --git a/open-sse/translator/request/claude-to-openai.ts b/open-sse/translator/request/claude-to-openai.ts index e96bf4b7e4..9b8b70f3c7 100644 --- a/open-sse/translator/request/claude-to-openai.ts +++ b/open-sse/translator/request/claude-to-openai.ts @@ -410,7 +410,10 @@ function convertClaudeMessage(msg, preserveCacheControl = false) { type: "function", function: { name: block.name, - arguments: JSON.stringify(block.input || {}), + arguments: + typeof block.input === "string" + ? block.input + : JSON.stringify(block.input || {}), }, }); break; diff --git a/open-sse/translator/request/openai-to-gemini.ts b/open-sse/translator/request/openai-to-gemini.ts index 39a7e16018..703f2cbe10 100644 --- a/open-sse/translator/request/openai-to-gemini.ts +++ b/open-sse/translator/request/openai-to-gemini.ts @@ -24,6 +24,20 @@ import { cleanJSONSchemaForAntigravity, } from "../helpers/geminiHelper.ts"; import { buildGeminiTools, sanitizeGeminiToolName } from "../helpers/geminiToolsSanitizer.ts"; +import { + type GeminiGenerationConfig, + isVertexGeminiProvider, + buildChangedToolNameMap, + extractClientThoughtSignature, + deepCleanUndefined, + applyAntigravityGenerationDefaults, + stringifyHistoricalToolArguments, + buildInertHistoricalToolCallText, + buildInertHistoricalToolResponseText, + escapeHistoricalContextAttribute, + escapeHistoricalContextContent, + buildHistoricalToolResultContext, +} from "./openai-to-gemini/helpers.ts"; // Observed Antigravity wrapper output cap, not an underlying model capability. // Keep this bridge-local: Antigravity currently caps visible output around 16K. @@ -43,20 +57,6 @@ const GEMINI_BUILTIN_TOOL_NAMES = new Set([ type GeminiPart = Record; type GeminiContent = { role: string; parts: GeminiPart[] }; -type GeminiGenerationConfig = { - temperature?: unknown; - topP?: unknown; - topK?: unknown; - maxOutputTokens?: unknown; - thinkingConfig?: { - thinkingBudget: number; - includeThoughts: boolean; - }; - responseMimeType?: string; - responseSchema?: unknown; - stopSequences?: string[] | unknown[]; -}; - type GeminiFunctionDeclaration = { name: string; description: string; @@ -118,124 +118,27 @@ type GeminiToolNameOptions = { supportsSignatureBypass?: boolean; }; -// Vertex AI (and Vertex Partner models) reject the OpenAI-style `id` field inside -// function_call / function_response parts. Detect these by the routed provider id. -function isVertexGeminiProvider(provider: unknown): boolean { - return provider === "vertex" || provider === "vertex-partner"; -} - -type OpenAIToolCallLike = { - thoughtSignature?: unknown; - thought_signature?: unknown; - function?: { - thoughtSignature?: unknown; - thought_signature?: unknown; - }; -}; - -function buildChangedToolNameMap(toolNameMap: Map): Map | null { - const changedEntries = [...toolNameMap.entries()].filter( - ([sanitizedName, originalName]) => sanitizedName !== originalName - ); - return changedEntries.length > 0 ? new Map(changedEntries) : null; -} - -function extractClientThoughtSignature(toolCall: unknown): string | null { - if (!toolCall || typeof toolCall !== "object") return null; - const candidate = toolCall as OpenAIToolCallLike; - - const signature = - candidate.thoughtSignature || - candidate.thought_signature || - candidate.function?.thoughtSignature || - candidate.function?.thought_signature || - null; - return typeof signature === "string" && signature.length > 0 ? signature : null; -} - -function deepCleanUndefined(value: unknown, depth = 0): void { - if (depth > 10 || !value || typeof value !== "object") { - return; - } - if (Array.isArray(value)) { - for (const item of value) { - deepCleanUndefined(item, depth + 1); - } - } else { - const obj = value as Record; - for (const key of Object.keys(obj)) { - const val = obj[key]; - if (typeof val === "string" && val === "[undefined]") { - delete obj[key]; - } else { - deepCleanUndefined(val, depth + 1); - } +// Gemini-family APIs (incl. Antigravity / Vertex) reject a `contents[]` array that +// has two adjacent entries with the same role: +// 400 INVALID_ARGUMENT "Request contains consecutive messages with the same role". +// Client history that carries consecutive user turns — or a tool-result turn (mapped +// to role:"user") immediately followed by a plain user turn — would otherwise leak +// that invalid alternation through. Merge adjacent same-role entries by concatenating +// their parts, the same normalization the Kiro and Claude request paths already apply +// (9router#2191). +export function mergeConsecutiveSameRoleContents(contents: GeminiContent[]): GeminiContent[] { + const merged: GeminiContent[] = []; + for (const entry of contents) { + const last = merged[merged.length - 1]; + if (last && last.role === entry.role) { + last.parts.push(...entry.parts); + } else { + // Shallow-copy the entry and its `parts` array so a later same-role merge + // (`last.parts.push(...)`) never mutates the caller's input objects. + merged.push({ ...entry, parts: [...entry.parts] }); } } -} - -function applyAntigravityGenerationDefaults(generationConfig: GeminiGenerationConfig) { - const config = { ...generationConfig }; - if (config.topK === undefined) { - config.topK = 40; - } - if (config.topP === undefined) { - config.topP = 1; - } - - const thinkingBudget = Number(config.thinkingConfig?.thinkingBudget); - const maxOutputTokens = Number(config.maxOutputTokens); - if ( - Number.isFinite(thinkingBudget) && - thinkingBudget > 0 && - (!Number.isFinite(maxOutputTokens) || maxOutputTokens <= thinkingBudget) - ) { - config.maxOutputTokens = Math.floor(thinkingBudget) + 1; - } - - return config; -} - -function stringifyHistoricalToolArguments(value: unknown): string { - if (typeof value === "string") return value; - try { - return JSON.stringify(value ?? {}); - } catch { - return String(value ?? "{}"); - } -} - -function buildInertHistoricalToolCallText(name: string | undefined, args: unknown): string { - const toolName = name || "unknown"; - return `[tool_history_call: ${toolName}] ${stringifyHistoricalToolArguments(args || "{}")}`; -} - -function buildInertHistoricalToolResponseText(name: string, response: unknown): string { - return `[tool_history_result: ${name || "unknown"}] ${typeof response === "string" ? response : stringifyHistoricalToolArguments(response)}`; -} - -function escapeHistoricalContextAttribute(value: string): string { - return value - .replaceAll("&", "&") - .replaceAll('"', """) - .replaceAll("<", "<") - .replaceAll(">", ">"); -} - -function escapeHistoricalContextContent(value: string): string { - return value.replaceAll("&", "&").replaceAll("<", "<").replaceAll(">", ">"); -} - -function buildHistoricalToolResultContext(name: string, response: unknown): string { - const source = escapeHistoricalContextAttribute(name || "unknown"); - const rawResult = - typeof response === "string" ? response : stringifyHistoricalToolArguments(response); - const result = escapeHistoricalContextContent(rawResult); - return [ - ``, - result, - "", - ].join("\n"); + return merged; } // Core: Convert OpenAI request to Gemini format (base for all variants) @@ -585,6 +488,9 @@ function openaiToGeminiBase( } } + // Collapse any consecutive same-role contents Gemini would reject (9router#2191). + result.contents = mergeConsecutiveSameRoleContents(result.contents ?? []); + // Convert tools const bodyTools = body.tools as Array> | undefined; const geminiTools = buildGeminiTools(bodyTools, { diff --git a/open-sse/translator/request/openai-to-gemini/helpers.ts b/open-sse/translator/request/openai-to-gemini/helpers.ts new file mode 100644 index 0000000000..3dfd35cef1 --- /dev/null +++ b/open-sse/translator/request/openai-to-gemini/helpers.ts @@ -0,0 +1,142 @@ +// Pure, self-contained helpers extracted verbatim from ../openai-to-gemini.ts +// (god-file decomposition): historical-tool-context string builders, undefined- +// pruning, thought-signature extraction, tool-name remapping, and the Vertex +// provider check + Antigravity generation-config defaults. No I/O or module state; +// the host imports them back internally (these were module-private — no public API +// change). The GeminiGenerationConfig shape lives here with its only mutator. + +export type GeminiGenerationConfig = { + temperature?: unknown; + topP?: unknown; + topK?: unknown; + maxOutputTokens?: unknown; + thinkingConfig?: { + thinkingBudget: number; + includeThoughts: boolean; + }; + responseMimeType?: string; + responseSchema?: unknown; + stopSequences?: string[] | unknown[]; +}; + +// Vertex AI (and Vertex Partner models) reject the OpenAI-style `id` field inside +// function_call / function_response parts. Detect these by the routed provider id. +export function isVertexGeminiProvider(provider: unknown): boolean { + return provider === "vertex" || provider === "vertex-partner"; +} + +type OpenAIToolCallLike = { + thoughtSignature?: unknown; + thought_signature?: unknown; + function?: { + thoughtSignature?: unknown; + thought_signature?: unknown; + }; +}; + +export function buildChangedToolNameMap( + toolNameMap: Map +): Map | null { + const changedEntries = [...toolNameMap.entries()].filter( + ([sanitizedName, originalName]) => sanitizedName !== originalName + ); + return changedEntries.length > 0 ? new Map(changedEntries) : null; +} + +export function extractClientThoughtSignature(toolCall: unknown): string | null { + if (!toolCall || typeof toolCall !== "object") return null; + const candidate = toolCall as OpenAIToolCallLike; + + const signature = + candidate.thoughtSignature || + candidate.thought_signature || + candidate.function?.thoughtSignature || + candidate.function?.thought_signature || + null; + return typeof signature === "string" && signature.length > 0 ? signature : null; +} + +export function deepCleanUndefined(value: unknown, depth = 0): void { + if (depth > 10 || !value || typeof value !== "object") { + return; + } + if (Array.isArray(value)) { + for (const item of value) { + deepCleanUndefined(item, depth + 1); + } + } else { + const obj = value as Record; + for (const key of Object.keys(obj)) { + const val = obj[key]; + if (typeof val === "string" && val === "[undefined]") { + delete obj[key]; + } else { + deepCleanUndefined(val, depth + 1); + } + } + } +} + +export function applyAntigravityGenerationDefaults(generationConfig: GeminiGenerationConfig) { + const config = { ...generationConfig }; + if (config.topK === undefined) { + config.topK = 40; + } + if (config.topP === undefined) { + config.topP = 1; + } + + const thinkingBudget = Number(config.thinkingConfig?.thinkingBudget); + const maxOutputTokens = Number(config.maxOutputTokens); + if ( + Number.isFinite(thinkingBudget) && + thinkingBudget > 0 && + (!Number.isFinite(maxOutputTokens) || maxOutputTokens <= thinkingBudget) + ) { + config.maxOutputTokens = Math.floor(thinkingBudget) + 1; + } + + return config; +} + +export function stringifyHistoricalToolArguments(value: unknown): string { + if (typeof value === "string") return value; + try { + return JSON.stringify(value ?? {}); + } catch { + return String(value ?? "{}"); + } +} + +export function buildInertHistoricalToolCallText(name: string | undefined, args: unknown): string { + const toolName = name || "unknown"; + return `[tool_history_call: ${toolName}] ${stringifyHistoricalToolArguments(args || "{}")}`; +} + +export function buildInertHistoricalToolResponseText(name: string, response: unknown): string { + return `[tool_history_result: ${name || "unknown"}] ${typeof response === "string" ? response : stringifyHistoricalToolArguments(response)}`; +} + +export function escapeHistoricalContextAttribute(value: string): string { + return value + .replaceAll("&", "&") + .replaceAll('"', """) + .replaceAll("<", "<") + .replaceAll(">", ">"); +} + +export function escapeHistoricalContextContent(value: string): string { + return value.replaceAll("&", "&").replaceAll("<", "<").replaceAll(">", ">"); +} + +export function buildHistoricalToolResultContext(name: string, response: unknown): string { + const source = escapeHistoricalContextAttribute(name || "unknown"); + const rawResult = + typeof response === "string" ? response : stringifyHistoricalToolArguments(response); + const result = escapeHistoricalContextContent(rawResult); + return [ + ``, + result, + "", + ].join("\n"); +} diff --git a/open-sse/translator/request/openai-to-kiro.ts b/open-sse/translator/request/openai-to-kiro.ts index 194484ded8..7a4c794b0d 100644 --- a/open-sse/translator/request/openai-to-kiro.ts +++ b/open-sse/translator/request/openai-to-kiro.ts @@ -158,7 +158,15 @@ function convertMessages(messages, tools, model) { const flushPending = () => { if (currentRole === "user") { - const content = pendingUserContent.join("\n\n").trim() || "(empty)"; + // Kiro accepts an empty user `content` when the turn carries toolResults or + // images (the agentic tool-loop case), so the "(empty)" placeholder is only + // needed for a genuinely bare turn. Without this check, a trailing + // tool-result-only turn (no follow-up user text) would get the literal + // "(empty)" injected as if the user had typed it, which can confuse Kiro. + // See decolua/9router#2183 for the same bug class. + const text = pendingUserContent.join("\n\n").trim(); + const hasContext = pendingToolResults.length > 0 || pendingImages.length > 0; + const content = text || (hasContext ? "" : "(empty)"); const userMsg: { userInputMessage: { content: string; @@ -669,8 +677,11 @@ export function buildKiroPayload(model, body, stream, credentials) { // Normalize model name: Claude Code sends dashes (claude-sonnet-4-6), // Kiro API expects dots (claude-sonnet-4.6). Convert trailing version segment. + // The minor group is bounded to 1-2 digits so date-suffixed ids (e.g. + // claude-opus-4-20250514) are never mistaken for a dash-separated minor + // version and corrupted into claude-opus-4.20250514 (upstream 9router #2270). const normalizedModel = model.replace( - /^(claude-(?:opus|sonnet|haiku|3-\d+)-\d+)-(\d+)$/, + /^(claude-(?:opus|sonnet|haiku|3-\d+)-\d+)-(\d{1,2})$/, "$1.$2" ); const messages = body.messages || []; diff --git a/open-sse/translator/response/openai-responses.ts b/open-sse/translator/response/openai-responses.ts index d6c397fbad..3a9a1299da 100644 --- a/open-sse/translator/response/openai-responses.ts +++ b/open-sse/translator/response/openai-responses.ts @@ -644,6 +644,51 @@ function computeFinishReason(state): "tool_calls" | "stop" { return (state.toolCallIndex || 0) > 0 || state.currentToolCallId ? "tool_calls" : "stop"; } +// #5786 — remember that a reasoning delta was streamed for a given reasoning item, so +// the terminal `response.output_item.done` snapshot for that item is NOT re-emitted +// (which would duplicate the reasoning channel). Keyed by item_id when present, with a +// global fallback for streams whose deltas carry no item_id. +function markResponsesReasoningDeltaEmitted(state, itemId) { + state.reasoningDeltaEmitted = true; + const id = itemId != null ? String(itemId) : ""; + if (!id) return; + if (!(state.reasoningItemsWithDelta instanceof Set)) { + state.reasoningItemsWithDelta = new Set(); + } + state.reasoningItemsWithDelta.add(id); +} + +// #5786 — build a Chat-format reasoning delta chunk in the shape the client renders in +// its thinking panel (`reasoning_content`, or `reasoning_text` for Copilot-compatible +// clients). Mirrors the `response.reasoning_summary_text.delta` branch. +function buildResponsesReasoningDeltaChunk(state, text) { + const delta = state.copilotCompatibleReasoning + ? { reasoning_text: text } + : { reasoning_content: text }; + return { + id: state.chatId, + object: "chat.completion.chunk", + created: state.created, + model: state.model || "gpt-4", + choices: [ + { + index: 0, + delta, + finish_reason: null, + }, + ], + }; +} + +function extractResponsesReasoningSummaryText(item) { + if (!item || !Array.isArray(item.summary)) return ""; + return item.summary + .map((part) => + part && typeof part === "object" && typeof part.text === "string" ? part.text : "" + ) + .join(""); +} + /** * Translate OpenAI Responses API chunk to OpenAI Chat Completions format * This is for when Codex returns data and we need to send it to an OpenAI-compatible client @@ -983,6 +1028,7 @@ function openaiResponsesToOpenAIResponseStream(chunk, state) { ) { const reasoningDelta = data.delta || ""; if (!reasoningDelta) return null; + markResponsesReasoningDeltaEmitted(state, data.item_id); return { id: state.chatId, object: "chat.completion.chunk", @@ -1007,22 +1053,33 @@ function openaiResponsesToOpenAIResponseStream(chunk, state) { if (eventType === "response.reasoning_summary_text.delta") { const reasoningDelta = data.delta || ""; if (!reasoningDelta) return null; - const reasoningDeltaShape = state.copilotCompatibleReasoning - ? { reasoning_text: reasoningDelta } - : { reasoning_content: reasoningDelta }; - return { - id: state.chatId, - object: "chat.completion.chunk", - created: state.created, - model: state.model || "gpt-4", - choices: [ - { - index: 0, - delta: reasoningDeltaShape, - finish_reason: null, - }, - ], - }; + markResponsesReasoningDeltaEmitted(state, data.item_id); + return buildResponsesReasoningDeltaChunk(state, reasoningDelta); + } + + // #5786 — reasoning summary exposed ONLY as a terminal snapshot on + // `response.output_item.done` (no preceding reasoning_summary_text.delta events — e.g. + // Codex reasoning models that surface the summary once at item close). Without this the + // reasoning channel is silently dropped and never reaches the client's thinking panel. + // Only synthesize when NO reasoning delta was already streamed for this item, so normal + // delta streams are never duplicated. + if (eventType === "response.output_item.done" && data.item?.type === "reasoning") { + const item = data.item; + const itemId = item.id != null ? String(item.id) : ""; + const emittedForItem = + state.reasoningItemsWithDelta instanceof Set && + itemId && + state.reasoningItemsWithDelta.has(itemId); + // Deltas were streamed but carried no item_id: fall back to the global flag and + // suppress synthesis to avoid duplicating that same reasoning text. + const emittedWithoutItemId = + state.reasoningDeltaEmitted && + !(state.reasoningItemsWithDelta instanceof Set && state.reasoningItemsWithDelta.size > 0); + if (emittedForItem || emittedWithoutItemId) return null; + + const summaryText = extractResponsesReasoningSummaryText(item); + if (!summaryText) return null; + return buildResponsesReasoningDeltaChunk(state, summaryText); } // Ignore other events diff --git a/open-sse/translator/response/openai-to-claude.ts b/open-sse/translator/response/openai-to-claude.ts index edc12816b8..e899778d4c 100644 --- a/open-sse/translator/response/openai-to-claude.ts +++ b/open-sse/translator/response/openai-to-claude.ts @@ -225,8 +225,15 @@ export function openaiToClaudeResponse(chunk, state) { } } - // Finish - if (choice.finish_reason) { + // Finish — guard against duplicate finish_reason chunks (common with OpenAI-compatible models). + // Use a dedicated `claudeFinishEmitted` flag rather than `state.finishReason`: in the + // Responses→Claude hub path the shared `state` object is also written by the + // openai-responses→openai translator, which sets `state.finishReason` on + // `response.completed` BEFORE this openai→claude step runs. Reusing `finishReason` as the + // guard therefore misfired and silently dropped the terminal message_delta/message_stop + // for Responses→Claude streams (#5828 regression). + if (choice.finish_reason && !state.claudeFinishEmitted) { + state.claudeFinishEmitted = true; stopThinkingBlock(state, results); stopTextBlock(state, results); diff --git a/open-sse/utils/agentGoalPolicy.ts b/open-sse/utils/agentGoalPolicy.ts new file mode 100644 index 0000000000..3f5aaf4f2a --- /dev/null +++ b/open-sse/utils/agentGoalPolicy.ts @@ -0,0 +1,112 @@ +type EnvSource = Record; +type HeaderLike = Headers | Record | null | undefined; + +export type AgentGoalPolicy = { + detected: boolean; + readinessMaxTimeoutMs: number; + streamRecoveryEnabled: boolean; +}; + +export const DEFAULT_AGENT_GOAL_READINESS_MAX_TIMEOUT_MS = 600_000; + +const GOAL_COMMAND_RE = /(^|[\s"'`])\/goal(?=$|[\s"'`:;,.!?])/i; +const MAX_VISITED_NODES = 5_000; +const MAX_STRING_CHARS = 256_000; + +function readHeader(headers: HeaderLike, name: string): string | null { + if (!headers) return null; + if (typeof (headers as Headers).get === "function") { + return (headers as Headers).get(name); + } + const lower = name.toLowerCase(); + for (const [key, value] of Object.entries(headers)) { + if (key.toLowerCase() !== lower) continue; + if (Array.isArray(value)) return value.join(","); + return typeof value === "string" ? value : null; + } + return null; +} + +function parseBoolean(value: string | undefined | null, fallback: boolean): boolean { + if (value == null || value.trim() === "") return fallback; + const normalized = value.trim().toLowerCase(); + if (["true", "1", "yes", "on"].includes(normalized)) return true; + if (["false", "0", "no", "off"].includes(normalized)) return false; + return fallback; +} + +function readPositiveMs(env: EnvSource, name: string, fallback: number): number { + const raw = env[name]; + if (raw == null || raw.trim() === "") return fallback; + const parsed = Number(raw); + return Number.isFinite(parsed) && parsed > 0 ? Math.floor(parsed) : fallback; +} + +function hasGoalCommandText(value: string): boolean { + const sample = value.length > MAX_STRING_CHARS ? value.slice(0, MAX_STRING_CHARS) : value; + return GOAL_COMMAND_RE.test(sample); +} + +export function isAgentGoalRequestBody(body: unknown): boolean { + const seen = new Set(); + let visited = 0; + + const visit = (value: unknown, depth: number): boolean => { + if (visited++ > MAX_VISITED_NODES || depth > 10) return false; + if (typeof value === "string") return hasGoalCommandText(value); + if (!value || typeof value !== "object") return false; + if (seen.has(value)) return false; + seen.add(value); + + if (Array.isArray(value)) { + for (const item of value) { + if (visit(item, depth + 1)) return true; + } + return false; + } + + for (const [key, child] of Object.entries(value as Record)) { + if (key === "metadata" || key === "usage") continue; + if (visit(child, depth + 1)) return true; + } + return false; + }; + + return visit(body, 0); +} + +export function resolveAgentGoalPolicy( + body: unknown, + headers: HeaderLike = null, + env: EnvSource = process.env +): AgentGoalPolicy { + // Kill-switch (default ON — preserves existing behavior). When explicitly + // disabled, the whole heuristic is a no-op: it never elevates readiness + // timeouts or stream recovery, regardless of request body/headers. This + // mitigates client-controlled timeout amplification when an operator does + // not want request bodies/headers to influence upstream timeout budgets. + const policyEnabled = parseBoolean(env.OMNIROUTE_AGENT_GOAL_POLICY_ENABLED, true); + if (!policyEnabled) { + return { + detected: false, + readinessMaxTimeoutMs: DEFAULT_AGENT_GOAL_READINESS_MAX_TIMEOUT_MS, + streamRecoveryEnabled: false, + }; + } + + const forcedByHeader = parseBoolean(readHeader(headers, "x-omniroute-agent-goal"), false); + const detected = forcedByHeader || isAgentGoalRequestBody(body); + const readinessMaxTimeoutMs = readPositiveMs( + env, + "OMNIROUTE_AGENT_GOAL_READINESS_MAX_TIMEOUT_MS", + DEFAULT_AGENT_GOAL_READINESS_MAX_TIMEOUT_MS + ); + const streamRecoveryEnabled = + detected && parseBoolean(env.OMNIROUTE_AGENT_GOAL_STREAM_RECOVERY, true); + + return { + detected, + readinessMaxTimeoutMs, + streamRecoveryEnabled, + }; +} diff --git a/open-sse/utils/cursorAgentProtobuf.ts b/open-sse/utils/cursorAgentProtobuf.ts index 24cee2332b..bb17d9c944 100644 --- a/open-sse/utils/cursorAgentProtobuf.ts +++ b/open-sse/utils/cursorAgentProtobuf.ts @@ -18,6 +18,25 @@ import zlib from "node:zlib"; import crypto from "node:crypto"; +import { + WT_VARINT, + WT_LEN, + encodeVarint, + encodeTag, + encodeBytes, + encodeString, + encodeMessage, + encodeUInt32Field, + encodeBoolField, + encodeDoubleField, + decodeVarint, + checkedLen, + decodeFields, + findField, + decodeStringField, + decodeVarintField, + type Field, +} from "./cursorAgentProtobuf/wire.ts"; // ─── Field numbers (from agent.proto descriptor) ─────────────────────────── @@ -225,125 +244,6 @@ const LIST_VALUES = 1; // ListValue.values = repeated Value const MAP_KEY = 1; const MAP_VALUE = 2; -// ─── Wire-type constants ─────────────────────────────────────────────────── - -const WT_VARINT = 0; -const WT_LEN = 2; - -// ─── Primitive encoders ──────────────────────────────────────────────────── - -function encodeVarint(value: number | bigint): Buffer { - let v = typeof value === "bigint" ? value : BigInt(value); - const bytes: number[] = []; - while (v > 0x7fn) { - bytes.push(Number(v & 0x7fn) | 0x80); - v >>= 7n; - } - bytes.push(Number(v)); - return Buffer.from(bytes); -} - -function encodeTag(fieldNumber: number, wireType: number): Buffer { - return encodeVarint((fieldNumber << 3) | wireType); -} - -function encodeBytes(fieldNumber: number, value: Buffer | Uint8Array): Buffer { - const buf = Buffer.isBuffer(value) ? value : Buffer.from(value); - return Buffer.concat([encodeTag(fieldNumber, WT_LEN), encodeVarint(buf.length), buf]); -} - -function encodeString(fieldNumber: number, value: string): Buffer { - return encodeBytes(fieldNumber, Buffer.from(value, "utf8")); -} - -function encodeMessage(fieldNumber: number, parts: Buffer[]): Buffer { - const inner = Buffer.concat(parts); - return Buffer.concat([encodeTag(fieldNumber, WT_LEN), encodeVarint(inner.length), inner]); -} - -function encodeUInt32Field(fieldNumber: number, value: number): Buffer { - return Buffer.concat([encodeTag(fieldNumber, WT_VARINT), encodeVarint(value)]); -} - -function encodeBoolField(fieldNumber: number, value: boolean): Buffer { - return Buffer.concat([encodeTag(fieldNumber, WT_VARINT), encodeVarint(value ? 1 : 0)]); -} - -function encodeDoubleField(fieldNumber: number, value: number): Buffer { - // wire type 1 = 64-bit fixed (double) - const buf = Buffer.alloc(8); - buf.writeDoubleLE(value, 0); - return Buffer.concat([encodeTag(fieldNumber, 1), buf]); -} - -// ─── Primitive decoders ──────────────────────────────────────────────────── - -function decodeVarint(buf: Buffer, offset: number): [bigint, number] { - let result = 0n; - let shift = 0n; - let pos = offset; - while (pos < buf.length) { - const byte = buf[pos++]; - result |= BigInt(byte & 0x7f) << shift; - if ((byte & 0x80) === 0) return [result, pos]; - shift += 7n; - } - throw new Error("varint truncated"); -} - -type Field = - | { fieldNumber: number; wireType: 0; varint: bigint } - | { fieldNumber: number; wireType: 2; bytes: Buffer }; - -/** - * Validate a length-delimited field's declared length against the bytes that - * actually remain in the buffer. Cursor's frames are well-formed, but a - * corrupted or hostile upstream could declare a length that overruns the - * buffer; without this guard `Buffer.subarray` silently clamps to EOF and a - * truncated tool argument (or any nested message) is decoded as empty/partial - * data instead of being recognized as malformed. Throwing lets the caller — - * `processFrame`, wrapped in driveH2's per-frame try/catch — skip the bad - * frame rather than act on corrupted fields. Also rejects absurd lengths that - * would not fit a JS safe integer. - */ -function checkedLen(len: bigint, pos: number, buf: Buffer): number { - if (len < 0n || len > BigInt(buf.length - pos)) { - throw new Error( - `length-delimited field overruns buffer (len=${len}, remaining=${buf.length - pos})` - ); - } - return Number(len); -} - -function decodeFields(buf: Buffer): Field[] { - const fields: Field[] = []; - let pos = 0; - while (pos < buf.length) { - const [tag, np] = decodeVarint(buf, pos); - pos = np; - const fieldNumber = Number(tag >> 3n); - const wireType = Number(tag & 0x7n); - if (wireType === WT_VARINT) { - const [v, np2] = decodeVarint(buf, pos); - pos = np2; - fields.push({ fieldNumber, wireType: 0, varint: v }); - } else if (wireType === WT_LEN) { - const [len, np2] = decodeVarint(buf, pos); - pos = np2; - const lenN = checkedLen(len, pos, buf); - fields.push({ fieldNumber, wireType: 2, bytes: buf.subarray(pos, pos + lenN) }); - pos += lenN; - } else if (wireType === 5) { - pos += 4; - } else if (wireType === 1) { - pos += 8; - } else { - throw new Error(`unsupported wireType ${wireType}`); - } - } - return fields; -} - // ─── Connect-RPC framing ─────────────────────────────────────────────────── const FLAG_NONE = 0x00; @@ -547,9 +447,7 @@ export function encodeAgentRunRequest(input: AgentRunInput): Buffer { const selectedContextParts: Buffer[] = []; if (input.images && input.images.length > 0) { for (const img of input.images) { - selectedContextParts.push( - encodeMessage(SC_SELECTED_IMAGES, [encodeSelectedImageBody(img)]) - ); + selectedContextParts.push(encodeMessage(SC_SELECTED_IMAGES, [encodeSelectedImageBody(img)])); } } // The empty selected_context placeholder and mode=1 match cursor-agent's @@ -652,24 +550,6 @@ export type DecodedDelta = | { kind: "kv_server_message" } | { kind: "unknown"; field: number }; -function findField(fields: Field[], fieldNumber: number): Field | undefined { - return fields.find((f) => f.fieldNumber === fieldNumber); -} - -function decodeStringField(buf: Buffer, fieldNumber: number): string { - const fields = decodeFields(buf); - const f = findField(fields, fieldNumber); - if (f && f.wireType === 2) return f.bytes.toString("utf8"); - return ""; -} - -function decodeVarintField(buf: Buffer, fieldNumber: number): number { - const fields = decodeFields(buf); - const f = findField(fields, fieldNumber); - if (f && f.wireType === 0) return Number(f.varint); - return 0; -} - export function decodeAgentServerMessage(payload: Buffer): DecodedDelta[] { const out: DecodedDelta[] = []; for (const top of decodeFields(payload)) { diff --git a/open-sse/utils/cursorAgentProtobuf/wire.ts b/open-sse/utils/cursorAgentProtobuf/wire.ts new file mode 100644 index 0000000000..bd2d0a9697 --- /dev/null +++ b/open-sse/utils/cursorAgentProtobuf/wire.ts @@ -0,0 +1,143 @@ +// Low-level protobuf wire-format primitives for the Cursor Agent codec, extracted +// verbatim from ../cursorAgentProtobuf.ts (god-file decomposition). Pure and +// dependency-free (Buffer only): varint/tag/length-delimited encode+decode and the +// generic field walker. Framing, the value codec, and the message encoders/decoders +// all build on this layer. Nothing here was part of the module's public API, so the +// host imports these back internally (no re-export). + +// ─── Wire-type constants ─────────────────────────────────────────────────── + +export const WT_VARINT = 0; +export const WT_LEN = 2; + +// ─── Primitive encoders ──────────────────────────────────────────────────── + +export function encodeVarint(value: number | bigint): Buffer { + let v = typeof value === "bigint" ? value : BigInt(value); + const bytes: number[] = []; + while (v > 0x7fn) { + bytes.push(Number(v & 0x7fn) | 0x80); + v >>= 7n; + } + bytes.push(Number(v)); + return Buffer.from(bytes); +} + +export function encodeTag(fieldNumber: number, wireType: number): Buffer { + return encodeVarint((fieldNumber << 3) | wireType); +} + +export function encodeBytes(fieldNumber: number, value: Buffer | Uint8Array): Buffer { + const buf = Buffer.isBuffer(value) ? value : Buffer.from(value); + return Buffer.concat([encodeTag(fieldNumber, WT_LEN), encodeVarint(buf.length), buf]); +} + +export function encodeString(fieldNumber: number, value: string): Buffer { + return encodeBytes(fieldNumber, Buffer.from(value, "utf8")); +} + +export function encodeMessage(fieldNumber: number, parts: Buffer[]): Buffer { + const inner = Buffer.concat(parts); + return Buffer.concat([encodeTag(fieldNumber, WT_LEN), encodeVarint(inner.length), inner]); +} + +export function encodeUInt32Field(fieldNumber: number, value: number): Buffer { + return Buffer.concat([encodeTag(fieldNumber, WT_VARINT), encodeVarint(value)]); +} + +export function encodeBoolField(fieldNumber: number, value: boolean): Buffer { + return Buffer.concat([encodeTag(fieldNumber, WT_VARINT), encodeVarint(value ? 1 : 0)]); +} + +export function encodeDoubleField(fieldNumber: number, value: number): Buffer { + // wire type 1 = 64-bit fixed (double) + const buf = Buffer.alloc(8); + buf.writeDoubleLE(value, 0); + return Buffer.concat([encodeTag(fieldNumber, 1), buf]); +} + +// ─── Primitive decoders ──────────────────────────────────────────────────── + +export function decodeVarint(buf: Buffer, offset: number): [bigint, number] { + let result = 0n; + let shift = 0n; + let pos = offset; + while (pos < buf.length) { + const byte = buf[pos++]; + result |= BigInt(byte & 0x7f) << shift; + if ((byte & 0x80) === 0) return [result, pos]; + shift += 7n; + } + throw new Error("varint truncated"); +} + +export type Field = + | { fieldNumber: number; wireType: 0; varint: bigint } + | { fieldNumber: number; wireType: 2; bytes: Buffer }; + +/** + * Validate a length-delimited field's declared length against the bytes that + * actually remain in the buffer. Cursor's frames are well-formed, but a + * corrupted or hostile upstream could declare a length that overruns the + * buffer; without this guard `Buffer.subarray` silently clamps to EOF and a + * truncated tool argument (or any nested message) is decoded as empty/partial + * data instead of being recognized as malformed. Throwing lets the caller — + * `processFrame`, wrapped in driveH2's per-frame try/catch — skip the bad + * frame rather than act on corrupted fields. Also rejects absurd lengths that + * would not fit a JS safe integer. + */ +export function checkedLen(len: bigint, pos: number, buf: Buffer): number { + if (len < 0n || len > BigInt(buf.length - pos)) { + throw new Error( + `length-delimited field overruns buffer (len=${len}, remaining=${buf.length - pos})` + ); + } + return Number(len); +} + +export function decodeFields(buf: Buffer): Field[] { + const fields: Field[] = []; + let pos = 0; + while (pos < buf.length) { + const [tag, np] = decodeVarint(buf, pos); + pos = np; + const fieldNumber = Number(tag >> 3n); + const wireType = Number(tag & 0x7n); + if (wireType === WT_VARINT) { + const [v, np2] = decodeVarint(buf, pos); + pos = np2; + fields.push({ fieldNumber, wireType: 0, varint: v }); + } else if (wireType === WT_LEN) { + const [len, np2] = decodeVarint(buf, pos); + pos = np2; + const lenN = checkedLen(len, pos, buf); + fields.push({ fieldNumber, wireType: 2, bytes: buf.subarray(pos, pos + lenN) }); + pos += lenN; + } else if (wireType === 5) { + pos += 4; + } else if (wireType === 1) { + pos += 8; + } else { + throw new Error(`unsupported wireType ${wireType}`); + } + } + return fields; +} + +export function findField(fields: Field[], fieldNumber: number): Field | undefined { + return fields.find((f) => f.fieldNumber === fieldNumber); +} + +export function decodeStringField(buf: Buffer, fieldNumber: number): string { + const fields = decodeFields(buf); + const f = findField(fields, fieldNumber); + if (f && f.wireType === 2) return f.bytes.toString("utf8"); + return ""; +} + +export function decodeVarintField(buf: Buffer, fieldNumber: number): number { + const fields = decodeFields(buf); + const f = findField(fields, fieldNumber); + if (f && f.wireType === 0) return Number(f.varint); + return 0; +} diff --git a/open-sse/utils/diagnostics.ts b/open-sse/utils/diagnostics.ts index d84eabdf79..176d0f7843 100644 --- a/open-sse/utils/diagnostics.ts +++ b/open-sse/utils/diagnostics.ts @@ -256,6 +256,24 @@ export function detectMalformedNonStream(resp: unknown): MalformedReason | null const c = choice as Record; const msg = c?.message as Record | undefined; if (typeof msg?.content === "string" && (msg.content as string).length > 0) return true; + // #5559: some OpenAI-compatible upstreams (e.g. Cline via OAuth) return + // `message.content` as an array of Anthropic-style content blocks rather than + // a plain string. An array with at least one non-empty text block is real + // output — without this it was falsely flagged as empty_choices → 502 + cooldown. + if ( + Array.isArray(msg?.content) && + (msg.content as unknown[]).some((block) => { + const b = block as Record | null; + return ( + !!b && + typeof b === "object" && + b.type === "text" && + typeof b.text === "string" && + (b.text as string).length > 0 + ); + }) + ) + return true; if (Array.isArray(msg?.tool_calls) && (msg.tool_calls as unknown[]).length > 0) return true; if (typeof msg?.reasoning_content === "string" && (msg.reasoning_content as string).length > 0) return true; diff --git a/open-sse/utils/opencodeHeaders.ts b/open-sse/utils/opencodeHeaders.ts new file mode 100644 index 0000000000..17746522a4 --- /dev/null +++ b/open-sse/utils/opencodeHeaders.ts @@ -0,0 +1,67 @@ +import { randomUUID } from "crypto"; +import { setUserAgentHeader } from "../executors/base.ts"; + +/** + * Header keys that are forwarded from the client to the upstream provider. + * Used by both OpencodeExecutor and DefaultExecutor. + */ +const OPENCODE_HEADER_KEYS = [ + "x-opencode-session", + "x-opencode-request", + "x-opencode-project", + "x-opencode-client", +] as const; + +/** + * Case-insensitive lookup for a header in a headers record. + */ +function findHeader(headers: Record, name: string): string | undefined { + return Object.entries(headers).find(([key]) => key.toLowerCase() === name.toLowerCase())?.[1]; +} + +/** + * Forward OpenCode client request metadata headers to the upstream provider. + * + * Shared logic used by OpencodeExecutor and DefaultExecutor: + * 1. Forwards User-Agent from clientHeaders via `setUserAgentHeader()` + * 2. Forwards x-opencode-session, x-opencode-request, x-opencode-project, + * x-opencode-client headers (case-insensitive match) + * + * @param headers - The outbound headers record to mutate + * @param clientHeaders - The client-provided headers to forward from + * @param options.synthesizeRequestId - When true (OpencodeExecutor only), maps + * x-session-affinity / x-session-id to x-opencode-session when the latter is + * missing, and synthesizes a UUID for x-opencode-request if also missing. + */ +export function forwardOpencodeClientHeaders( + headers: Record, + clientHeaders: Record, + options?: { synthesizeRequestId?: boolean } +): void { + // 1. Forward User-Agent + const clientUA = clientHeaders["User-Agent"] || clientHeaders["user-agent"]; + if (clientUA) { + setUserAgentHeader(headers, clientUA); + } + + // 2. Forward x-opencode-* metadata headers + for (const headerName of OPENCODE_HEADER_KEYS) { + const value = findHeader(clientHeaders, headerName); + if (value) { + headers[headerName] = value; + } + } + + // 3. OpencodeExecutor-only: synthesize session/request id from fallback headers + if (options?.synthesizeRequestId && !headers["x-opencode-session"]) { + const sessionAffinity = + findHeader(clientHeaders, "x-session-affinity") || findHeader(clientHeaders, "x-session-id"); + if (sessionAffinity) { + headers["x-opencode-session"] = sessionAffinity; + + if (!headers["x-opencode-request"]) { + headers["x-opencode-request"] = randomUUID(); + } + } + } +} diff --git a/open-sse/utils/progressTracker.ts b/open-sse/utils/progressTracker.ts index aaaede2223..fda582d12d 100644 --- a/open-sse/utils/progressTracker.ts +++ b/open-sse/utils/progressTracker.ts @@ -90,6 +90,10 @@ export function createProgressTransform({ } } }, + + cancel() { + clearInterval(intervalId); + }, }, { highWaterMark: 16384 }, { highWaterMark: 16384 } diff --git a/open-sse/utils/proxyFetch.ts b/open-sse/utils/proxyFetch.ts index 134aaf8311..62ed9356be 100644 --- a/open-sse/utils/proxyFetch.ts +++ b/open-sse/utils/proxyFetch.ts @@ -107,6 +107,29 @@ export function describeFetchCause(err: unknown): string { return parts.join(" | ") || String(err); } + +function isStreamLikeBody(body: unknown): boolean { + return ( + body !== null && + body !== undefined && + typeof body === "object" && + (typeof (body as Record).getReader === "function" || + typeof (body as Record).stream === "function") + ); +} + +function requestHasNonReplayableBody( + input: RequestInfo | URL, + options: FetchWithDispatcherOptions +): boolean { + if (isStreamLikeBody(options.body as unknown)) return true; + if (typeof Request !== "undefined" && input instanceof Request) { + if (input.bodyUsed) return true; + if (input.body !== null) return true; + } + return false; +} + /** Injectable dependencies for testability (Approach B DI). */ export type ProxyFetchDeps = { undiciFetch?: FetchWithDispatcher; @@ -436,17 +459,13 @@ async function patchedFetch( // Falls back to original native fetch if dispatcher initialization fails (#1054). // Retries once on transient dispatcher errors before falling back (fix: proxyfetch-undici-retry). // - // ReadableStream/Blob body guard: if the body is non-replayable, skip the retry because - // the first attempt drains the stream; a second attempt would silently send an empty body. - // ReadableStream check: cast through unknown to avoid explicit-any budget (T11). - const _bodyUnknown = options.body as unknown; - const bodyIsStream = - _bodyUnknown !== null && - _bodyUnknown !== undefined && - typeof _bodyUnknown === "object" && - (typeof (_bodyUnknown as Record).getReader === "function" || // ReadableStream - typeof (_bodyUnknown as Record).stream === "function"); // Blob - const maxAttempts = bodyIsStream ? 1 : 2; + // Non-replayable body guard: if the body is stream-like (ReadableStream/Blob) + // or the input is a Request that carries a body, the first dispatcher attempt + // owns that body. Retrying or falling back to native fetch would replay a + // consumed/locked body and can mask the original transport error with + // "Response body object should not be disturbed or locked". + const hasNonReplayableBody = requestHasNonReplayableBody(input, options); + const maxAttempts = hasNonReplayableBody ? 1 : 2; const _undiciDirect = deps.undiciFetch ?? (undiciFetch as unknown as (...args: unknown[]) => Promise); const _nativeFallback = @@ -492,6 +511,17 @@ async function patchedFetch( await new Promise((r) => setTimeout(r, 25 + Math.random() * 50)); continue; } + if (hasNonReplayableBody) { + const detail = `dispatcher=[${describeFetchCause(dispatcherError)}] native=[skipped: non-replayable request body]`; + console.warn( + `[ProxyFetch] skipping native fetch fallback for non-replayable body: ${detail}` + ); + if (dispatcherError instanceof Error) { + (dispatcherError as Error & { proxyFetchDetail?: string }).proxyFetchDetail = detail; + } + throw dispatcherError; + } + // All attempts exhausted — try proxy fallback before native fetch if (source === "direct" && isFeatureFlagEnabled("PROXY_AUTO_SELECT_ENABLED")) { let targetHostname = ""; diff --git a/open-sse/utils/requestLogger.ts b/open-sse/utils/requestLogger.ts index bdb78be2fb..560c65a1a6 100644 --- a/open-sse/utils/requestLogger.ts +++ b/open-sse/utils/requestLogger.ts @@ -270,7 +270,8 @@ function makeStreamChunkMethods(options: RequestLoggerOptions, captureChunks: bo const append = (arr: string[], bytes: { value: number; truncated: boolean }, chunk: string) => { if (!captureChunks) return; push(); - appendBoundedChunk(arr, bytes, chunk, maxBytes, maxItems); + const ts = new Date().toISOString().slice(11, 23); + appendBoundedChunk(arr, bytes, `[${ts}] ${chunk}`, maxBytes, maxItems); }; return { diff --git a/open-sse/utils/sseHeartbeat.ts b/open-sse/utils/sseHeartbeat.ts index f0ef75a26f..9a12214145 100644 --- a/open-sse/utils/sseHeartbeat.ts +++ b/open-sse/utils/sseHeartbeat.ts @@ -111,5 +111,9 @@ export function createSseHeartbeatTransform({ flush() { stop(); }, + + cancel() { + stop(); + }, }); } diff --git a/open-sse/utils/stream.ts b/open-sse/utils/stream.ts index b6feb86e0a..8357ce2506 100644 --- a/open-sse/utils/stream.ts +++ b/open-sse/utils/stream.ts @@ -684,6 +684,18 @@ export function createSSEStream(options: StreamOptions = {}) { let passthroughResponsesId: string | null = null; let passthroughResponsesCurrentFunctionCallKey: string | null = null; const passthroughResponsesReasoningSummarySeen = new Set(); + // #5786 — highest Responses-API `sequence_number` already forwarded on this stream. + // The Responses API guarantees a strictly increasing sequence_number, so any event at + // or below this watermark is an upstream reconnect/retry replay and must be dropped — + // otherwise the replayed deltas glue duplicated text into the client stream. Applies to + // both translate mode (openai-responses → claude/openai) and Responses passthrough. + let lastSeenResponsesSequenceNumber = -1; + const isDuplicateResponsesSequence = (value: unknown): boolean => { + if (typeof value !== "number" || !Number.isFinite(value)) return false; + if (value <= lastSeenResponsesSequenceNumber) return true; + lastSeenResponsesSequenceNumber = value; + return false; + }; const streamStartedAt = Date.now(); let lastToolCallChunkTime: number | null = null; @@ -1185,6 +1197,18 @@ export function createSSEStream(options: StreamOptions = {}) { }) : null; + // #5786 — drop replayed Responses-API events (a re-sent event carrying an + // already-seen sequence_number) so their deltas are not forwarded twice. + if ( + parsedPassthroughData && + typeof parsedPassthroughData.type === "string" && + parsedPassthroughData.type.startsWith("response.") && + isDuplicateResponsesSequence(parsedPassthroughData.sequence_number) + ) { + clearPendingPassthroughEvent(); + continue; + } + if (trimmed.startsWith("data:")) { const providerPayload = parsedPassthroughData ?? parseSSELine(trimmed); if (providerPayload) { @@ -1859,6 +1883,17 @@ export function createSSEStream(options: StreamOptions = {}) { const parsed = parseSSELine(trimmed); if (!parsed) continue; + + // #5786 — drop replayed Responses-API events (identical/lower sequence_number + // re-sent on an upstream reconnect) so their deltas are not glued twice into + // the translated client stream. + if ( + targetFormat === FORMATS.OPENAI_RESPONSES && + isDuplicateResponsesSequence((parsed as JsonRecord).sequence_number) + ) { + continue; + } + providerPayloadCollector.push(parsed); if (parsed && parsed.done) { diff --git a/open-sse/utils/streamPayloadCollector.ts b/open-sse/utils/streamPayloadCollector.ts index 630c964e89..26e9f4418e 100644 --- a/open-sse/utils/streamPayloadCollector.ts +++ b/open-sse/utils/streamPayloadCollector.ts @@ -3,6 +3,7 @@ import { FORMATS } from "../translator/formats.ts"; type StructuredSSEEvent = { index: number; + timestamp?: string; event?: string; data: unknown; }; @@ -356,9 +357,21 @@ function buildClaudeSummary(events: StructuredSSEEvent[], fallbackModel?: string let role = "assistant"; let stopReason = "end_turn"; let stopSequence: string | null = null; + // Context Editing (`anthropic-beta: context-management-2025-06-27`) surfaces + // `context_management.applied_edits[]` on the final `message_delta` snapshot. Preserve it + // so streaming context-clear savings reach `extractContextEditingTelemetry`, mirroring the + // non-streaming JSON path. Last-writer-wins: the final snapshot is authoritative. + let contextManagement: JsonRecord | null = null; for (const payload of payloads) { const eventType = toString(payload.type); + if ( + payload.context_management && + typeof payload.context_management === "object" && + !Array.isArray(payload.context_management) + ) { + contextManagement = asRecord(payload.context_management); + } if (eventType === "message_start") { const message = asRecord(payload.message); messageId = toString(message.id, messageId || `msg_${Date.now()}`); @@ -506,6 +519,7 @@ function buildClaudeSummary(events: StructuredSSEEvent[], fallbackModel?: string stop_reason: stopReason, ...(stopSequence ? { stop_sequence: stopSequence } : {}), ...(Object.keys(usage).length > 0 ? { usage } : {}), + ...(contextManagement ? { context_management: contextManagement } : {}), }; } @@ -647,6 +661,7 @@ export function createStructuredSSECollector(options: CollectorOptions = {}) { const event: StructuredSSEEvent = { index: events.length + droppedEvents, + timestamp: new Date().toISOString(), data: cloneLogPayload(payload), }; diff --git a/open-sse/utils/streamReadinessPolicy.ts b/open-sse/utils/streamReadinessPolicy.ts index 065256044b..56e58a03b9 100644 --- a/open-sse/utils/streamReadinessPolicy.ts +++ b/open-sse/utils/streamReadinessPolicy.ts @@ -14,7 +14,7 @@ export type StreamReadinessPolicyResult = { reasons: string[]; }; -const DEFAULT_MAX_TIMEOUT_MS = 120_000; +const DEFAULT_MAX_TIMEOUT_MS = 180_000; const LARGE_ITEM_THRESHOLD = 150; const VERY_LARGE_ITEM_THRESHOLD = 400; const TOOL_HEAVY_THRESHOLD = 15; diff --git a/open-sse/utils/urlSanitize.ts b/open-sse/utils/urlSanitize.ts index d598d268cb..3cdb8ddd90 100644 --- a/open-sse/utils/urlSanitize.ts +++ b/open-sse/utils/urlSanitize.ts @@ -12,3 +12,13 @@ export function stripTrailingSlashes(value: string): string { } return end === value.length ? value : value.slice(0, end); } + +/** + * Normalize a base URL by trimming whitespace and stripping trailing slashes. + * Handles non-string inputs gracefully (returns empty string). + * Single source of truth — replaces per-file inline copies in config/*.ts. + */ +export function normalizeBaseUrl(value: string | null | undefined): string { + const str = typeof value === "string" ? value : ""; + return stripTrailingSlashes(str.trim()); +} diff --git a/package-lock.json b/package-lock.json index b3be55e1f6..655662db13 100644 --- a/package-lock.json +++ b/package-lock.json @@ -1,12 +1,12 @@ { "name": "omniroute", - "version": "3.8.42", + "version": "3.8.43", "lockfileVersion": 3, "requires": true, "packages": { "": { "name": "omniroute", - "version": "3.8.42", + "version": "3.8.43", "hasInstallScript": true, "license": "MIT", "workspaces": [ @@ -107,6 +107,7 @@ "@types/safe-regex": "^1.1.6", "@types/ws": "^8.18.0", "@vitejs/plugin-react": "^6.0.2", + "bun": "1.3.10", "c8": "^11.0.0", "concurrently": "^10.0.3", "cross-env": "^10.1.0", @@ -4944,6 +4945,174 @@ "node": ">= 20.0.0" } }, + "node_modules/@oven/bun-darwin-aarch64": { + "version": "1.3.10", + "resolved": "https://registry.npmjs.org/@oven/bun-darwin-aarch64/-/bun-darwin-aarch64-1.3.10.tgz", + "integrity": "sha512-PXgg5gqcS/rHwa1hF0JdM1y5TiyejVrMHoBmWY/DjtfYZoFTXie1RCFOkoG0b5diOOmUcuYarMpH7CSNTqwj+w==", + "cpu": [ + "arm64" + ], + "dev": true, + "license": "MIT", + "optional": true, + "os": [ + "darwin" + ] + }, + "node_modules/@oven/bun-darwin-x64": { + "version": "1.3.10", + "resolved": "https://registry.npmjs.org/@oven/bun-darwin-x64/-/bun-darwin-x64-1.3.10.tgz", + "integrity": "sha512-Nhssuh7GBpP5PiDSOl3+qnoIG7PJo+ec2oomDevnl9pRY6x6aD2gRt0JE+uf+A8Om2D6gjeHCxjEdrw5ZHE8mA==", + "cpu": [ + "x64" + ], + "dev": true, + "license": "MIT", + "optional": true, + "os": [ + "darwin" + ] + }, + "node_modules/@oven/bun-darwin-x64-baseline": { + "version": "1.3.10", + "resolved": "https://registry.npmjs.org/@oven/bun-darwin-x64-baseline/-/bun-darwin-x64-baseline-1.3.10.tgz", + "integrity": "sha512-w1gaTlqU0IJCmJ1X+PGHkdNU1n8Gemx5YKkjhkJIguvFINXEBB5U1KG82QsT65Tk4KyNMfbLTlmy4giAvUoKfA==", + "cpu": [ + "x64" + ], + "dev": true, + "license": "MIT", + "optional": true, + "os": [ + "darwin" + ] + }, + "node_modules/@oven/bun-linux-aarch64": { + "version": "1.3.10", + "resolved": "https://registry.npmjs.org/@oven/bun-linux-aarch64/-/bun-linux-aarch64-1.3.10.tgz", + "integrity": "sha512-OUgPHfL6+PM2Q+tFZjcaycN3D7gdQdYlWnwMI31DXZKY1r4HINWk9aEz9t/rNaHg65edwNrt7dsv9TF7xK8xIA==", + "cpu": [ + "arm64" + ], + "dev": true, + "license": "MIT", + "optional": true, + "os": [ + "linux" + ] + }, + "node_modules/@oven/bun-linux-aarch64-musl": { + "version": "1.3.10", + "resolved": "https://registry.npmjs.org/@oven/bun-linux-aarch64-musl/-/bun-linux-aarch64-musl-1.3.10.tgz", + "integrity": "sha512-Ui5pAgM7JE9MzHokF0VglRMkbak3lTisY4Mf1AZutPACXWgKJC5aGrgnHBfkl7QS6fEeYb0juy1q4eRznRHOsw==", + "cpu": [ + "arm64" + ], + "dev": true, + "license": "MIT", + "optional": true, + "os": [ + "linux" + ] + }, + "node_modules/@oven/bun-linux-x64": { + "version": "1.3.10", + "resolved": "https://registry.npmjs.org/@oven/bun-linux-x64/-/bun-linux-x64-1.3.10.tgz", + "integrity": "sha512-bzUgYj/PIZziB/ZesIP9HUyfvh6Vlf3od+TrbTTyVEuCSMKzDPQVW/yEbRp0tcHO3alwiEXwJDrWrHAguXlgiQ==", + "cpu": [ + "x64" + ], + "dev": true, + "license": "MIT", + "optional": true, + "os": [ + "linux" + ] + }, + "node_modules/@oven/bun-linux-x64-baseline": { + "version": "1.3.10", + "resolved": "https://registry.npmjs.org/@oven/bun-linux-x64-baseline/-/bun-linux-x64-baseline-1.3.10.tgz", + "integrity": "sha512-oqvMDYpX6dGJO03HgO5bXuccEsH3qbdO3MaAiAlO4CfkBPLUXz3N0DDElg5hz0L6ktdDVKbQVE5lfe+LAUISQg==", + "cpu": [ + "x64" + ], + "dev": true, + "license": "MIT", + "optional": true, + "os": [ + "linux" + ] + }, + "node_modules/@oven/bun-linux-x64-musl": { + "version": "1.3.10", + "resolved": "https://registry.npmjs.org/@oven/bun-linux-x64-musl/-/bun-linux-x64-musl-1.3.10.tgz", + "integrity": "sha512-poVXvOShekbexHq45b4MH/mRjQKwACAC8lHp3Tz/hEDuz0/20oncqScnmKwzhBPEpqJvydXficXfBYuSim8opw==", + "cpu": [ + "x64" + ], + "dev": true, + "license": "MIT", + "optional": true, + "os": [ + "linux" + ] + }, + "node_modules/@oven/bun-linux-x64-musl-baseline": { + "version": "1.3.10", + "resolved": "https://registry.npmjs.org/@oven/bun-linux-x64-musl-baseline/-/bun-linux-x64-musl-baseline-1.3.10.tgz", + "integrity": "sha512-/hOZ6S1VsTX6vtbhWVL9aAnOrdpuO54mAGUWpTdMz7dFG5UBZ/VUEiK0pBkq9A1rlBk0GeD/6Y4NBFl8Ha7cRA==", + "cpu": [ + "x64" + ], + "dev": true, + "license": "MIT", + "optional": true, + "os": [ + "linux" + ] + }, + "node_modules/@oven/bun-windows-aarch64": { + "version": "1.3.10", + "resolved": "https://registry.npmjs.org/@oven/bun-windows-aarch64/-/bun-windows-aarch64-1.3.10.tgz", + "integrity": "sha512-GXbz2swvN2DLw2dXZFeedMxSJtI64xQ9xp9Eg7Hjejg6mS2E4dP1xoQ2yAo2aZPi/2OBPAVaGzppI2q20XumHA==", + "cpu": [ + "arm64" + ], + "dev": true, + "license": "MIT", + "optional": true, + "os": [ + "win32" + ] + }, + "node_modules/@oven/bun-windows-x64": { + "version": "1.3.10", + "resolved": "https://registry.npmjs.org/@oven/bun-windows-x64/-/bun-windows-x64-1.3.10.tgz", + "integrity": "sha512-qaS1In3yfC/Z/IGQriVmF8GWwKuNqiw7feTSJWaQhH5IbL6ENR+4wGNPniZSJFaM/SKUO0e/YCRdoVBvgU4C1g==", + "cpu": [ + "x64" + ], + "dev": true, + "license": "MIT", + "optional": true, + "os": [ + "win32" + ] + }, + "node_modules/@oven/bun-windows-x64-baseline": { + "version": "1.3.10", + "resolved": "https://registry.npmjs.org/@oven/bun-windows-x64-baseline/-/bun-windows-x64-baseline-1.3.10.tgz", + "integrity": "sha512-gh3UAHbUdDUG6fhLc1Csa4IGdtghue6U8oAIXWnUqawp6lwb3gOCRvp25IUnLF5vUHtgfMxuEUYV7YA2WxVutw==", + "cpu": [ + "x64" + ], + "dev": true, + "license": "MIT", + "optional": true, + "os": [ + "win32" + ] + }, "node_modules/@oxc-parser/binding-android-arm-eabi": { "version": "0.137.0", "resolved": "https://registry.npmjs.org/@oxc-parser/binding-android-arm-eabi/-/binding-android-arm-eabi-0.137.0.tgz", @@ -5071,9 +5240,6 @@ "arm64" ], "dev": true, - "libc": [ - "glibc" - ], "license": "MIT", "optional": true, "os": [ @@ -5091,9 +5257,6 @@ "arm64" ], "dev": true, - "libc": [ - "musl" - ], "license": "MIT", "optional": true, "os": [ @@ -5111,9 +5274,6 @@ "ppc64" ], "dev": true, - "libc": [ - "glibc" - ], "license": "MIT", "optional": true, "os": [ @@ -5131,9 +5291,6 @@ "riscv64" ], "dev": true, - "libc": [ - "glibc" - ], "license": "MIT", "optional": true, "os": [ @@ -5151,9 +5308,6 @@ "riscv64" ], "dev": true, - "libc": [ - "musl" - ], "license": "MIT", "optional": true, "os": [ @@ -5171,9 +5325,6 @@ "s390x" ], "dev": true, - "libc": [ - "glibc" - ], "license": "MIT", "optional": true, "os": [ @@ -5191,9 +5342,6 @@ "x64" ], "dev": true, - "libc": [ - "glibc" - ], "license": "MIT", "optional": true, "os": [ @@ -5211,9 +5359,6 @@ "x64" ], "dev": true, - "libc": [ - "musl" - ], "license": "MIT", "optional": true, "os": [ @@ -5479,9 +5624,6 @@ "arm64" ], "dev": true, - "libc": [ - "glibc" - ], "license": "MIT", "optional": true, "os": [ @@ -5496,9 +5638,6 @@ "arm64" ], "dev": true, - "libc": [ - "musl" - ], "license": "MIT", "optional": true, "os": [ @@ -5513,9 +5652,6 @@ "ppc64" ], "dev": true, - "libc": [ - "glibc" - ], "license": "MIT", "optional": true, "os": [ @@ -5530,9 +5666,6 @@ "riscv64" ], "dev": true, - "libc": [ - "glibc" - ], "license": "MIT", "optional": true, "os": [ @@ -5547,9 +5680,6 @@ "riscv64" ], "dev": true, - "libc": [ - "musl" - ], "license": "MIT", "optional": true, "os": [ @@ -5564,9 +5694,6 @@ "s390x" ], "dev": true, - "libc": [ - "glibc" - ], "license": "MIT", "optional": true, "os": [ @@ -5581,9 +5708,6 @@ "x64" ], "dev": true, - "libc": [ - "glibc" - ], "license": "MIT", "optional": true, "os": [ @@ -5598,9 +5722,6 @@ "x64" ], "dev": true, - "libc": [ - "musl" - ], "license": "MIT", "optional": true, "os": [ @@ -7187,9 +7308,6 @@ "arm64" ], "dev": true, - "libc": [ - "glibc" - ], "license": "MIT", "optional": true, "os": [ @@ -7207,9 +7325,6 @@ "arm64" ], "dev": true, - "libc": [ - "musl" - ], "license": "MIT", "optional": true, "os": [ @@ -7227,9 +7342,6 @@ "ppc64" ], "dev": true, - "libc": [ - "glibc" - ], "license": "MIT", "optional": true, "os": [ @@ -7247,9 +7359,6 @@ "s390x" ], "dev": true, - "libc": [ - "glibc" - ], "license": "MIT", "optional": true, "os": [ @@ -7267,9 +7376,6 @@ "x64" ], "dev": true, - "libc": [ - "glibc" - ], "license": "MIT", "optional": true, "os": [ @@ -7287,9 +7393,6 @@ "x64" ], "dev": true, - "libc": [ - "musl" - ], "license": "MIT", "optional": true, "os": [ @@ -8441,9 +8544,6 @@ "arm64" ], "dev": true, - "libc": [ - "glibc" - ], "license": "MIT", "optional": true, "os": [ @@ -8461,9 +8561,6 @@ "arm64" ], "dev": true, - "libc": [ - "musl" - ], "license": "MIT", "optional": true, "os": [ @@ -8481,9 +8578,6 @@ "x64" ], "dev": true, - "libc": [ - "glibc" - ], "license": "MIT", "optional": true, "os": [ @@ -8501,9 +8595,6 @@ "x64" ], "dev": true, - "libc": [ - "musl" - ], "license": "MIT", "optional": true, "os": [ @@ -11372,6 +11463,41 @@ "url": "https://github.com/sponsors/sindresorhus" } }, + "node_modules/bun": { + "version": "1.3.10", + "resolved": "https://registry.npmjs.org/bun/-/bun-1.3.10.tgz", + "integrity": "sha512-S/CXaXXIyA4CMjdMkYQ4T2YMqnAn4s0ysD3mlsY4bUiOCqGlv28zck4Wd4H4kpvbekx15S9mUeLQ7Uxd0tYTLA==", + "cpu": [ + "arm64", + "x64" + ], + "dev": true, + "hasInstallScript": true, + "license": "MIT", + "os": [ + "darwin", + "linux", + "win32" + ], + "bin": { + "bun": "bin/bun.exe", + "bunx": "bin/bunx.exe" + }, + "optionalDependencies": { + "@oven/bun-darwin-aarch64": "1.3.10", + "@oven/bun-darwin-x64": "1.3.10", + "@oven/bun-darwin-x64-baseline": "1.3.10", + "@oven/bun-linux-aarch64": "1.3.10", + "@oven/bun-linux-aarch64-musl": "1.3.10", + "@oven/bun-linux-x64": "1.3.10", + "@oven/bun-linux-x64-baseline": "1.3.10", + "@oven/bun-linux-x64-musl": "1.3.10", + "@oven/bun-linux-x64-musl-baseline": "1.3.10", + "@oven/bun-windows-aarch64": "1.3.10", + "@oven/bun-windows-x64": "1.3.10", + "@oven/bun-windows-x64-baseline": "1.3.10" + } + }, "node_modules/bun-types": { "version": "1.3.14", "resolved": "https://registry.npmjs.org/bun-types/-/bun-types-1.3.14.tgz", @@ -28545,7 +28671,7 @@ }, "open-sse": { "name": "@omniroute/open-sse", - "version": "3.8.42", + "version": "3.8.43", "dependencies": { "@toon-format/toon": "^2.3.0", "safe-regex": "^2.1.1" diff --git a/package.json b/package.json index ca8a21c4ef..ce571e997c 100644 --- a/package.json +++ b/package.json @@ -1,6 +1,6 @@ { "name": "omniroute", - "version": "3.8.42", + "version": "3.8.43", "description": "Unified AI router with 160+ providers, RTK+Caveman compression, auto fallback, MCP/A2A, desktop, PWA, and OpenAI-compatible APIs.", "type": "module", "bin": { @@ -76,8 +76,8 @@ "scripts": { "dev": "node --max-old-space-size=8192 scripts/dev/run-next.mjs dev", "prebuild:docs": "node scripts/docs/gen-openapi-module.mjs", - "gen:provider-reference": "node --import tsx scripts/docs/gen-provider-reference.ts", - "bench:compression": "node --import tsx scripts/compression/benchmark.ts", + "gen:provider-reference": "bun scripts/docs/gen-provider-reference.ts", + "bench:compression": "bun scripts/compression/benchmark.ts", "eval:compression": "node --import tsx scripts/compression-eval/index.ts", "release:sync-changelog-i18n": "node scripts/release/sync-changelog-i18n.mjs", "build": "node scripts/build/build-next-isolated.mjs", @@ -95,13 +95,13 @@ "electron:build:mac": "npm run build && cd electron && npm run build:mac", "electron:build:linux": "npm run build && cd electron && npm run build:linux", "electron:smoke:packaged": "node scripts/dev/smoke-electron-packaged.mjs", - "test": "cross-env DISABLE_SQLITE_AUTO_BACKUP=true node --max-old-space-size=8192 --import tsx --import ./open-sse/utils/setupPolyfill.ts --import ./tests/_setup/isolateDataDir.ts --test --test-concurrency=20 tests/unit/*.test.ts \"tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts\"", - "test:unit": "cross-env DISABLE_SQLITE_AUTO_BACKUP=true node --max-old-space-size=8192 --import tsx --import ./open-sse/utils/setupPolyfill.ts --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit --test-concurrency=20 tests/unit/*.test.ts \"tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts\"", - "test:unit:ci": "cross-env DISABLE_SQLITE_AUTO_BACKUP=true node --max-old-space-size=8192 --import tsx --import ./open-sse/utils/setupPolyfill.ts --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit --test-concurrency=4 tests/unit/*.test.ts \"tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts\"", - "test:unit:fast": "cross-env DISABLE_SQLITE_AUTO_BACKUP=true node --max-old-space-size=8192 --import tsx --import ./open-sse/utils/setupPolyfill.ts --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit --test-isolation=none tests/unit/*.test.ts \"tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts\"", + "test": "cross-env DISABLE_SQLITE_AUTO_BACKUP=true node --max-old-space-size=8192 --import tsx --import ./open-sse/utils/setupPolyfill.ts --import ./tests/_setup/isolateDataDir.ts --test --test-concurrency=20 tests/unit/*.test.ts \"tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,memory,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts\"", + "test:unit": "cross-env DISABLE_SQLITE_AUTO_BACKUP=true node --max-old-space-size=8192 --import tsx --import ./open-sse/utils/setupPolyfill.ts --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit --test-concurrency=20 tests/unit/*.test.ts \"tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,memory,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts\"", + "test:unit:ci": "cross-env DISABLE_SQLITE_AUTO_BACKUP=true node --max-old-space-size=8192 --import tsx --import ./open-sse/utils/setupPolyfill.ts --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit --test-concurrency=4 tests/unit/*.test.ts \"tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,memory,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts\"", + "test:unit:fast": "cross-env DISABLE_SQLITE_AUTO_BACKUP=true node --max-old-space-size=8192 --import tsx --import ./open-sse/utils/setupPolyfill.ts --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit --test-isolation=none tests/unit/*.test.ts \"tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,memory,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts\"", "test:unit:shard": "concurrently --kill-others-on-fail -n s1,s2 \"npm:test:unit:shard:1\" \"npm:test:unit:shard:2\"", - "test:unit:shard:1": "cross-env DISABLE_SQLITE_AUTO_BACKUP=true node --max-old-space-size=8192 --import tsx --import ./open-sse/utils/setupPolyfill.ts --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit --test-concurrency=10 --test-shard=1/2 tests/unit/*.test.ts \"tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts\"", - "test:unit:shard:2": "cross-env DISABLE_SQLITE_AUTO_BACKUP=true node --max-old-space-size=8192 --import tsx --import ./open-sse/utils/setupPolyfill.ts --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit --test-concurrency=10 --test-shard=2/2 tests/unit/*.test.ts \"tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts\"", + "test:unit:shard:1": "cross-env DISABLE_SQLITE_AUTO_BACKUP=true node --max-old-space-size=8192 --import tsx --import ./open-sse/utils/setupPolyfill.ts --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit --test-concurrency=10 --test-shard=1/2 tests/unit/*.test.ts \"tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,memory,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts\"", + "test:unit:shard:2": "cross-env DISABLE_SQLITE_AUTO_BACKUP=true node --max-old-space-size=8192 --import tsx --import ./open-sse/utils/setupPolyfill.ts --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit --test-concurrency=10 --test-shard=2/2 tests/unit/*.test.ts \"tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,memory,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts\"", "test:plan3": "cross-env DISABLE_SQLITE_AUTO_BACKUP=true node --import tsx --import ./open-sse/utils/setupPolyfill.ts --import ./tests/_setup/isolateDataDir.ts --test tests/unit/plan3-p0.test.ts", "test:fixes": "cross-env DISABLE_SQLITE_AUTO_BACKUP=true node --import tsx --import ./open-sse/utils/setupPolyfill.ts --import ./tests/_setup/isolateDataDir.ts --test tests/unit/fixes-p1.test.ts", "test:security": "cross-env DISABLE_SQLITE_AUTO_BACKUP=true node --import tsx --import ./open-sse/utils/setupPolyfill.ts --import ./tests/_setup/isolateDataDir.ts --test tests/unit/security-fase01.test.ts", @@ -113,7 +113,7 @@ "check:env-doc-sync": "node scripts/check/check-env-doc-sync.mjs", "check:docs-counts": "node scripts/check/check-docs-counts-sync.mjs", "check:deprecated-versions": "node scripts/check/check-deprecated-versions.mjs", - "check:compression-budget": "node --import tsx scripts/check/check-compression-budget.ts", + "check:compression-budget": "bun scripts/check/check-compression-budget.ts", "check:doc-links": "node scripts/check/check-doc-links.mjs", "check:fabricated-docs": "node scripts/check/check-fabricated-docs.mjs --strict", "check:docs-all": "npm run check:docs-sync && npm run check:docs-counts && npm run check:env-doc-sync && npm run check:deprecated-versions && npm run check:doc-links && npm run check:fabricated-docs", @@ -130,7 +130,7 @@ "check:cli-i18n": "node scripts/check/check-cli-i18n.mjs", "check:openapi-coverage": "node scripts/check/check-openapi-coverage.mjs", "check:openapi-security-tiers": "node scripts/check/check-openapi-security-tiers.mjs", - "check:provider-consistency": "node --import tsx scripts/check/check-provider-consistency.ts", + "check:provider-consistency": "bun scripts/check/check-provider-consistency.ts", "check:provider-assets": "node scripts/check/check-provider-assets.mjs", "check:fetch-targets": "node scripts/check/check-fetch-targets.mjs", "check:openapi-routes": "node scripts/check/check-openapi-routes.mjs", @@ -147,9 +147,10 @@ "check:public-creds": "node scripts/check/check-public-creds.mjs", "check:db-rules": "node scripts/check/check-db-rules.mjs", "check:docs-symbols": "node scripts/check/check-docs-symbols.mjs", - "check:known-symbols": "node --import tsx scripts/check/check-known-symbols.ts", + "check:known-symbols": "bun scripts/check/check-known-symbols.ts", "check:route-guard-membership": "node --import tsx scripts/check/check-route-guard-membership.ts", "check:test-discovery": "node scripts/check/check-test-discovery.mjs", + "check:mutation-test-coverage": "node scripts/check/check-mutation-test-coverage.mjs --strict", "check:complexity": "node scripts/check/check-complexity.mjs", "check:dead-code": "node scripts/check/check-dead-code.mjs", "check:cognitive-complexity": "node scripts/check/check-cognitive-complexity.mjs", @@ -190,7 +191,7 @@ "test:mutation": "stryker run", "test:ecosystem": "node scripts/dev/run-ecosystem-tests.mjs", "test:system": "cross-env DISABLE_SQLITE_AUTO_BACKUP=true node --import tsx --import ./open-sse/utils/setupPolyfill.ts --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit --test-concurrency=1 tests/e2e/system-failover.test.ts", - "test:coverage": "cross-env DISABLE_SQLITE_AUTO_BACKUP=true NODE_OPTIONS=--max-old-space-size=8192 c8 --merge-async --output-dir coverage --exclude=tests/** --exclude=**/*.test.* --reporter=text-summary --reporter=html --reporter=json-summary --reporter=lcov --check-coverage --statements 60 --lines 60 --functions 60 --branches 60 node --max-old-space-size=8192 --import tsx --import ./open-sse/utils/setupPolyfill.ts --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit --test-concurrency=8 tests/unit/*.test.ts \"tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts\"", + "test:coverage": "cross-env DISABLE_SQLITE_AUTO_BACKUP=true NODE_OPTIONS=--max-old-space-size=8192 c8 --merge-async --output-dir coverage --exclude=tests/** --exclude=**/*.test.* --reporter=text-summary --reporter=html --reporter=json-summary --reporter=lcov --check-coverage --statements 60 --lines 60 --functions 60 --branches 60 node --max-old-space-size=8192 --import tsx --import ./open-sse/utils/setupPolyfill.ts --import ./tests/_setup/isolateDataDir.ts --test --test-force-exit --test-concurrency=8 tests/unit/*.test.ts \"tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,memory,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts\"", "test:coverage:legacy": "c8 --output-dir coverage --exclude=open-sse --check-coverage --lines 50 --functions 50 --branches 50 node --import tsx --test tests/unit/*.test.ts", "coverage:report": "cross-env NODE_OPTIONS=--max-old-space-size=8192 c8 report --merge-async --output-dir coverage --exclude=tests/** --exclude=**/*.test.* --reporter=text --reporter=text-summary --reporter=html --reporter=json-summary --reporter=lcov", "coverage:summary": "node scripts/check/test-report-summary.mjs --input coverage/coverage-summary.json --output coverage/coverage-report.md", @@ -216,6 +217,7 @@ "@monaco-editor/react": "^4.7.0", "@ngrok/ngrok": "^1.7.0", "@swc/helpers": "0.5.23", + "@toon-format/toon": "^2.3.0", "@types/mdx": "^2.0.13", "@xyflow/react": "^12.11.1", "axios": "^1.16.1", @@ -264,6 +266,7 @@ "react-markdown": "^10.1.0", "react-reconciler": "^0.33.0", "recharts": "^3.8.1", + "safe-regex": "^2.1.1", "selfsigned": "^5.5.0", "socks": "^2.8.7", "sql.js": "^1.14.1", @@ -307,6 +310,7 @@ "@types/safe-regex": "^1.1.6", "@types/ws": "^8.18.0", "@vitejs/plugin-react": "^6.0.2", + "bun": "1.3.10", "c8": "^11.0.0", "concurrently": "^10.0.3", "cross-env": "^10.1.0", diff --git a/scripts/build/validate-pack-artifact.ts b/scripts/build/validate-pack-artifact.ts index 94b09fe166..202cb87b51 100644 --- a/scripts/build/validate-pack-artifact.ts +++ b/scripts/build/validate-pack-artifact.ts @@ -20,8 +20,11 @@ const npmCommand: string = process.platform === "win32" ? "npm.cmd" : "npm"; function runNpm(args: string[], stdio: "inherit" | "pipe" = "pipe"): string { const npmExecPath = process.env.npm_execpath; - const command = npmExecPath ? process.execPath : npmCommand; - return execFileSync(command, [...(npmExecPath ? [npmExecPath] : []), ...args], { + const isBunRuntime = "Bun" in globalThis; + const command = npmExecPath && !isBunRuntime ? process.execPath : npmCommand; + const commandArgs = npmExecPath && !isBunRuntime ? [npmExecPath, ...args] : args; + + return execFileSync(command, commandArgs, { cwd: ROOT, encoding: "utf8", stdio: stdio === "inherit" ? "inherit" : ["ignore", "pipe", "pipe"], diff --git a/scripts/check/check-fabricated-docs.mjs b/scripts/check/check-fabricated-docs.mjs index 785ca9395b..13dcc42ab1 100644 --- a/scripts/check/check-fabricated-docs.mjs +++ b/scripts/check/check-fabricated-docs.mjs @@ -305,6 +305,7 @@ const ENV_VAR_DENYLIST = new Set([ "KNOWN_STALE_DOC_REFS", // export const in check-docs-symbols.mjs "KNOWN_MISSING", // export const in check-fetch-targets.mjs "KNOWN_RAW_SQL", // export const in check-db-rules.mjs + "ROUTER_BACKENDS", // typed router-backend registry constant documented in the ADR (ROUTER_BACKENDS.md); code lands with PR #5868 (#5798) // ── Error / Node codes documented in prose (string-literal codes, not env vars) ── "URL_GUARD_BLOCKED", // HTTP 422 guard-violation code (ARCHITECTURE.md) "AUTHZ_NOT_INITIALIZED", // AuthzAssertionError code (AUTHZ_GUIDE.md) diff --git a/scripts/check/check-mutation-ratchet.mjs b/scripts/check/check-mutation-ratchet.mjs index 990436b7ed..8cd6747d98 100644 --- a/scripts/check/check-mutation-ratchet.mjs +++ b/scripts/check/check-mutation-ratchet.mjs @@ -31,19 +31,30 @@ const DEFAULT_REPORT = path.join(ROOT, "reports/mutation/mutation.json"); const BASELINE_PREFIX = "mutationScore."; const RATCHET = process.argv.includes("--ratchet"); +// Anti-flake tolerance (percentage points). The tap-runner runs at concurrency 4 and a +// mutant whose tests sit near the `timeoutMS` boundary can flip Killed/Timeout/Survived +// between runs, jittering a module's covered score by a fraction of a point. Only a drop +// LARGER than this counts as a regression — same idea as the coverage ratchet's `eps`. It +// does NOT lower the baseline; it absorbs run-to-run noise so the blocking gate stays +// deterministic. Real test-quality regressions (the headers.ts/telemetry-scale drops that +// motivated the tap.testFiles coverage fix) are far larger than EPS and still fire. +export const MUTATION_RATCHET_EPS = 1.0; + const DETECTED = new Set(["Killed", "Timeout"]); const SURVIVED = new Set(["Survived"]); /** * Avalia o score MEDIDO de um módulo contra o baseline. Direction: UP (o score só - * pode subir — menor = regressão). + * pode subir — menor = regressão), com tolerância anti-flake `eps`: só conta como + * regressão uma queda MAIOR que `eps` pontos. * @param {number} current * @param {number} baseline + * @param {number} [eps] tolerância anti-flake em pontos percentuais * @returns {{ regressed: boolean, improved: boolean }} */ -export function evaluateMutationRatchet(current, baseline) { +export function evaluateMutationRatchet(current, baseline, eps = MUTATION_RATCHET_EPS) { return { - regressed: current < baseline, + regressed: current < baseline - eps, improved: current > baseline, }; } diff --git a/scripts/check/check-mutation-test-coverage.mjs b/scripts/check/check-mutation-test-coverage.mjs new file mode 100644 index 0000000000..e256521c81 --- /dev/null +++ b/scripts/check/check-mutation-test-coverage.mjs @@ -0,0 +1,121 @@ +#!/usr/bin/env node +// check-mutation-test-coverage — guards against tap.testFiles drift. +// +// WHY: Stryker (nightly-mutation) only runs the test files listed in +// stryker.conf.json `tap.testFiles` against each mutant. When a NEW unit test +// that covers a mutated module is added (or an existing one is split/renamed) +// but NOT added to tap.testFiles, that test's kills stop counting. The mutants +// it would kill then go COVERED-but-unkilled = SURVIVED on a cold run, so the +// module's COVERED mutation score collapses and the blocking mutationScore +// ratchet (nightly-mutation.yml) false-fails — but only on cold-cache nights, +// because the warm incremental run reuses the older (passing) verdicts. The +// pass/fail then tracks GitHub cache state, not code quality. Root cause: +// tap.testFiles is a hand-maintained list with no drift guard. This is it. +// +// INVARIANT: every UNIT test (tests/unit/**) that imports a mutated module +// (stryker.conf.json `mutate`) MUST be in `tap.testFiles`. Integration/e2e +// tests are intentionally excluded (the tap-runner runs node:test units only). +// +// MODE: advisory by default (prints drift, exit 0). `--strict` exits 1 on drift +// so CI can block. Skip-graceful (exit 0) if stryker.conf.json is absent. +// +// USAGE: +// node scripts/check/check-mutation-test-coverage.mjs # advisory +// node scripts/check/check-mutation-test-coverage.mjs --strict # blocking + +import fs from "node:fs"; +import { execFileSync } from "node:child_process"; + +/** 3-segment path suffix without extension (unique enough; matches relative + alias imports). */ +export function moduleFragment(modulePath) { + return modulePath.replace(/\.ts$/, "").split("/").slice(-3).join("/"); +} + +/** + * True if `content` imports the module identified by `fragment`, in any form: + * static `... from "…fragment…"`, dynamic `import("…fragment…")` (incl. split + * across lines), or `require("…fragment…")`. The fragment must appear INSIDE the + * import string literal — a bare mention in a comment does not match. + */ +export function testImportsModule(content, fragment) { + const esc = fragment.replace(/[.*+?^${}()|[\]\\]/g, "\\$&"); + const re = new RegExp( + "(?:from\\s+|\\bimport\\s*\\(\\s*|\\brequire\\s*\\(\\s*)[\"'][^\"']*" + esc + ); + return re.test(content); +} + +/** + * @param {{mutate: string[], tapTestFiles: string[], unitTests: {path:string, content:string}[]}} input + * @returns {Record} module -> covering unit tests NOT in tap.testFiles (drift) + */ +export function findCoverageDrift({ mutate, tapTestFiles, unitTests }) { + const tap = new Set(tapTestFiles); + const drift = {}; + for (const mod of mutate) { + if (mod.startsWith("_") || !mod.endsWith(".ts")) continue; // skip comment/non-ts entries + const frag = moduleFragment(mod); + const missing = unitTests + .filter((t) => testImportsModule(t.content, frag) && !tap.has(t.path)) + .map((t) => t.path) + .sort(); + if (missing.length > 0) drift[mod] = missing; + } + return drift; +} + +function listUnitTests() { + // Static argv — no shell, no interpolation. + const out = execFileSync("git", ["ls-files", "tests/unit"], { encoding: "utf8" }); + return out + .split("\n") + .filter((f) => /\.test\.ts$/.test(f)) + // Exclude tests/unit/build/: these test the build TOOLING (scripts/), not the + // mutated runtime modules. They legitimately embed module paths as fixture + // strings (e.g. this gate's own test), which would otherwise false-match. + .filter((f) => !f.startsWith("tests/unit/build/")) + .map((path) => ({ path, content: fs.readFileSync(path, "utf8") })); +} + +function main() { + const STRICT = process.argv.includes("--strict"); + let conf; + try { + conf = JSON.parse(fs.readFileSync("stryker.conf.json", "utf8")); + } catch { + console.warn("[mutation-test-coverage] stryker.conf.json not found — skipping (advisory)."); + process.exit(0); + } + const mutate = (conf.mutate || []).filter((m) => typeof m === "string"); + const tapTestFiles = conf.tap?.testFiles || []; + const unitTests = listUnitTests(); + + const drift = findCoverageDrift({ mutate, tapTestFiles, unitTests }); + const modules = Object.keys(drift); + const total = modules.reduce((n, m) => n + drift[m].length, 0); + + console.log("Mutation test-coverage gate — tap.testFiles drift detection"); + console.log("==========================================================="); + console.log( + `Scanned ${unitTests.length} unit test file(s) against ${mutate.filter((m) => m.endsWith(".ts")).length} mutated module(s).` + ); + + if (total === 0) { + console.log("✓ No drift — every covering unit test is listed in tap.testFiles."); + process.exit(0); + } + + for (const m of modules) { + console.log(`\n ${m}`); + for (const t of drift[m]) console.log(` + ${t}`); + } + const msg = `${total} covering unit test(s) across ${modules.length} module(s) are missing from stryker.conf.json tap.testFiles.`; + if (STRICT) { + console.error(`\n✗ ${msg} Add them so their mutant kills count (--strict).`); + process.exit(1); + } + console.warn(`\n⚠ ${msg} Re-run with --strict to fail. Add them to tap.testFiles.`); + process.exit(0); +} + +if (import.meta.url === `file://${process.argv[1]}`) main(); diff --git a/scripts/check/check-route-guard-membership.ts b/scripts/check/check-route-guard-membership.ts index f547a2b4af..9f21c13d13 100644 --- a/scripts/check/check-route-guard-membership.ts +++ b/scripts/check/check-route-guard-membership.ts @@ -68,6 +68,7 @@ export const KNOWN_UNCLASSIFIED: Record = {}; */ export function routeFileToApiPath(routeFile: string): string { return routeFile + .replace(/\\/g, "/") .replace(/^src\/app/, "") .replace(/\/route\.ts$/, "") .replace(/\[([^\]]+)\]/g, "_$1_"); diff --git a/scripts/check/check-test-discovery.mjs b/scripts/check/check-test-discovery.mjs index 5a1bfe6b3f..9ea80e0774 100644 --- a/scripts/check/check-test-discovery.mjs +++ b/scripts/check/check-test-discovery.mjs @@ -53,7 +53,7 @@ export const COLLECTORS = [ // "vitest" e explodem no node runner). Subdir novo: adicione aqui E nos scripts // (o drift-check + o gate de órfãos forçam a manutenção em sincronia). { - glob: "tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts", + glob: "tests/unit/{api,auth,authz,build,cli,cli-helper,combo,compression,correctness,cors,dashboard,db,db-adapters,docs,gamification,guardrails,lib,mcp,memory,runtime,security,services,settings,shared,ui,usage}/**/*.test.ts", sources: ["package.json", ".github/workflows/ci.yml"], }, // Node native runner — test:integration (top-level only; tests/integration/services/ NÃO roda) diff --git a/scripts/dev/responses-ws-proxy.mjs b/scripts/dev/responses-ws-proxy.mjs index 5fa562d752..411cb78828 100644 --- a/scripts/dev/responses-ws-proxy.mjs +++ b/scripts/dev/responses-ws-proxy.mjs @@ -615,11 +615,17 @@ class ResponsesWsSession { toStringOrNull(responseBody.service_tier) || toStringOrNull(responseBody.serviceTier), }; - const upstream = await this.wsFactory(prepared.json.upstreamUrl, { - browser: prepared.json.browser || "chrome_149", + const wsOptions = { + // #5591: chrome_149 is not a wreq-js 2.3.1 profile (max chrome_147); the + // prepare route now sends chrome_142, this fallback matches it. + browser: prepared.json.browser || "chrome_142", os: prepared.json.os || "windows", headers: prepared.json.headers || {}, - }); + }; + // #5611: forward the configured proxy so the upstream WS connect honors the + // Proxy Registry in no-direct-egress deployments. + if (prepared.json.proxy) wsOptions.proxy = prepared.json.proxy; + const upstream = await this.wsFactory(prepared.json.upstreamUrl, wsOptions); upstream.onmessage = (event) => { if (this.closed) return; diff --git a/scripts/quality/validate-release-green.mjs b/scripts/quality/validate-release-green.mjs index a13253e32b..2914db0b50 100644 --- a/scripts/quality/validate-release-green.mjs +++ b/scripts/quality/validate-release-green.mjs @@ -119,7 +119,27 @@ export function computeVerdict(results) { // ─── Orchestration (only when run directly) ───────────────────────────────── -function run(cmd, cmdArgs) { +/** + * Map a thrown `execFileSync` error to a {code, out} gate result. Exported as a pure helper + * so the timeout/hang path has a regression test: a gate that exceeds its ceiling (e.g. the unit + * suite wedged on an unreleased SQLite handle — see CLAUDE.md "Database Handles in Tests") is + * killed by `execFileSync` (`err.killed === true`) and MUST surface as a visible non-zero gate, + * never an infinite block that the release captain mistakes for a hang and kills the pre-flight. + */ +export function classifyRunError(err, timeoutMs) { + if (err && err.killed && timeoutMs) { + return { + code: 124, + out: `gate exceeded its ${Math.round(timeoutMs / 1000)}s ceiling and was killed — treat as a hung/failed gate (e.g. an unreleased DB handle in the unit suite); does NOT pass`, + }; + } + return { + code: typeof err?.status === "number" ? err.status : 1, + out: `${err?.stdout || ""}${err?.stderr || ""}`, + }; +} + +function run(cmd, cmdArgs, opts = {}) { try { const out = execFileSync(cmd, cmdArgs, { cwd: ROOT, @@ -127,13 +147,13 @@ function run(cmd, cmdArgs) { stdio: ["ignore", "pipe", "pipe"], maxBuffer: 256 * 1024 * 1024, env: { ...process.env, FORCE_COLOR: "0" }, + // A hard ceiling for the long, silent test suites (execFileSync buffers all output until + // exit, so they show no progress while running). undefined = no timeout for fast gates. + ...(opts.timeout ? { timeout: opts.timeout } : {}), }); return { code: 0, out }; } catch (err) { - return { - code: typeof err.status === "number" ? err.status : 1, - out: `${err.stdout || ""}${err.stderr || ""}`, - }; + return classifyRunError(err, opts.timeout); } } @@ -150,8 +170,14 @@ function main() { process.stderr.write(`${icon} [${r.kind}] ${r.label}${r.detail ? ` — ${r.detail}` : ""}\n`); }; - const hardCmd = (id, label, cmd, cmdArgs) => { - const { code, out } = run(cmd, cmdArgs); + // Announce a gate BEFORE running it. The long suites (unit/vitest/integration) run silently + // for many minutes (execFileSync buffers their output until exit), which previously looked like + // a hang and got the pre-flight killed before it surfaced the unit reds (the v3.8.42 miss). + const announce = (label) => process.stderr.write(`▶ ${label}…\n`); + + const hardCmd = (id, label, cmd, cmdArgs, opts) => { + announce(label); + const { code, out } = run(cmd, cmdArgs, opts); record({ id, label, kind: "hard", ok: code === 0, detail: code === 0 ? "pass" : firstFailureLine(out) }); }; @@ -160,8 +186,9 @@ function main() { // non-zero exit here is drift to rebaseline at release, never a contributor block. // ALL checks run regardless of earlier failures (the report is collected, not // fail-fast) so one pass surfaces every red instead of revealing them in layers. - const driftCmd = (id, label, cmd, cmdArgs, okDetail = "within baseline") => { - const { code, out } = run(cmd, cmdArgs); + const driftCmd = (id, label, cmd, cmdArgs, okDetail = "within baseline", opts) => { + announce(label); + const { code, out } = run(cmd, cmdArgs, opts); record({ id, label, kind: "drift", ok: code === 0, detail: code === 0 ? okDetail : firstFailureLine(out) }); }; @@ -171,7 +198,8 @@ function main() { // ESLint: ONE pass → errors (hard) + warnings (drift) { - const { out } = run("npx", ["eslint", ".", "--format", "json"]); + announce("ESLint (errors + warnings — ~3-6min)"); + const { out } = run("npx", ["eslint", ".", "--format", "json"], { timeout: 15 * 60 * 1000 }); const parsed = parseEslintJson(out); if (!parsed) { record({ id: "lint", label: "ESLint", kind: "hard", ok: false, detail: "could not parse eslint json" }); @@ -198,6 +226,7 @@ function main() { // Cognitive-complexity (drift) { + announce("Cognitive complexity (ratchet)"); const { out } = run(npmCmd, ["run", "check:cognitive-complexity"]); const current = parseCognitiveCount(out); const base = baselineValue("cognitiveComplexity"); @@ -243,15 +272,20 @@ function main() { hardCmd("docs-all", "Docs sync + fabricated-docs (strict)", npmCmd, ["run", "check:docs-all"]); if (!QUICK) { - hardCmd("unit", "Unit tests (full, CI concurrency)", npmCmd, ["run", "test:unit:ci"]); - hardCmd("vitest", "Vitest (MCP / autoCombo / cache)", npmCmd, ["run", "test:vitest"]); + // These are the gates that catch inherited base-red tests from cycle PRs (the fast-path + // PR→release does NOT run unit/vitest/integration per-PR — the v3.8.42 release PR exploded + // with 15 such reds). They run SILENTLY for many minutes; the announce line above + these + // hard ceilings keep a long-but-healthy run from being mistaken for a hang (the ceiling also + // converts a genuine DB-handle hang into a visible failure instead of an infinite block). + hardCmd("unit", "Unit tests (full suite, CI concurrency — runs ~20-35min silently)", npmCmd, ["run", "test:unit:ci"], { timeout: 45 * 60 * 1000 }); + hardCmd("vitest", "Vitest (MCP / autoCombo / cache — ~3-8min)", npmCmd, ["run", "test:vitest"], { timeout: 15 * 60 * 1000 }); // Integration tests run ONLY on the release PR full CI (PR→main), so an assertion // regression here (e.g. a contributor flipping a Codex fingerprint key order) is // invisible until release — run them in the pre-flight as a HARD gate. - hardCmd("integration", "Integration tests", npmCmd, ["run", "test:integration"]); + hardCmd("integration", "Integration tests (~3-10min)", npmCmd, ["run", "test:integration"], { timeout: 20 * 60 * 1000 }); } if (WITH_BUILD) { - hardCmd("pack-artifact", "Package artifact (npm pack policy)", npmCmd, ["run", "check:pack-artifact"]); + hardCmd("pack-artifact", "Package artifact (npm pack policy)", npmCmd, ["run", "check:pack-artifact"], { timeout: 20 * 60 * 1000 }); } const { releaseGreen, hardFailures, drift } = computeVerdict(results); diff --git a/semcheck.yaml b/semcheck.yaml deleted file mode 100644 index f86a9bb94c..0000000000 --- a/semcheck.yaml +++ /dev/null @@ -1,160 +0,0 @@ -# semcheck.yaml — docs↔code semantic consistency (Task 14 — Fase 7) -# -# PURPOSE -# Fuzzy (LLM-assisted) layer that catches documentation describing behaviour -# that the code no longer implements. Complements the deterministic -# check-docs-symbols.mjs (Fase 6) which catches renamed/deleted exports. -# semcheck catches semantic drift: a doc that says "returns 200 on success" -# when the code now returns 204, or that describes a feature that was removed. -# -# OPERATING MODE — ADVISORY / NON-BLOCKING -# - This gate is LLM-backed → runs in the NIGHTLY job only (not per-PR). -# - It NEVER blocks a PR merge automatically. -# - Findings appear as annotations in the CI summary and as a comment on the -# PR if triggered by a label (see ci.yml `semcheck` job). -# - To trigger on a specific PR: add the label `semcheck` to the PR. -# - To promote to blocking: set `fail-on-issues: true` and wire the job -# under `required-checks` in the branch protection rule. -# -# FAIL MODE -# fail-on-issues: false ← advisory (current) -# fail-on-issues: true ← blocking (future, opt-in) -# -# TOOL -# semcheck OSS (https://github.com/semcheck/semcheck — MIT). -# Install: npm install --save-dev semcheck (not bundled; nightly-only) -# Run: npx semcheck --config semcheck.yaml -# -# ADDING NEW RULES -# 1. Pick a doc file and the matching source file(s). -# 2. Write a `claim` that the doc makes about the code behaviour. -# 3. Optionally add `context-file` paths for the LLM to read. -# 4. Set `severity: warning` for informational, `error` for blocking candidates. -# Rules are evaluated independently; a failure in one does not block others. - -version: "1" - -fail-on-issues: false # ADVISORY — change to true to make blocking - -model: "gpt-4o-mini" # cheap model for advisory pass; upgrade to gpt-4o for precision - -rules: - - # ── Routing & Combo ────────────────────────────────────────────────────── - - - id: combo-strategies-count - doc: docs/routing/AUTO-COMBO.md - claim: > - The document states that OmniRoute supports 14 combo routing strategies. - Verify that open-sse/services/combo.ts exports or references exactly 14 - strategy identifiers (or that any discrepancy has a comment explaining why). - context-files: - - open-sse/services/combo.ts - severity: warning - - - id: auto-combo-9-factors - doc: docs/routing/AUTO-COMBO.md - claim: > - The document describes exactly 9 scoring factors for the Auto-Combo - strategy. Verify that the implementation references 9 distinct factors - (e.g. in the autoCombo scoring function or its constants/comments). - context-files: - - open-sse/services/autoCombo/ - severity: warning - - - id: resilience-3-layers - doc: docs/architecture/RESILIENCE_GUIDE.md - claim: > - The document describes 3 and only 3 resilience mechanisms: Provider - Circuit Breaker, Connection Cooldown, and Model Lockout. Verify that no - 4th mechanism is implemented without being documented. - context-files: - - src/shared/utils/circuitBreaker.ts - - open-sse/services/accountFallback.ts - severity: warning - - # ── Security ───────────────────────────────────────────────────────────── - - - id: error-sanitization-flow - doc: docs/security/ERROR_SANITIZATION.md - claim: > - The document states that all error responses MUST route through - buildErrorBody() or sanitizeErrorMessage() from open-sse/utils/error.ts. - Verify that the described API (function names, module path) matches what - is actually exported in that file. - context-files: - - open-sse/utils/error.ts - severity: error - - - id: public-creds-pattern - doc: docs/security/PUBLIC_CREDS.md - claim: > - The document describes a mandatory resolvePublicCred() function in - open-sse/utils/publicCreds.ts. Verify that function exists with roughly - the API signature the doc describes (accepts a key, returns a string). - context-files: - - open-sse/utils/publicCreds.ts - severity: error - - # ── MCP Server ─────────────────────────────────────────────────────────── - - - id: mcp-tool-count - doc: docs/frameworks/MCP-SERVER.md - claim: > - The document states the MCP server exposes 43 tools in total (30 base + - 3 memory + 4 skills + 6 notion). Verify this count is consistent with - the number of tool definitions found in open-sse/mcp-server/tools/. - context-files: - - open-sse/mcp-server/tools/ - severity: warning - - # ── A2A Server ─────────────────────────────────────────────────────────── - - - id: a2a-skills-count - doc: docs/frameworks/A2A-SERVER.md - claim: > - The document lists exactly 5 built-in A2A skills: smart-routing, - quota-management, provider-discovery, cost-analysis, health-report. - Verify that src/lib/a2a/skills/ contains entries matching these names - and that no additional undocumented skills exist. - context-files: - - src/lib/a2a/skills/ - - src/lib/a2a/taskExecution.ts - severity: warning - - # ── API Reference ───────────────────────────────────────────────────────── - - - id: api-route-v1-prefix - doc: docs/reference/API_REFERENCE.md - claim: > - The document states all API routes are under /v1/ (e.g. /v1/chat/completions). - Verify that src/app/api/v1/ exists and that the key routes mentioned in - the doc (chat/completions, embeddings, models) are present. - context-files: - - src/app/api/v1/ - severity: warning - - # ── Embedded Services ───────────────────────────────────────────────────── - - - id: embedded-services-api-pattern - doc: docs/frameworks/EMBEDDED-SERVICES.md - claim: > - The document says each embedded service exposes 7 API endpoints: - install, start, stop, restart, update, status, auto-start (plus a shared - logs endpoint). Verify that at least one service under src/app/api/services/ - follows this pattern. - context-files: - - src/app/api/services/ - severity: warning - - # ── Route Guard ─────────────────────────────────────────────────────────── - - - id: route-guard-local-only - doc: docs/security/ROUTE_GUARD_TIERS.md - claim: > - The document describes LOCAL_ONLY_API_PREFIXES enforced by - src/server/authz/routeGuard.ts via isLocalOnlyPath(). Verify these - identifiers exist in the source file with roughly the described semantics. - context-files: - - src/server/authz/routeGuard.ts - severity: error diff --git a/src/app/(dashboard)/dashboard/HomePageClient.tsx b/src/app/(dashboard)/dashboard/HomePageClient.tsx index d27cc013a6..017d14de00 100644 --- a/src/app/(dashboard)/dashboard/HomePageClient.tsx +++ b/src/app/(dashboard)/dashboard/HomePageClient.tsx @@ -1109,7 +1109,7 @@ export default function HomePageClient({ machineId }: HomePageClientProps) {

{t.rich("step1Desc", { endpoint: (chunks) => ( - + {chunks} ), diff --git a/src/app/(dashboard)/dashboard/api-manager/ApiManagerPageClient.tsx b/src/app/(dashboard)/dashboard/api-manager/ApiManagerPageClient.tsx index 15224fb377..362196321d 100644 --- a/src/app/(dashboard)/dashboard/api-manager/ApiManagerPageClient.tsx +++ b/src/app/(dashboard)/dashboard/api-manager/ApiManagerPageClient.tsx @@ -23,6 +23,7 @@ import { readActiveOnlyPreference, writeActiveOnlyPreference } from "./apiManage import { buildApiKeyCreateScopes, mergeApiKeyPermissionScopes } from "./apiManagerScopes"; import { SELF_ACCOUNT_QUOTA_SCOPE, SELF_USAGE_SCOPE } from "@/shared/constants/selfServiceScopes"; import { extractApiErrorMessage } from "@/shared/http/apiErrorMessage"; +import { hasProviderQuotaBypassScope } from "@/shared/constants/apiKeyPolicyScopes"; import { UsageLimitSettings } from "./components/UsageLimitSettings"; // Constants for validation @@ -818,17 +819,15 @@ export default function ApiManagerPageClient() { if (!debouncedSearchModel.trim()) return modelsByProvider; return modelsByProvider - .map( - ([provider, models]): ProviderGroup => [ - provider, - models.filter( - (m) => - matchesSearch(m.id, debouncedSearchModel) || - matchesSearch(m.name || "", debouncedSearchModel) || - matchesSearch(provider, debouncedSearchModel) - ), - ] - ) + .map(([provider, models]): ProviderGroup => [ + provider, + models.filter( + (m) => + matchesSearch(m.id, debouncedSearchModel) || + matchesSearch(m.name || "", debouncedSearchModel) || + matchesSearch(provider, debouncedSearchModel) + ), + ]) .filter(([, models]) => models.length > 0); }, [modelsByProvider, debouncedSearchModel]); @@ -958,6 +957,7 @@ export default function ApiManagerPageClient() { : 0; const hasThrottle = throttleDelayMs > 0; const hasManageScope = Array.isArray(key.scopes) && key.scopes.includes("manage"); + const hasProviderQuotaBypass = hasProviderQuotaBypassScope(key.scopes); const hasJsonStreamDefault = key.streamDefaultMode === "json"; const hasLocalUsageCommand = key.allowUsageCommand === true; const maxSessions = typeof key.maxSessions === "number" ? key.maxSessions : 0; @@ -985,9 +985,7 @@ export default function ApiManagerPageClient() {

- {visibleKeys.has(key.id) - ? (revealedKeys.get(key.id) ?? key.key) - : key.key} + {visibleKeys.has(key.id) ? (revealedKeys.get(key.id) ?? key.key) : key.key} {allowKeyReveal ? ( <> @@ -1114,6 +1112,12 @@ export default function ApiManagerPageClient() { USD quota )} + {hasProviderQuotaBypass && ( + + alt_route + Bypass quota policy + + )} {hasSessionLimit && ( group @@ -1588,6 +1592,9 @@ const PermissionsModal = memo(function PermissionsModal({ const [selfAccountQuotaEnabled, setSelfAccountQuotaEnabled] = useState( Array.isArray(apiKey?.scopes) && apiKey.scopes.includes(SELF_ACCOUNT_QUOTA_SCOPE) ); + const [bypassProviderQuotaPolicyEnabled, setBypassProviderQuotaPolicyEnabled] = useState( + hasProviderQuotaBypassScope(apiKey?.scopes) + ); const [maxSessions, setMaxSessions] = useState( typeof apiKey?.maxSessions === "number" && apiKey.maxSessions > 0 ? apiKey.maxSessions : 0 ); @@ -1822,6 +1829,7 @@ const PermissionsModal = memo(function PermissionsModal({ manageEnabled, selfUsageEnabled, selfAccountQuotaEnabled, + bypassProviderQuotaPolicyEnabled, }), allowAllEndpoints ? [] : selectedEndpoints, streamDefaultMode, @@ -1851,6 +1859,7 @@ const PermissionsModal = memo(function PermissionsModal({ manageEnabled, selfUsageEnabled, selfAccountQuotaEnabled, + bypassProviderQuotaPolicyEnabled, scheduleEnabled, scheduleFrom, scheduleUntil, @@ -2451,6 +2460,31 @@ const PermissionsModal = memo(function PermissionsModal({ />
+ {/* Advanced Provider Quota Policy Override */} +
+
+

Bypass provider quota cutoffs

+

+ Allows this key to ignore upstream provider/account cutoff policy during routing. API + key USD quotas still apply. +

+
+ +
+ {/* Disable Non-Public Models Toggle */}
diff --git a/src/app/(dashboard)/dashboard/api-manager/apiManagerScopes.ts b/src/app/(dashboard)/dashboard/api-manager/apiManagerScopes.ts index e8adb3546a..3e1692b179 100644 --- a/src/app/(dashboard)/dashboard/api-manager/apiManagerScopes.ts +++ b/src/app/(dashboard)/dashboard/api-manager/apiManagerScopes.ts @@ -1,7 +1,5 @@ -import { - SELF_ACCOUNT_QUOTA_SCOPE, - SELF_USAGE_SCOPE, -} from "@/shared/constants/selfServiceScopes"; +import { SELF_ACCOUNT_QUOTA_SCOPE, SELF_USAGE_SCOPE } from "@/shared/constants/selfServiceScopes"; +import { API_KEY_BYPASS_PROVIDER_QUOTA_SCOPE } from "@/shared/constants/apiKeyPolicyScopes"; const MANAGEMENT_SCOPE = "manage"; @@ -9,12 +7,14 @@ export interface CreateScopeOptions { manageEnabled: boolean; selfUsageEnabled?: boolean; selfAccountQuotaEnabled?: boolean; + bypassProviderQuotaPolicyEnabled?: boolean; } export interface PermissionScopeOptions { manageEnabled: boolean; selfUsageEnabled: boolean; selfAccountQuotaEnabled: boolean; + bypassProviderQuotaPolicyEnabled: boolean; } export function buildApiKeyCreateScopes(options: CreateScopeOptions): string[] { @@ -25,6 +25,9 @@ export function buildApiKeyCreateScopes(options: CreateScopeOptions): string[] { if (selfUsageEnabled && options.selfAccountQuotaEnabled === true) { scopes.push(SELF_ACCOUNT_QUOTA_SCOPE); } + if (options.bypassProviderQuotaPolicyEnabled === true) { + scopes.push(API_KEY_BYPASS_PROVIDER_QUOTA_SCOPE); + } return scopes; } @@ -41,6 +44,7 @@ export function mergeApiKeyPermissionScopes( SELF_ACCOUNT_QUOTA_SCOPE, options.selfUsageEnabled && options.selfAccountQuotaEnabled ); + setScope(scopes, API_KEY_BYPASS_PROVIDER_QUOTA_SCOPE, options.bypassProviderQuotaPolicyEnabled); return [...scopes]; } diff --git a/src/app/(dashboard)/dashboard/cli-code/CliCodePageClient.tsx b/src/app/(dashboard)/dashboard/cli-code/CliCodePageClient.tsx index 86e4af3360..1a5de54891 100644 --- a/src/app/(dashboard)/dashboard/cli-code/CliCodePageClient.tsx +++ b/src/app/(dashboard)/dashboard/cli-code/CliCodePageClient.tsx @@ -9,6 +9,7 @@ import { EXPECTED_CODE_COUNT } from "@/shared/schemas/cliCatalog"; import { CliToolCard, CliConceptCard, CliComparisonCard } from "@/shared/components/cli"; import { useToolBatchStatuses } from "@/shared/hooks/cli/useToolBatchStatuses"; import type { CliCatalogEntry } from "@/shared/schemas/cliCatalog"; +import CliProfileAutoSyncToggles from "./components/CliProfileAutoSyncToggles"; // ── Static catalogue slice ──────────────────────────────────────────────────── @@ -58,7 +59,9 @@ export default function CliCodePageClient({ machineId: _machineId }: CliCodePage useEffect(() => { let cancelled = false; fetch("/api/providers") - .then((res) => (res.ok ? res.json() : Promise.resolve({ connections: [] }))) + .then((res) => + res.ok ? res.json() : Promise.resolve({ connections: [] }) + ) .then((data) => { if (cancelled) return; const active = (data.connections ?? []).filter((c) => c.isActive !== false); @@ -99,8 +102,7 @@ export default function CliCodePageClient({ machineId: _machineId }: CliCodePage return CODE_TOOLS.filter(([id, tool]) => { // Search filter if (q) { - const haystack = - `${tool.name} ${tool.vendor} ${tool.description}`.toLowerCase(); + const haystack = `${tool.name} ${tool.vendor} ${tool.description}`.toLowerCase(); if (!haystack.includes(q)) return false; } @@ -131,6 +133,8 @@ export default function CliCodePageClient({ machineId: _machineId }: CliCodePage {/* Comparison card */} + + {/* Header bar */}
{/* Title + subtitle */} diff --git a/src/app/(dashboard)/dashboard/cli-code/components/CliProfileAutoSyncToggles.tsx b/src/app/(dashboard)/dashboard/cli-code/components/CliProfileAutoSyncToggles.tsx new file mode 100644 index 0000000000..0069de66d8 --- /dev/null +++ b/src/app/(dashboard)/dashboard/cli-code/components/CliProfileAutoSyncToggles.tsx @@ -0,0 +1,99 @@ +"use client"; + +import { useCallback, useEffect, useState } from "react"; +import { Toggle } from "@/shared/components"; + +type FlagEntry = { key: string; effectiveValue: string }; + +const CODEX_KEY = "OMNIROUTE_AUTO_SYNC_CODEX_PROFILES"; +const CLAUDE_KEY = "OMNIROUTE_AUTO_SYNC_CLAUDE_PROFILES"; + +function isOn(value: string | undefined): boolean { + return value === "true" || value === "1" || value === "yes" || value === "on"; +} + +/** + * Toggle card for the opt-in "auto-sync CLI profiles after model discovery" feature. + * Reads/writes the OMNIROUTE_AUTO_SYNC_{CODEX,CLAUDE}_PROFILES feature flags via + * /api/settings/feature-flags. Both default off; enabling one makes a provider model + * sync regenerate that tool's profile files from the live catalog. + */ +export default function CliProfileAutoSyncToggles() { + const [codexOn, setCodexOn] = useState(false); + const [claudeOn, setClaudeOn] = useState(false); + const [loading, setLoading] = useState(true); + const [savingKey, setSavingKey] = useState(null); + const [error, setError] = useState(null); + + const load = useCallback(async () => { + setLoading(true); + try { + const res = await fetch("/api/settings/feature-flags"); + if (!res.ok) throw new Error(`HTTP ${res.status}`); + const data = await res.json(); + const flags: FlagEntry[] = Array.isArray(data.flags) ? data.flags : []; + setCodexOn(isOn(flags.find((f) => f.key === CODEX_KEY)?.effectiveValue)); + setClaudeOn(isOn(flags.find((f) => f.key === CLAUDE_KEY)?.effectiveValue)); + setError(null); + } catch (err) { + setError(err instanceof Error ? err.message : "Failed to load settings"); + } finally { + setLoading(false); + } + }, []); + + useEffect(() => { + load(); + }, [load]); + + const persist = useCallback( + async (key: string, next: boolean, previous: boolean, apply: (v: boolean) => void) => { + apply(next); // optimistic + setSavingKey(key); + setError(null); + try { + const res = await fetch("/api/settings/feature-flags", { + method: "PUT", + headers: { "Content-Type": "application/json" }, + body: JSON.stringify({ key, value: next ? "true" : "false" }), + }); + if (!res.ok) throw new Error(`HTTP ${res.status}`); + } catch (err) { + apply(previous); // revert on failure + setError(err instanceof Error ? err.message : "Failed to save"); + } finally { + setSavingKey(null); + } + }, + [] + ); + + return ( +
+
+

CLI profile auto-sync

+

+ After a provider model sync, automatically regenerate CLI tool profiles from the live + catalog. Off by default — only profile files are written, never the active/default config. +

+
+
+ persist(CODEX_KEY, v, codexOn, setCodexOn)} + label="Codex profiles" + description="Regenerate ~/.codex/*.config.toml after model discovery." + /> + persist(CLAUDE_KEY, v, claudeOn, setClaudeOn)} + label="Claude Code profiles" + description="Regenerate ~/.claude/profiles//settings.json after model discovery." + /> +
+ {error ?

{error}

: null} +
+ ); +} diff --git a/src/app/(dashboard)/dashboard/combos/ResponseValidationEditor.tsx b/src/app/(dashboard)/dashboard/combos/ResponseValidationEditor.tsx new file mode 100644 index 0000000000..f31d9fe979 --- /dev/null +++ b/src/app/(dashboard)/dashboard/combos/ResponseValidationEditor.tsx @@ -0,0 +1,199 @@ +"use client"; + +import React from "react"; + +// Feature 4985 — per-combo response-body validation editor. Extracted into its own file +// so the combos god-component (page.tsx) stays lean. Emits the same declarative shape the +// Zod `responseValidationSchema` validates; backend evaluates it in validateResponseQuality. + +export type JsonPathCondition = "exists" | "nonEmpty" | "equals" | "notEquals"; + +export interface ResponseValidationValue { + forbiddenSubstrings?: string[]; + requiredSubstrings?: string[]; + minContentLength?: number; + jsonPathPredicates?: Array<{ + path: string; + condition: JsonPathCondition; + value?: string | number | boolean; + }>; +} + +function tr(t: ((key: string) => string) | undefined, key: string, fallback: string): string { + if (!t) return fallback; + try { + const v = t(key); + return v && v !== key ? v : fallback; + } catch { + return fallback; + } +} + +const linesToArray = (text: string): string[] => + text + .split("\n") + .map((s) => s.trim()) + .filter(Boolean); + +const arrayToLines = (arr?: string[]): string => (arr ?? []).join("\n"); + +const INPUT_CLASS = + "w-full text-xs py-1.5 px-2 rounded border border-black/10 dark:border-white/10 bg-transparent focus:border-primary focus:outline-none"; +const LABEL_CLASS = "text-[11px] font-medium text-text-muted block mb-0.5"; + +const CONDITIONS: JsonPathCondition[] = ["exists", "nonEmpty", "equals", "notEquals"]; + +export function ResponseValidationEditor({ + value, + onChange, + t, +}: { + value?: ResponseValidationValue | null; + onChange: (next: ResponseValidationValue | undefined) => void; + t?: (key: string) => string; +}) { + const v: ResponseValidationValue = value && typeof value === "object" ? value : {}; + + const emit = (draft: ResponseValidationValue) => { + const cleaned: ResponseValidationValue = {}; + if (draft.forbiddenSubstrings && draft.forbiddenSubstrings.length) + cleaned.forbiddenSubstrings = draft.forbiddenSubstrings; + if (draft.requiredSubstrings && draft.requiredSubstrings.length) + cleaned.requiredSubstrings = draft.requiredSubstrings; + if (typeof draft.minContentLength === "number" && draft.minContentLength > 0) + cleaned.minContentLength = draft.minContentLength; + if (draft.jsonPathPredicates && draft.jsonPathPredicates.length) + cleaned.jsonPathPredicates = draft.jsonPathPredicates; + onChange(Object.keys(cleaned).length ? cleaned : undefined); + }; + + const predicates = v.jsonPathPredicates ?? []; + + const updatePredicate = (index: number, patch: Partial<(typeof predicates)[number]>) => { + const next = predicates.map((p, i) => (i === index ? { ...p, ...patch } : p)); + emit({ ...v, jsonPathPredicates: next }); + }; + + return ( +
+

+ {tr( + t, + "responseValidationHelp", + "Fail over to the next target when a 200 OK body fails these checks (assistant content)." + )} +

+ +
+ +