fix(i18n): restore canonical Traditional Chinese glossary

feat(core): add Layer A capability filter at router (#5696)

.env.example

@@ -1414,7 +1414,7 @@ APP_LOG_TO_FILE=true
 
 # Whether call log pipeline capture stores stream chunks when enabled in settings.
 # Only applies when call_log_pipeline_enabled=true.
-# Default: true
+# Default: false (opt-in — saves disk: stream chunks are the biggest call-log artifact)
 # CALL_LOG_PIPELINE_CAPTURE_STREAM_CHUNKS=true
 
 # Maximum call log artifact size for pipeline captures, in KB.
@@ -1426,7 +1426,7 @@ APP_LOG_TO_FILE=true
 # bodies is retained in the database.
 # Used by: open-sse/handlers/chatCore.ts — cloneBoundedChatLogPayload()
 # CHAT_LOG_TEXT_LIMIT=65536                  # Max string length before truncation (default: 64 KB)
-# CHAT_LOG_ARRAY_TAIL_ITEMS=24               # Number of array items retained from tail (default: 24)
+# CHAT_LOG_ARRAY_TAIL_ITEMS=128              # Number of array items retained from tail (default: 128)
 # CHAT_LOG_MAX_DEPTH=6                       # Max nesting depth before truncation (default: 6)
 # CHAT_LOG_MAX_OBJECT_KEYS=80                # Max object keys retained (default: 80, 0 = no limit)
 
@@ -1893,7 +1893,7 @@ APP_LOG_TO_FILE=true
 
 # Log request shape (content-type + content-length) for large chat payloads.
 # Used by: src/app/api/v1/chat/completions/route.ts. Set to "0" to silence.
-# Default: enabled.
+# Default: disabled (opt-in).
 # OMNIROUTE_LOG_REQUEST_SHAPE=1
 
 # Write raw (untruncated) request/response JSON in call log artifacts.
@@ -2514,3 +2514,46 @@ QUOTA_STORE_DRIVER=sqlite              # sqlite | redis
 # URL the dashboard's "Support the project" button opens (payment/plans
 # page). No pricing/value lives in this repo — only the link.
 # RADAR_SUPPORTER_PLANS_URL=https://radar.omniroute.online/planos
+
+# ═══════════════════════════════════════════════════════════════════════════════
+# 27. RELEASE v3.8.50 ADDITIONS
+# ═══════════════════════════════════════════════════════════════════════════════
+
+# Heavy chat admission queue wait before returning retryable 503. Set 0 for the
+# legacy immediate rejection. Used by: src/shared/middleware/chatBodyAdmission.ts.
+# Default: 5000 (5 seconds)
+# OMNIROUTE_CHAT_ADMISSION_QUEUE_MS=5000
+
+# Timeout for /api/jobs/:id/run-now while it waits for an in-flight run.
+# Used by: src/app/api/jobs/[id]/run-now/route.ts. Default: 30000 (30 seconds)
+# OMNIROUTE_RUNNOW_TIMEOUT_MS=30000
+
+# Maximum request/response body size before chat-log summarization, in KiB.
+# Used by: src/lib/chatLogTruncation.ts. Default: 1024
+# CHAT_LOG_MAX_BODY_KB=1024
+
+# Adobe Firefly browser renewal and durable session cache (enabled by default).
+# Used by: open-sse/services/adobeFireflySession.ts.
+# ADOBE_FIREFLY_BROWSER_REFRESH=1
+# ADOBE_FIREFLY_SESSION_DISK=1
+# Minimum spacing between submissions and the extra pause after every third success.
+# ADOBE_FIREFLY_MIN_SUBMIT_GAP_MS=12000
+# ADOBE_FIREFLY_BATCH_EXTRA_GAP_MS=15000
+# Chrome CDP runtime used by Adobe Firefly renewal. True headless is debug-only:
+# Adobe colligo normally rejects risk tokens minted without a headed browser.
+# ADOBE_FIREFLY_CHROME_CDP_PORT=9334
+# ADOBE_FIREFLY_CHROME_VISIBLE=0
+# ADOBE_FIREFLY_CHROME_HEADLESS=0
+# ADOBE_FIREFLY_CHROME_FORCE_RESTART=0
+# ADOBE_FIREFLY_CHROME_PING=auto
+# ADOBE_FIREFLY_LOGIN_WAIT_MS=0
+# ADOBE_FIREFLY_FORTER_WAIT_MS=45000
+# Optional absolute Chrome executable; auto-detected when unset.
+# CHROME_PATH=
+
+# Telegram Mini App bridge. The update endpoint remains disabled while the bot
+# token is unset. Used by: src/lib/telegram/* and src/app/api/telegram/update/route.ts.
+# TELEGRAM_BOT_TOKEN=
+# TELEGRAM_DEFAULT_MODEL=auto/chat
+# TELEGRAM_BOT_API_BASE=https://api.telegram.org
+# TELEGRAM_WEBHOOK_TIMEOUT_MS=60000

.github/workflows/ci.yml

@@ -811,7 +811,12 @@ jobs:
 
   test-bun-sqlite:
     name: Bun SQLite Compatibility
-    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        os: [ubuntu-latest, windows-latest]
+      fail-fast: false
+    runs-on: ${{ matrix.os }}
+    continue-on-error: ${{ matrix.os == 'windows-latest' }}
     timeout-minutes: 10
     needs: changes
     if: ${{ github.event_name != 'pull_request' || (needs.changes.outputs.code == 'true' && github.event.pull_request.draft == false) }}
@@ -824,6 +829,15 @@ jobs:
           node-version: ${{ env.CI_NODE_VERSION }}
           cache: npm
       - uses: ./.github/actions/npm-ci-retry
+      - name: Install Bun (Windows)
+        if: runner.os == 'Windows'
+        shell: pwsh
+        run: |
+          powershell -c "iwr bun.sh/install.ps1 -useb | iex"
+          echo "$env:USERPROFILE\.bun\bin" | Out-File -FilePath $env:GITHUB_PATH -Append
+      - name: Install Bun (non-Windows)
+        if: runner.os != 'Windows'
+        run: npm install -g bun
       - run: npm run test:bun:db
 
   test-vitest:

.gitignore

@@ -72,7 +72,6 @@ yarn-error.log*
 # env files (can opt-in for committing if needed)
 .env*
 !.env.example
-!.env.devin-bridge.example
 !.env.homolog.example
 # Provider API keys (never commit)
 *.api-key
@@ -172,6 +171,7 @@ config/quality/test-impact-map.json
 # GitNexus local index
 .gitnexus
 .worktrees
+bin/omniroute.mjs
 
 # Consistent with .dockerignore / .npmignore
 .omc/
@@ -201,17 +201,12 @@ scripts/i18n/_pending-keys.json
 .codegraph/
 
 # Fumadocs generated source
-/.source/
-
-# Temporary local worktrees used to build unpublished npm tarballs
-/.deploy-build-*/
+.source/
 
 # AI agent local settings and configs
 .agents/
 .antigravitycli/
 .claude/
-!tests/fixtures/devin-bridge/e2e-workspace/.claude/
-!tests/fixtures/devin-bridge/e2e-workspace/.claude/**
 
 # PR Reviews and local feedback files
 pr_reviews*.json
@@ -226,6 +221,26 @@ CODEX-SETUP-PROMPT.md
 # Quality ratchet — métricas efêmeras (baseline commitado em config/quality/; métricas não)
 config/quality/quality-metrics.json
 
+# Electron desktop build output unpacked into the repo root.
+# `electron-builder` (squirrel-windows target) unpacks the packaged app — the
+# entire Chromium runtime, ~24k files — directly into the repository root.
+# Every rule below is ROOT-ANCHORED (leading `/`) on purpose: a bare `locales/`
+# or `resources/` would also swallow tracked sources such as the CLI
+# translations in `bin/cli/locales/*.json`.
+/OmniRoute.exe
+/Uninstall OmniRoute.exe
+/uninstallerIcon.ico
+/locales/
+/resources/
+/*.pak
+/*.dll
+/icudtl.dat
+/snapshot_blob.bin
+/v8_context_snapshot.bin
+/vk_swiftshader_icd.json
+/LICENSE.electron.txt
+/LICENSES.chromium.html
+
 # Runtime logs (diretório local, nunca versionado)
 /logs/
 -home-diegosouzapw-dev-automações-bots-yt-downloader-20260504 .txt
@@ -238,10 +253,7 @@ omniroute.md
 
 # mise configuration
 mise.toml
-# release-green artifacts (.gitignore has no inline comments — a trailing
-# `# ...` becomes part of the pattern, so it must sit on its own line).
-# Already covered by /_*/ above; kept explicit for discoverability.
-_artifacts/
+_artifacts/          # release-green artifacts
 .claude-flow/
 
 # ESLint file cache (npm run lint --cache / complexity ratchets)
@@ -251,8 +263,6 @@ _artifacts/
 
 # CI/local quality artifacts (eslint-results.json, quality-ratchet.md, etc.)
 .artifacts/
-# Isolated Devin bridge workspaces, evidence, and test databases
-.sandbox/
 
 # Homologation E2E suite (npm run homolog) — real-environment credentials + report output
 .env.homolog
@@ -260,12 +270,11 @@ tests/homolog/.auth/
 tests/homolog/ui/.auth/
 homolog-report/
 docker-compose.yml.bak
-.playwright-cli/
-# Playwright screenshot/log output. Today every artifact happens to land inside
-# output/**/.playwright-cli/ (covered above), but anything written directly to
-# output/ would otherwise show up as untracked.
-/output/
 
-# _tasks e um repo git SEPARADO (ver AGENTS.md). _tasks/ (com barra) NAO ignora um
-# SYMLINK _tasks; /_tasks (ancorado) cobre symlink/dir na raiz (incidente 2026-08-08).
+# _tasks e um repo git SEPARADO (ver AGENTS.md). A linha _tasks/ (com barra) NAO
+# ignora um SYMLINK chamado _tasks; /_tasks (ancorado) cobre arquivo/symlink/dir na raiz
+# e impede que um git add -A recapture o symlink (incidente 2026-08-08).
 /_tasks
+
+# CLI local cache/state
+.playwright-cli

AGENTS.md

@@ -627,7 +627,7 @@ procedures are in [`docs/architecture/QUALITY_GATES.md`](docs/architecture/QUALI
   complexity) must not regress vs `quality-baseline.json`. Update via
   `npm run quality:ratchet -- --update` when a metric genuinely improves.
 - Job `test-vitest` runs `npm run test:vitest` (MCP tools, autoCombo, cache) — blocking.
-  `test:vitest:ui` is advisory until UI component tests are triaged.
+  `test:vitest:ui` has been blocking since PR #7127.
 
 **Allowlist policy (short form):** Fix the cause; use the allowlist only for pre-existing
 violations you cannot fix in the same PR. Add a comment with justification + issue number.

Dockerfile

@@ -93,7 +93,15 @@ RUN --mount=type=cache,id=npm-cache,target=/root/.npm \
 # build from 17min to 9min on the same 32-core box. Webpack stays available as the
 # escape hatch: `--build-arg`/-e OMNIROUTE_USE_TURBOPACK=0.
 # See docs/ops/QUALITY_GATE_PLAYBOOK.md Parte 6.
-ENV OMNIROUTE_USE_TURBOPACK=1
+#
+# Declared as ARG+ENV, not a bare ENV: a bare ENV shadows any same-named ARG for
+# the rest of the stage, so `--build-arg OMNIROUTE_USE_TURBOPACK=0` was silently
+# ignored and the escape hatch above only ever worked via `-e` at runtime, never
+# at build time. Turbopack compiles in native Rust memory that lives outside the
+# V8 heap, so OMNIROUTE_BUILD_MEMORY_MB cannot bound it and a memory-constrained
+# build host gets SIGKILLed by the cgroup OOM killer with no error message.
+ARG OMNIROUTE_USE_TURBOPACK=1
+ENV OMNIROUTE_USE_TURBOPACK="${OMNIROUTE_USE_TURBOPACK}"
 
 # Next.js basePath is fixed at build time; pass OMNIROUTE_BASE_PATH here when the
 # image should serve under a reverse-proxy subpath without a runtime patch.

Makefile

@@ -0,0 +1,69 @@
+.PHONY: help install dev start build build-release lint typecheck typecheck-strict \
+        test test-unit test-vitest test-coverage test-all test-integration test-e2e \
+        check check-cycles check-docs env-sync clean
+
+# OmniRoute — convenience wrapper around the npm scripts.
+# All targets delegate to the canonical package.json scripts (single source of truth).
+
+help: ## Show this help
+	@grep -E '^[a-zA-Z_-]+:.*?## .*$$' $(MAKEFILE_LIST) | awk 'BEGIN {FS = ":.*?## "}; {printf "  \033[36m%-18s\033[0m %s\n", $$1, $$2}'
+
+install: ## Install dependencies (auto-generates .env from .env.example)
+	npm install
+
+dev: ## Dev server at http://localhost:20128
+	npm run dev
+
+start: ## Production server (requires a prior build)
+	npm run start
+
+build: ## Production build (Next.js 16 standalone)
+	npm run build
+
+build-release: ## Release build
+	npm run build:release
+
+lint: ## ESLint (0 errors expected)
+	npm run lint
+
+typecheck: ## TypeScript check (core)
+	npm run typecheck:core
+
+typecheck-strict: ## Strict check (no implicit any)
+	npm run typecheck:noimplicit:core
+
+test: ## Unit tests (Node native runner)
+	npm run test:unit
+
+test-unit: ## Alias for `test`
+	npm run test:unit
+
+test-vitest: ## Vitest (MCP server, autoCombo, cache)
+	npm run test:vitest
+
+test-coverage: ## Unit tests + coverage gate (60/60/60/60)
+	npm run test:coverage
+
+test-all: ## All suites (unit + vitest + ecosystem + e2e)
+	npm run test:all
+
+test-integration: ## Integration tests
+	npm run test:integration
+
+test-e2e: ## E2E (Playwright)
+	npm run test:e2e
+
+check: ## lint + test combined
+	npm run check
+
+check-cycles: ## Detect circular dependencies
+	npm run check:cycles
+
+check-docs: ## Validate documentation (incl. fabricated-docs)
+	npm run check:docs-all
+
+env-sync: ## Sync .env from .env.example
+	npm run env:sync
+
+clean: ## Remove build artifacts
+	rm -rf .build dist coverage .eslintcache

changelog.d/features/5696-layer-a-capability-filter.md

@@ -0,0 +1 @@
+- **feat(core):** add Layer A capability filter at router (#5696)

changelog.d/features/8468-bun-windows-ci-coverage.md

@@ -0,0 +1 @@
+- feat(ci): add windows-latest leg to test-bun-sqlite job (#8468)

changelog.d/features/9268-gemini-schema-recursive-type-empty-choices.md

@@ -0,0 +1 @@
+- **feat(gemini):** recursive type:object injection in schema normalizer + empty choices interceptor for streaming (#9268)

changelog.d/features/9322-nanogpt-endpoint-surface.md

@@ -0,0 +1 @@
+- **feat(providers):** expanded the NanoGPT (`nano-gpt.com`) upstream provider from chat-only to the full OpenAI-compatible endpoint surface: audio transcriptions (`/api/v1/audio/transcriptions`), audio speech (`/api/v1/audio/speech`), video generation (`/api/v1/video/generations`), embeddings (`/v1/embeddings`), and the Responses API (`responsesBaseUrl` → `/api/v1/responses`) ([#9322](https://github.com/diegosouzapw/OmniRoute/issues/9322))

changelog.d/features/9415-newapi-sub2api-aggregator-balance.md

@@ -1 +1 @@
-- **sse:** New-API / One-API / Sub2API aggregator balance detection for compatible nodes — with the "Aggregator Gateway" toggle on, OmniRoute queries the aggregator's `/api/user/self` to read the account balance, shows it as a dashboard badge and lets quota-preflight routing skip exhausted accounts. Gated by the `NEWAPI_AGGREGATOR_BALANCE` feature flag (default off), with a `quotaPerUnit` override for aggregators that do not use the default 500000 units/$1 rate ([#9415](https://github.com/diegosouzapw/OmniRoute/issues/9415))
+- **feat(sse):** New-API/One-API/Sub2API aggregator balance detection for compatible provider nodes — when the "Aggregator Gateway" toggle is enabled, OmniRoute queries the aggregator's `/api/user/self` endpoint to detect the account balance; the dashboard shows a balance badge and quota-preflight routing skips exhausted accounts. Gated by the `NEWAPI_AGGREGATOR_BALANCE` feature flag (default: off), with a custom `quotaPerUnit` override for aggregators that use a different rate than the default 500000 units/$1 ([#9415](https://github.com/diegosouzapw/OmniRoute/issues/9415))

changelog.d/features/9782-modality-bridge-settings.md

@@ -0,0 +1 @@
+- **feat(settings):** add a dedicated Modality Bridge settings page with Vision controls, runtime stats, and URL-addressable Audio and Video tabs ([#9782](https://github.com/diegosouzapw/OmniRoute/pull/9782))

changelog.d/fixes/9496-kimi-k3-responses-replay.md

@@ -0,0 +1 @@
+- fix(translator): preserve authentic K3 Responses reasoning by model across providers, keep it on the matching assistant turn, and make Kimi Coding prefer client reasoning then cached replay before its empty-marker fallback (#9496)

changelog.d/fixes/9675-anonymous-fallback-disable-toggle.md

@@ -0,0 +1 @@
+- **fix(providers):** new per-provider `noAuthFallbackDisabledProviders` setting lets operators disable the synthetic anonymous (no-auth) credential fallback for API-key providers whose static definition declares `anonymousFallback: true` (e.g. `opencode-go`, `opencode-zen`) — upstream endpoints now reject anonymous requests with `401 Missing API key`, so the fallback added latency and caused UI health/reconnect churn. Real keyed connections keep working and recover automatically once quota state clears; true no-auth providers (`opencode`, `mimocode`, …) are unaffected, with `blockedProviders` remaining their disable mechanism. Default behavior is unchanged ([#9675](https://github.com/diegosouzapw/OmniRoute/pull/9675))

changelog.d/fixes/9695-docker-bundler-build-arg.md

@@ -0,0 +1 @@
+- **fix(docker):** `--build-arg OMNIROUTE_USE_TURBOPACK=0` now reaches the builder stage — a bare `ENV` was shadowing the `ARG`, so the documented webpack escape hatch was silently ignored and memory-constrained hosts were OOM-killed with no error output ([#9695](https://github.com/diegosouzapw/OmniRoute/pull/9695))

changelog.d/fixes/9719-combo-connection-pins.md

@@ -0,0 +1 @@
+- fix(db): clear stale combo connection pins when provider connections are deleted (#9719)

changelog.d/fixes/9730-persist-rtk-renderers.md

@@ -0,0 +1 @@
+- fix(compression): persist `enableRenderers` through `normalizeRtkConfig` so RTK renderer settings survive a DB round-trip ([#9730](https://github.com/diegosouzapw/OmniRoute/pull/9730))

changelog.d/fixes/9737-vietnamese-locale-parity.md

@@ -0,0 +1 @@
+- Restore Vietnamese locale parity after the entity-normalization sync dropped Radar, provider, and mini-playground messages.

changelog.d/fixes/9788-model-catalog-gateway-permissions.md

@@ -0,0 +1 @@
+- **fix(api):** Model catalogs no longer expose functional gateway mirrors unless the API key permits the mirror's final public model ID ([#9788](https://github.com/diegosouzapw/OmniRoute/pull/9788)) — thanks @xz-dev

changelog.d/fixes/9826-command-code-responses-usage.md

@@ -0,0 +1 @@
+- **fix(executors):** preserve Command Code usage in `/v1/responses` streams so Codex clients receive real input, output, cache, and reasoning token counts ([#9826](https://github.com/diegosouzapw/OmniRoute/pull/9826)) — thanks @MrShitFox

changelog.d/fixes/9828-codex-redundant-oneof-enum.md

@@ -0,0 +1 @@
+- **fix(executors):** prevent intermittent Codex `upstream_empty_response` errors for tool schemas that combine `oneOf` const branches with a matching sibling `enum` by removing only the semantically redundant `oneOf`; bare, narrowing, non-matching, and type-discriminated `oneOf` schemas remain unchanged. ([#9828](https://github.com/diegosouzapw/OmniRoute/pull/9828))

changelog.d/fixes/9834-cursor-selected-image-blobid.md

@@ -0,0 +1 @@
+- **fix(cursor):** SelectedImage uses `blobIdWithData` + session blobStore, with JPEG soft-cap prep via sharp ([#9834](https://github.com/diegosouzapw/OmniRoute/pull/9834)) — thanks @yansigit

changelog.d/fixes/i18n-cc-alias-unclosed-tags.md

@@ -0,0 +1 @@
+- fix(i18n): re-escape CC discovery-alias `claude/<provider>/<model>` to HTML entities so next-intl stops logging INVALID_MESSAGE: UNCLOSED_TAG on provider detail pages (#8747 regression)

changelog.d/fixes/release-v3850-base-drifted-test-expectations.md

@@ -0,0 +1 @@
+- **fix(test):** reconcile test expectations that drifted from the code they guard on `release/v3.8.50` — auth/vision/provider schema snapshots, and three context-aware combo compatibility assertions that contradicted the same file's own stated contract (catalog-too-small targets stay available as runtime fallback rather than being dropped). The combo assertions were masked by an unresolved import that stopped `combo.ts` from loading at all, so they only become reachable once that import is repaired.

changelog.d/fixes/release-v3850-post-sweep-locale-lint.md

@@ -0,0 +1 @@
+- Repair release-sweep regressions in locale and environment contracts, package metadata, scripts, dependency, size and dead-code ratchets, OpenAPI coverage, Telegram error sanitization, Openference public-credential handling, DB-module classification, resilience UI test assertions, strict CodeBuddy CN tests, Lite compression typing, and the job-registry migration number.

changelog.d/fixes/release-v3850-validator-test-masking-timeout.md

@@ -0,0 +1 @@
+- Let the release-green validator finish the test-masking gate on loaded runners while preserving the existing timeout for every other full-CI gate, and report Node.js `ETIMEDOUT` errors as explicit timeout failures.

changelog.d/maintenance/9839-v3850-quality-ratchets.md

@@ -0,0 +1 @@
+- Reconcile the final v3.8.50 bundle-size and file-size ratchets against the measured release tip, preserving exact direction-down ceilings and their source attribution.

config/quality/dependency-allowlist.json

@@ -44,6 +44,7 @@
     "commander",
     "concurrently",
     "cross-env",
+    "cron-parser",
     "csv-stringify",
     "ctrf",
     "dompurify",
@@ -114,6 +115,7 @@
     "recharts",
     "safe-regex",
     "selfsigned",
+    "sharp",
     "size-limit",
     "smol-toml",
     "socks",

config/quality/eslint-suppressions.json

@@ -1673,7 +1673,7 @@
   },
   "tests/unit/base-executor-sanitize-effort.test.ts": {
     "@typescript-eslint/no-explicit-any": {
-      "count": 48
+      "count": 6
     }
   },
   "tests/unit/batch-deletion.test.ts": {
@@ -2036,11 +2036,6 @@
       "count": 3
     }
   },
-  "tests/unit/codebuddy-cn-provider.test.ts": {
-    "@typescript-eslint/no-explicit-any": {
-      "count": 3
-    }
-  },
   "tests/unit/codex-banked-reset-credits-5199.test.ts": {
     "@typescript-eslint/no-explicit-any": {
       "count": 7
@@ -3339,4 +3334,4 @@
       "count": 5
     }
   }
-}
+}

config/quality/file-size-baseline.json

@@ -1,5 +1,9 @@
 {
   "_rebaseline_2026_08_09_9296_adobe_media_capabilities": "PR #9296 (artickc, fix/adobe-firefly-model-capabilities) own growth: src/app/api/v1/models/catalog.ts 1590->1597 (+7). The image and video catalog serializers now expose the already-normalized Adobe Firefly discovery capability data (media_capabilities, plus the existing video modality/size fields) at their only response-emission chokepoints. The discovery parser and capability normalization remain in open-sse/services/adobeFireflyModels.ts; extracting these seven serialization fields would obscure the catalog contract. Covered by tests/unit/adobe-firefly.test.ts and tests/unit/image-upscale.test.ts.",
+  "_rebaseline_2026_08_08_v3850_base_drift_batch_9757": "Base drift on release/v3.8.50, not own growth: the 08-06..08-08 merge batches grew 12 already-frozen (or newly-landed) files without carrying their rebaselines — the dedicated rebaseline PR #9616 was closed as 'superseded' but its file-size entries never actually reached the base, and later merges (#8894 combos page, #9539 EditConnectionModal, #8895 models route, #9294/#9293 catalog, #9541 db/core, #8970 tokenHealthCheck, #8925 mcp schemas+server, #8890 accountFallback, #9467 chat.ts, #8931 openai-to-kiro, ProxyRegistryManager) kept growing them. All 12 values re-measured on THIS branch's tree (= pure tip + this PR's 1-line chat.ts fix, which adds zero lines). This PR's own source changes (chat.ts identifier restore, stream.ts format carve-out) do not grow any frozen file past these values.",
+  "_rebaseline_2026_08_08_migration_135_collision": "fix(db): resolve migration version 135 numbering collision — #9449's 135_connection_runtime_state.sql and #8908's 135_migrate_model_capability_max_token.sql both claimed version 135 (#9449 branched before #8908 merged and never got renumbered before landing on release/v3.8.50), which threw 'Migration version collision detected' the moment ANY code touched the database — a fresh install/deploy from this tip cannot even boot. Renumbered the later-landing file to 140 (next free slot) and added the matching isSchemaAlreadyApplied('140') retroactive guard, matching the established pattern already used for the prior 135/136 -> 137/138 renumber in the same file. Own growth: src/lib/db/migrationRunner.ts 1084->1094 (+10, the new case block) — irreducible, matches the existing per-case guard pattern exactly. Covered by tests/unit/migration-135-numbering-collision.test.ts (2/2), confirmed failing (reproducing the exact live crash) against the pre-fix colliding filenames, passing after.",
+  "_rebaseline_2026_08_08_9183_reasoning_cache_index_sync": "Extracted fix(responses-api): sync reasoning-cache write index with the fixed read side (from the originally-authored #9183) — chatCore.ts's write side cached every response under a hardcoded messageIndex:0, and translator/index.ts's plain-turn (non-tool-call) cache-key lookup ALSO still hardcoded messageIndex 0 at its call site (a second, previously-undiscovered instance of the same hardcoding bug, found while re-verifying this fix against the current upstream tip — the two never agreed once a conversation went past its first assistant turn, so DeepSeek/Xiaomi-mimo plain-turn reasoning replay silently missed the cache). Own growth: open-sse/handlers/chatCore.ts 5034->5042 (+8, computing messageIndex from the incoming request's message count at both the streaming and non-streaming cache-write call sites) — irreducible call-site wiring. Covered by tests/unit/reasoning-cache.test.ts (new end-to-end write/read regression test, rebaselined below) and tests/unit/translator-helper-branches.test.ts fixture updates. Other #9183 sub-fixes (output_index collision prevention, reasoning-content-alias generalization) were originally assumed already superseded by upstream's own independent fix — a live incident 2026-08-08 disproved that for the message-vs-tool-call collision case specifically (fixed separately in #9822); not re-extracted here since this PR's own scope is the narrower messageIndex sync only.",
+  "_rebaseline_2026_08_02_9259_rolling_rpm": "PR #9259 (issue #8733) own growth: open-sse/services/rateLimitManager.ts baseline 1060->1167 (+107; final source 1153). The existing withRateLimit chokepoint now composes process-local rolling RPM leases with Bottleneck admission, releases pre-dispatch leases on queue timeout/abort/connection disable, preserves caller abort reasons, and wires 429/header state into the extracted rollingRpmGate.ts. The remaining growth is irreducible lifecycle wiring at the dispatch boundary plus the real watchdog test hooks needed to verify queued-wedge recovery; moving it further would obscure lease ownership and Bottleneck cleanup. Covered by the focused rate-limit manager/sliding-window suite (33/33); distributed multi-instance coordination remains explicitly out of scope.",
   "_rebaseline_2026_07_24_8470_hyperagent_sticky_thread": "PR #8470 (artickc, fix/hyperagent-tool-loop-thread-sticky) own growth: open-sse/executors/hyperagent.ts 936->1025 (wc -l; check-file-size.mjs counts via split(\"\\n\").length so the gate sees 937->1026, +89, crosses the 1000 cap). Fixes a real bug where a reverse-conversion proxy (text-Intent/JSON to Claude Code native tool_calls) rewrites assistant messages between agentic tool-loop turns, breaking HyperAgent’s conversation-prefix fingerprint and cold-starting the thread mid tool-loop. Adds Anthropic tool_use/tool_result flattening to extractMessageText() plus a new rootUserFingerprint()/root-key lookup tier in resolveHyperAgentThreadBinding()/storeHyperAgentThreadAfterTurn() so the thread stays sticky across the tool loop. Cohesive additions inside the existing single-file executor; not extractable without splitting the executor mid-request-flow. Covered by tests/unit/executor-hyperagent.test.ts (19/19, +5 new cases for tool_result/tool_use flattening + root-key stickiness). Pre-merge review flagged a cross-conversation root-key collision risk (tracked in the PR’s own mandatory pre-merge checklist, not yet addressed) — unrelated to this file-size ratchet, tracked separately by /fix-prs.",
   "_rebaseline_2026_07_25_8494_capability_filter_fail_closed": "PR #8494 (fix/capability-filters-fail-closed, #8488) own growth: open-sse/services/combo.ts 3640->3693 (+53) adds a fail-closed guard after filterTargetsByRequestCompatibility() — when every eligible target is excluded by request-capability filtering (vision/tools/etc) instead of quota/health, the combo now returns an explicit `capability_mismatch` 400 (describeCapabilityFilterExhaustion, imported from combo/comboStructure.ts) rather than silently falling through to a generic no-targets error, plus a `compatFilterFailOpen` escape hatch (combo config OR settings) mirrored at both the main/auto and round-robin call sites for symmetry. combo/comboStructure.ts (previously under cap, un-frozen) grows 794->918 (+124) — new home for describeCapabilityFilterExhaustion + providerSupportsEmulatedToolCalling (#5240 emulated tool-calling exemption so fail-closed does not regress prompt-emulation-only combos like all-chatgpt-web). Irreducible orchestration wiring at the existing filter chokepoint (same precedent as #7301's universal-cooldown-retry generalization). Companion test tests/unit/combo-routing-engine.test.ts 3409->3449 (+40, fail-closed/fail-open coverage across both call sites) also rebaselined. Covered by tests/unit/8488-capability-filter-fail-closed.test.ts (new) + 95/95 passing across both files. Structural shrink of combo.ts tracked in #3501.",
   "_rebaseline_2026_07_25_8499_ts7_result_union_predicates": "PR #8499 (backryun, chore/ts7-types-executor-scattered) own growth: muse-spark-web.ts 1396->1405 (+9, irreducible). Under this workspace's `strictNullChecks: false`, the boolean-literal discriminant on `GraphqlResult` (`{ ok: true } | { ok: false; error: string }`) narrows the positive `.ok===true` branch but leaves `!result.ok` at the full union under TS7, making `.error` unreachable to the checker at the two call sites (warmup, mode-switch). Fixed by adding a single `isGraphqlFailure()` type-predicate helper (doc comment + 3-line body) reused at both call sites instead of duplicating the predicate inline — not extractable to a shared module without splitting a single-file executor's local narrowing helper out of its own file. Covered by the existing muse-spark-web executor test suite (no behavior change, pure narrowing fix).",
@@ -159,9 +163,12 @@
   "_rebaseline_2026_06_20_4389_thinking_toolchoice": "Re-baseline base.ts 1387->1399 (#4389): tool_choice-forced thinking guard at the existing Claude wire-image injection chokepoint (effThinking gate avoids the Anthropic 400 when tool_choice forces a tool). Cohesive guard; structural shrink tracked in #3501.",
   "_rebaseline_2026_07_18_6979_codex_test": "PR #6979 own growth: executor-codex.test.ts 1340->1347 (+7 = generalized ensureThinkingBudget assertion added to the existing codex thinking-budget cases). antigravity-test bump 942->977 REVERTED here: #7408's test split dropped that file to 888, so this PR's +35 fits under the original 942 frozen cap.",
   "_rebaseline_2026_07_24_8354_logs_timeline_sidebar": "PR #8354 (hartmark, feature/scrolling-log) own growth: src/shared/constants/sidebarVisibility/sections.ts 812->820 (+8, the single new logs-timeline SidebarItemDefinition entry added to LOGS_GROUP.items for the new /dashboard/logs/timeline scrolling request-timeline page). Irreducible data-literal wiring at the existing sidebar-sections chokepoint, same shape as every other item in the file; not extractable without an ad-hoc single-item exception to the file's otherwise-uniform multi-line item style.",
+  "_rebaseline_2026_08_09_v3850_post_sweep_tip": "Release-captain reconciliation of absolute file-size drift on pure tip 382449d593 after the authorized cherry-pick wave. The affected production growth already belongs to merged, tested commits: Adobe Firefly CDP/session recovery (#9881), model capability serialization (#9296), Modality Bridge request wiring (#9759), disconnect-grace/reasoning-cache chatCore wiring (#9653/#9183), stacked Lite precedence, and Responses tool-call index/argument handling (#9843 plus the release translator fixes). This repair adds only the compact migration-146 retroactive guard, covered by db-job-registry-migration-renumber-139.test.ts. Values are the exact check:file-size split-newline measurements and remain shrink-only; structural decomposition remains tracked by the existing #3501 notes.",
   "cap": 1000,
   "testCap": 1000,
   "testFrozen": {
+    "tests/unit/adobe-firefly.test.ts": 1136,
+    "tests/unit/reasoning-cache.test.ts": 1035,
     "_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).",
     "_rebaseline_2026_07_02_5928_base_red": "web-cookie-providers-new.test.ts 845->850: #5928 (test(security) Kimi Web URL host parse, CodeQL #689) grew the file +5 lines and merged into release/v3.8.44 WITHOUT rebaselining, leaving a fast-gates base-red that blocked every subsequent PR->release. Test growth is legitimate (a security regression test); maintainer absorbs the drift here. Frozen at 850.",
     "_rebaseline_2026_07_09_6126_clinepass_dualauth": "#6126 (ClinePass dual-auth) own test growth: oauth-providers-config.test.ts 842->845 (+3: clinepass key/config/required-fields entries reusing the Cline WorkOS flow config, needed after registering clinepass in the oauth.ts PROVIDERS enum).",
@@ -350,7 +357,7 @@
     "open-sse/executors/deepseek-web.ts": 1148,
     "open-sse/executors/grok-web.ts": 1044,
     "open-sse/executors/muse-spark-web.ts": 1405,
-    "open-sse/handlers/chatCore.ts": 5034,
+    "open-sse/handlers/chatCore.ts": 5061,
     "open-sse/handlers/imageGeneration.ts": 3101,
     "open-sse/handlers/responseSanitizer.ts": 1128,
     "open-sse/handlers/search.ts": 1536,
@@ -359,12 +366,15 @@
     "open-sse/mcp-server/server.ts": 1448,
     "open-sse/mcp-server/tools/advancedTools.ts": 1120,
     "open-sse/services/accountFallback.ts": 1978,
-    "open-sse/services/adobeFireflyClient.ts": 2385,
+    "open-sse/services/adobeFireflyBrowserLogin.ts": 1362,
+    "open-sse/services/adobeFireflyChromeRuntime.ts": 1201,
+    "open-sse/services/adobeFireflyClient.ts": 2999,
+    "open-sse/services/adobeFireflySession.ts": 1003,
     "open-sse/services/claudeCodeCompatible.ts": 1202,
     "open-sse/services/combo.ts": 3648,
-    "open-sse/services/compression/strategySelector.ts": 1060,
+    "open-sse/services/compression/strategySelector.ts": 1061,
     "open-sse/services/rateLimitManager.ts": 1167,
-    "open-sse/translator/response/openai-responses.ts": 1204,
+    "open-sse/translator/response/openai-responses.ts": 1271,
     "open-sse/utils/cursorAgentProtobuf.ts": 1505,
     "open-sse/utils/stream.ts": 2889,
     "src/app/(dashboard)/dashboard/HomePageClient.tsx": 1388,
@@ -391,7 +401,7 @@
     "src/app/api/v1/models/catalog.ts": 1597,
     "src/lib/db/apiKeys.ts": 1529,
     "src/lib/db/core.ts": 1639,
-    "src/lib/db/migrationRunner.ts": 1094,
+    "src/lib/db/migrationRunner.ts": 1101,
     "src/lib/db/models.ts": 1097,
     "src/lib/db/providers.ts": 1034,
     "src/lib/memory/retrieval.ts": 1073,
@@ -401,8 +411,8 @@
     "src/shared/components/RequestLoggerV2.tsx": 1629,
     "src/shared/components/analytics/charts.tsx": 1035,
     "src/shared/services/cliRuntime.ts": 1122,
-    "src/sse/handlers/chat.ts": 1904,
-    "src/sse/services/auth.ts": 2520,
+    "src/sse/handlers/chat.ts": 1918,
+    "src/sse/services/auth.ts": 2508,
     "tests/unit/account-fallback-service.test.ts": 1572,
     "tests/unit/provider-validation-specialty.test.ts": 2985,
     "open-sse/executors/hyperagent.ts": 1026,
@@ -421,9 +431,6 @@
   "_rebaseline_2026_07_28_8863_firefly_detail_level": "PR #8863 (fix/adobe-firefly-gpt-detail-level-max) own growth: adobeFireflyClient.ts 2317->2322 (+5 = gpt-image detailLevel defaulting to maximal at the existing payload-build site). Covered by tests/unit/adobe-firefly.test.ts.",
   "_rebaseline_2026_07_29_8281_home_quickstart_prefetch": "Release v3.8.49 base-red fix (no PR — captain sweep): src/app/(dashboard)/dashboard/HomePageClient.tsx 1377->1381 (+4). #8292 added prefetch={false} to the sidebar but left /home's five quick-start Links prefetching, so first paint still fired 12 speculative RSC requests — caught by navigation.spec.ts only after the e2e helper bug (APP_ROUTE_PATTERN missing /home) was repaired in the same cycle. Growth is the five prefetch attributes; it was offset first by extracting the repeated className literals (INLINE_LINK x4, DOCS_LINK x1), which collapsed five wrapped <Link> blocks back to one line each — a naive fix measured 1391. Guard: tests/unit/sidebar-prefetch-policy-8281.test.ts.",
   "_rebaseline_2026_08_02_v3850_agentrouter_responses": "Release v3.8.50 AgentRouter/Codex compatibility reconciliation. open-sse/executors/base.ts 1562->1578: #9190 wires AgentRouter's selected Claude/OpenAI/Responses protocol through the existing executor URL, auth, identity-header and fingerprint chokepoints; the reusable alternate resolver remains outside base.ts. open-sse/utils/stream.ts 2887->2889: #9213 evaluates Responses ID and usage normalization independently so response.completed always receives finite usage.total_tokens instead of short-circuiting after an ID rewrite. tests/unit/chatcore-translation-paths.test.ts 2769->2776: #9191 updates the existing Claude-Code bridge assertions for the dynamic AgentRouter wire image. PR #9224 offsets its own chatCore growth by extracting the AgentRouter protocol decisions into chatCore/agentRouterProtocol.ts, leaving chatCore below its frozen ceiling. Covered by agentrouter executor/chatCore protocol tests, chatcore translation-path tests, and responses-commentary-passthrough tests.",
-  "_rebaseline_2026_08_08_v3850_base_drift_batch_9757": "Base drift on release/v3.8.50, not own growth: the 08-06..08-08 merge batches grew 12 already-frozen (or newly-landed) files without carrying their rebaselines — the dedicated rebaseline PR #9616 was closed as 'superseded' but its file-size entries never actually reached the base, and later merges (#8894 combos page, #9539 EditConnectionModal, #8895 models route, #9294/#9293 catalog, #9541 db/core, #8970 tokenHealthCheck, #8925 mcp schemas+server, #8890 accountFallback, #9467 chat.ts, #8931 openai-to-kiro, ProxyRegistryManager) kept growing them. All 12 values re-measured on THIS branch's tree (= pure tip + this PR's 1-line chat.ts fix, which adds zero lines). This PR's own source changes (chat.ts identifier restore, stream.ts format carve-out) do not grow any frozen file past these values.",
-  "_rebaseline_2026_08_08_migration_135_collision": "fix(db): resolve migration version 135 numbering collision — #9449's 135_connection_runtime_state.sql and #8908's 135_migrate_model_capability_max_token.sql both claimed version 135 (#9449 branched before #8908 merged and never got renumbered before landing on release/v3.8.50), which threw 'Migration version collision detected' the moment ANY code touched the database — a fresh install/deploy from this tip cannot even boot. Renumbered the later-landing file to 140 (next free slot) and added the matching isSchemaAlreadyApplied('140') retroactive guard, matching the established pattern already used for the prior 135/136 -> 137/138 renumber in the same file. Own growth: src/lib/db/migrationRunner.ts 1084->1094 (+10, the new case block) — irreducible, matches the existing per-case guard pattern exactly. Covered by tests/unit/migration-135-numbering-collision.test.ts (2/2), confirmed failing (reproducing the exact live crash) against the pre-fix colliding filenames, passing after.",
-  "_rebaseline_2026_08_02_9259_rolling_rpm": "PR #9259 (issue #8733) own growth: open-sse/services/rateLimitManager.ts baseline 1060->1167 (+107; final source 1153). The existing withRateLimit chokepoint now composes process-local rolling RPM leases with Bottleneck admission, releases pre-dispatch leases on queue timeout/abort/connection disable, preserves caller abort reasons, and wires 429/header state into the extracted rollingRpmGate.ts. The remaining growth is irreducible lifecycle wiring at the dispatch boundary plus the real watchdog test hooks needed to verify queued-wedge recovery; moving it further would obscure lease ownership and Bottleneck cleanup. Covered by the focused rate-limit manager/sliding-window suite (33/33); distributed multi-instance coordination remains explicitly out of scope.",
   "_rebaseline_2026_07_25_dario_upstream_proxy_selector": "PR #8523 (Dario embedded service): upstream-proxy mode selector replaces the binary CLIProxyAPI toggle with Native/CLIProxyAPI/Dario/Fallback + a fallback-backend picker. ProviderDetailPageClient.tsx 798->804 (+6, new hook fields threaded through to ConnectionsListPanel), ConnectionRow.tsx 942->958 (+16, the mode <select> + conditional fallback-backend <select> replacing a single pill button), useProviderConnections.ts 954->986 (+32, upstreamProxyMode/upstreamProxyFallbackBackend state + handleSetUpstreamProxyMode, handleToggleCliproxyapiMode kept as a thin backward-compat wrapper for the existing hook-shape test). All additive UI/state for the new modes — no unrelated refactor.",
   "_rebaseline_2026_08_02_9242_token_health_transient": "PR #9242 (fix/refresh-circuit-transient): src/lib/tokenHealthCheck.ts 1021 (new file, above cap 1000). The file consolidates token-refresh health checking logic that was previously scattered across auth.ts and tokenRefresh.ts. Cohesive single-responsibility module for refresh circuit state management; not extractable without splitting the refresh state machine. Covered by tests/unit/tokenHealthCheck-transient.test.ts.",
   "_rebaseline_2026_07_28_8870_firefly_ref_cap_timeout": "PR #8870 (fix/adobe-firefly-gpt-ref-cap-timeout) own growth: adobeFireflyClient.ts 2322->2385 (+63 = gpt-image subject-ref hard cap at 2 + adaptive poll timeout budget (base 300s + 60s/ref, max 600s) + defensive .slice on referenceBlobs for gpt/nano/generic families). Fixes live 504s on multi-screenshot listing jobs (Featured Promo / Box Art) where 3–4+ subject refs stall colligo until the old 180s poll budget expires. Helpers adobeFireflyMaxImageRefs/adobeFireflyImageTimeoutMs live next to the existing payload/poll chokepoint (not extractable without splitting the wire recipe mid-PR). Covered by tests/unit/adobe-firefly.test.ts (ref-cap + timeout cases). Structural shrink tracked in #3501.",
@@ -435,6 +442,7 @@
   "_rebaseline_2026_08_06b_v3850_sweepreds_drift": "Segunda reconciliacao de 2026-08-06 (/sweep-reds sobre o tip puro 2ddbbc61a6): 3 arquivos voltaram a passar do frozen apos os merges do mesmo dia, com atribuicao 1:1 por commit. (1) src/app/(dashboard)/dashboard/providers/page.tsx 1928->1944 e (2) open-sse/executors/base.ts 1635->1640, ambos do #9515 (feat(radar): flag-gated signed free-model catalog overlay, commit e7f6b1d130) — o overlay do Radar entra por wiring nos chokepoints ja existentes (a resolucao/verificacao do catalogo assinado mora fora destes dois arquivos); +16 e +5 linhas liquidas nao sao extraiveis sem inventar um leaf por callsite. (3) open-sse/services/accountFallback.ts 1966->1972 do #8704 (commit c4527f97bd), +6 linhas de dados em CREDITS_EXHAUSTED_SIGNALS ('has been exhausted', fixes #8631). src/sse/handlers/chat.ts 1880>1877 tambem estava violando e NAO entra aqui de proposito: e drenado por encolhimento na PR #9598, sem rebaseline. Crescimento proprio DESTA PR: src/lib/db/migrationRunner.ts 1077->1084 (+7) — o guard retroativo em isSchemaAlreadyApplied para os arquivos renumerados 137/138, exigido pela propria mensagem de erro de colisao do runner (ambas as migracoes sao ALTER TABLE ADD COLUMN puro, nao idempotente). Dois `case` + dois `return hasColumn(...)` + 3 linhas de comentario dentro do switch existente; nao extraivel.",
   "_rebaseline_2026_08_06c_v3850_sweepreds_pr2": "Segunda PR do /sweep-reds (fix/release-v3.8.50-basereds-0806b): tests/unit/provider-models-route.test.ts 1784->1787 (medido pelo gate, que conta split(\"\\n\").length) (+2 apos compressao de comentarios) — alinhamento de contrato forcado por dois merges do dia: #9106 tornou gemini-3.1-pro-high user-callable (a entry do alias entra na lista esperada do teste de discovery-retry, +1 linha de dado + 1 de comentario) e ff012ff420 adicionou onboardUser como bootstrap hop (exclusao no mock, ja comprimida a 1 linha). Nao ha o que encolher sem apagar o comentario que explica o porque.",
   "_rebaseline_2026_08_07_v3850_sweepreds_pr2_toolnamemap": "tests/unit/translator-openai-to-gemini.test.ts 1616->1619 (+3). O frozen estava EXATAMENTE no tamanho da base, entao qualquer linha nova viola. #9568 (c9a3361e5a) fez buildChangedToolNameMap emitir entradas IDENTIDADE (o Gemini minusculiza nomes de tool nas respostas, entao o tradutor de resposta precisa da chave para mapear de volta), o que passou a incluir `_toolNameMap` no envelope Antigravity de qualquer request com tools. As 3 linhas sao: a chave nova na lista esperada de Object.keys, 1 comentario explicando POR QUE ela aparece (sem ele o proximo leitor tenta remove-la de novo) e 1 assert do CONTEUDO do map — presenca de chave sozinha nao provaria a entrada identidade, que e justamente o comportamento novo. Nao ha o que extrair: e alinhamento de contrato dentro de um teste existente.",
+  "_rebaseline_2026_08_08_9634_migration_139_guard": "PR #9634 (fix/release-v3850-basereds) own growth, re-measured on e0ce95c59 after rebase: src/lib/db/migrationRunner.ts 1094->1096 (+2, the isSchemaAlreadyApplied case-139 retroactive guard for the renumbered ccr migration). Irreducible, matches the per-case guard pattern exactly. Covered by tests/unit/migration-135-numbering-collision.test.ts.",
   "_rebaseline_2026_06_22_4644_deepseek_web_tools": "PR #4644 (BugsBag/robust deepseek-web tool-call parsing): open-sse/executors/deepseek-web.ts 1117->1125 (+8). The new agentic tool-call path emits surrounding text + reasoning before tool_calls and swaps to the dedicated deepseekWebTools.ts parser; the +8 lines are cohesive wiring at the existing transformSSE chokepoint (the parser itself lives in the new deepseekWebTools.ts file, already under cap). The PR's own fast-gate (PR->release) does not run check:file-size, so this surfaced only at release reconcile. Covered by tests/unit/deepseek-web-tools-variants.test.ts + deepseek-web-tools-execute.test.ts.",
   "_rebaseline_2026_06_23_4712_deepseek_web_tool_results": "PR for #4712 (deepseek-web drops role:tool): open-sse/executors/deepseek-web.ts 1125->1148 (+23). messagesToPrompt() now folds role:\\\"tool\\\" results into the single-prompt transcript (recovering the tool name from the preceding assistant tool_calls by tool_call_id) instead of silently dropping them; the lines are cohesive wiring inside the existing function. Covered by tests/unit/deepseek-web-tool-result-prompt-4712.test.ts.",
   "_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, <cap) and the async orderer orderTargetsByHeadroom is appended to the existing open-sse/services/combo/quotaStrategies.ts (<cap) next to its sibling reset-aware/reset-window orderers (reuses their connection-expansion machinery). headroom = 1 - max(util_5h, util_7d) from getSaturation (src/lib/quota/saturationSignals.ts), prefers the connection with the most free capacity. Only the dispatch wiring is irreducible at the existing combo strategy chokepoint (mirrors the reset-aware/reset-window/context-optimized branches); not extractable without hiding the call site. fill-first stays default; all existing strategies untouched. Covered by tests/unit/combo-headroom-ranking.test.ts (pure helper) + tests/unit/combo-headroom-strategy.test.ts (orderer, saturation injected). Structural shrink of combo.ts tracked in #3501.",
@@ -515,7 +523,7 @@
   "open-sse/services/rateLimitManager.ts": "1167",
   "open-sse/translator/response/openai-responses.ts": "1204",
   "open-sse/utils/cursorAgentProtobuf.ts": "1505",
-  "open-sse/utils/stream.ts": "2889",
+  "open-sse/utils/stream.ts": 2915,
   "src/app/(dashboard)/dashboard/HomePageClient.tsx": "1388",
   "src/app/(dashboard)/dashboard/analytics/ComboHealthTab.tsx": "1031",
   "src/app/(dashboard)/dashboard/api-manager/ApiManagerPageClient.tsx": "3117",
@@ -541,7 +549,7 @@
   "src/lib/tokenHealthCheck.ts": "1053",
   "src/lib/db/apiKeys.ts": "1529",
   "src/lib/db/core.ts": "1639",
-  "src/lib/db/migrationRunner.ts": "1094",
+  "src/lib/db/migrationRunner.ts": "1096",
   "src/lib/db/models.ts": "1097",
   "src/lib/db/providers.ts": "1034",
   "src/lib/memory/retrieval.ts": "1073",
@@ -560,5 +568,11 @@
   "open-sse/executors/kiro.ts": "1069",
   "open-sse/translator/request/openai-to-kiro.ts": "1057",
   "open-sse/utils/sseHeartbeat.ts": "142",
-  "_rebaseline_2026_08_04_9305_sse_comments": "#9305 fix: broadened sseCommentsEnabled()"
+  "_rebaseline_2026_08_04_9305_sse_comments": "#9305 fix: broadened sseCommentsEnabled()",
+  "_rebaseline_2026_08_09_v3850_release_close": "Release v3.8.50 close reconciliation on e0ce95c592: src/sse/handlers/chat.ts 1904->1918 is the irreducible request-pipeline wiring from #9759 that invokes the Modality Bridge guardrail without moving its implementation into the handler; covered by the 17 Vision Bridge canaries plus the PR-1 focused suite. open-sse/translator/response/openai-responses.ts 1204->1215 is #9168's Responses tool-call argument delta buffering/normalization at the existing translator state-machine chokepoint; covered by its dedicated translator regression tests. Both values are measured by check:file-size (split-newline semantics), and the gate remains frozen at the new exact sizes.",
+  "_rebaseline_2026_08_08_toolcall_message_index_collision": "fix(responses-api): tool call after a text message collided on the same output_index. own growth: open-sse/translator/response/openai-responses.ts 1204->1224 (+20, extracted toolCallOutputIndexBase() shared helper so emitToolCall/closeToolCall can no longer compute a tool call's output_index independently and collide with a text message emitted in the same turn). Live incident (2026-08-08, OpenClaw agent): a client that tracks response items by output_index saw the tool call's added/delta/done events land on an index it had already marked complete (the just-closed text message), and silently dropped them — the agent spoke its preamble and never executed the tool call, even though OmniRoute's own recorded responseBody had a complete, valid tool_calls entry. Covered by the new regression test in tests/unit/translator-resp-openai-responses.test.ts reproducing the exact live scenario.",
+  "_rebaseline_2026_08_03_9255_adobe_firefly_durable_sessions": "PR #9255 own cohesive growth: open-sse/services/adobeFireflyClient.ts 2322->2894 adds authenticated-vs-guest IMS classification, browser-risk ARP validation/rebuild, bounded 408 retry/recovery, sticky accepted-session handling, and matching image/video submit recovery at the existing Adobe upstream client chokepoints. This client was already explicitly frozen as a single self-contained upstream integration by #8006/#8510; splitting only the retry/auth helpers now would scatter one request state machine while structural shrink remains tracked in #3501. tests/unit/adobe-firefly.test.ts 871->1136 adds direct regression coverage for guest-token rejection, cookie/ARP rebuilding, 408 retries, sticky accepted ARP reuse, forced auth recovery, and cookie-to-IMS exchange. The obsolete 1179-line managed-Chrome fallback module was deleted rather than rebaselined after the packaged-safe pure-CDP path became authoritative. Focused Adobe suite: 61/61.",
+  "_rebaseline_2026_08_07_9653_disconnect_grace_period": "Extracted fix(sse): grace period before finalizing a client disconnect as 499 (#9653) — a client that closes its connection right after reading a fully-completed SSE stream can race OmniRoute's own completion bookkeeping, getting persisted as a false 499/0-tokens even though it delivered the full response (live-confirmed: a real disconnect at 18236ms was corrected to 200/82814+1292 tokens). Own growth: open-sse/handlers/chatCore.ts 5030->5039 (+9, wiring createClientDisconnectGraceHandler at the existing onClientDisconnectFinalize call site) — irreducible call-site wiring, the actual grace-period logic lives in the new leaf createClientDisconnectGraceHandler (open-sse/utils/streamFailureFinalization.ts, not frozen). Re-measured to 5042 after rebasing onto a newer release/v3.8.50 tip: the file carries an unrelated +3 base drift from already-merged upstream commits between this PR's original branch point and the rebase target, not covered by this entry. Covered by tests/unit/stream-disconnect-grace-period-9653.test.ts (4/4, fake-timer driven). Other file-size gate violations present on this base tip are pre-existing/unrelated to this change (base-red #9679, re-verify current issue number at merge time).",
+  "_rebaseline_2026_08_04_9268_gemini_schema_empty_choices": "Feature #9268 own growth: open-sse/utils/stream.ts 2889->2915 (+26 = irreducible call-site wiring for the empty-choices interceptor). The translate-mode flush now rejects a stream that completed without forwarding any valuable chunk (all-empty `choices: []`, no content/tool_calls/finish_reason) as a retryable 502 \"empty content\" instead of a clean empty 200 — the missing streaming counterpart of chatCore.ts's non-streaming isEmptyContentResponse. All rejection logic lives in the NEW leaf module open-sse/utils/streamEmptyChoices.ts (<cap, not frozen, unit-tested via tests/unit/stream-empty-choices-interceptor.test.ts); stream.ts only carries the `forwardedValuableChunk` boolean (declared at createSSEStream scope, set in emitTranslatedClientItem where the sole hasValuableContent check passes) plus the one flush-time rejectEmptyChoicesStream() call — the wait/orchestration at the chokepoint, not a movable block (mirrors the comboCooldownRetry.ts precedent). Schema-side twin fix: recursive type:\"object\" injection in open-sse/translator/helpers/geminiHelper.ts (not frozen, +33) for nested schemas with properties but no type (Gemini 400).",
+  "_rebaseline_2026_08_09_5696_capability_filter": "PR #9424 own growth: open-sse/handlers/chatCore.ts 5050->5061 (+11). The Layer A capability gate is irreducible wiring at the existing pre-dispatch chokepoint: feature-flag check, capability derivation, compatibility decision, sanitized 400 response, pending-request cleanup, and warning telemetry. All matching and message logic lives outside the god-file in src/shared/constants/capabilities/capabilityFilter.ts; only orchestration remains here. Covered by tests/unit/capability-filter.test.ts (20 cases, including flag-off and sanitized error behavior). Structural shrink remains tracked separately."
 }

config/quality/quality-baseline.json

@@ -92,17 +92,19 @@
       "_rebaseline_2026_07_13_v3847_release": "39.3 -> 38.0 (-1.3, beyond the 0.5 eps). v3.8.47 cycle drift: the cycle merged ~45 PRs adding API routes (relay repair/free-pool #6909, backpressure #6590, combo context requirements #6907, services/usage endpoints) faster than openapi.yaml documentation; same class as the v3.8.34/v3.8.39 rebaselines. Documented follow-up: raise coverage next cycle via docs/openapi.yaml additions."
     },
     "i18nUiCoverage.pct": {
-      "value": 99,
+      "value": 100,
       "direction": "up",
       "eps": 0.5,
+      "_tighten_2026_08_08_modality_bridge": "99 -> 100. Tighten required by the PR quality gate after the Modality Bridge UI keys were translated across all 42 non-English locales. CI collect-metrics on PR #9782 measured i18nUiCoverage.pct=100 with 0 ESLint warnings and 0 ESLint errors; locale dry-sync and UI coverage also report 100% with no missing keys or placeholders.",
       "_rebaseline_2026_07_04_v3844_release": "77.5 -> 76.8 (-0.7, beyond the 0.5 eps). v3.8.44 cycle drift surfaced only on the release PR (i18n-ui-coverage does NOT run on PR->release fast-gates). The cycle added ~1352 new UI keys to the en.json denominator (Discovery dashboard tab #5939, Bifrost/Mux embedded-service tabs #5817/#6034, proxy batch-ops #5918, fusion defaults #5598, tool-source toggle #5978, quota-row collapse #5977, CodeWhale/Crush CLI cards #5996/#5970, etc.) that the async i18n translation workflow has not yet back-filled (worst locales measure 76.8; __MISSING__ placeholders count as uncovered by design). Same shape and remedy as _rebaseline_2026_06_28_v3839_release. Recover via the i18n workflow next cycle; tighten with --require-tighten once translations land.",
       "_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).",
       "_rebaseline_2026_07_13_v3847_release": "76.8 -> 75.5 (-1.3, beyond the 0.5 eps). v3.8.47 cycle drift: merged UI features added EN strings (relay repair UI #6909, combo builder #6907/#6991, capability override UI #6727) ahead of the 42-locale mirrors; same class as the v3.8.39/v3.8.44 rebaselines.",
       "_rebaseline_2026_07_28_v3849_release": "75.5 -> 99 (+23.5). Aperto EXIGIDO pelo modo --require-tighten do ratchet: a métrica melhorou de verdade no ciclo v3.8.49. A causa é o workflow assíncrono de tradução, que finalmente alcançou o denominador em EN — as rebaselines anteriores (v3.8.39/.44/.47) foram todas afrouxamentos registrando o atraso das traduções, e agora ele foi pago. O coletor SUBTRAI os placeholders (present - placeholder em scripts/quality/collect-metrics.mjs), então os 317 marcadores __MISSING__ que esta release introduziu para o drift de valor já estão descontados dos 99 — o número é honesto, não inflado por placeholder. Medido pelo collect-metrics do CI no run 30404226939."
     },
     "deadExports": {
-      "value": 227,
+      "value": 230,
       "direction": "down",
+      "_rebaseline_2026_08_09_v3850_post_sweep": "227 -> 230. Measured by npm run check:dead-code on the unmodified release/v3.8.50 tip 382449d593 during the mandatory --full-ci pre-flight. The +3 is inherited cycle drift from the authorized merge sweep; this repair adds no production exports. Rebaseline records the actual tip so ci.yml quality-gate can run, while structural cleanup remains separate debt.",
       "_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.",
@@ -177,12 +179,13 @@
       "dedicatedGate": true
     },
     "bundleSize": {
-      "value": 7666,
+      "value": 8045,
       "direction": "down",
       "dedicatedGate": true,
       "_rebaseline_2026_07_07_v3846_release_close": "5601->6534 (+933). v3.8.46 release close: gzip of the 4 bin/*.mjs entrypoints (size-limit + @size-limit/file) grew from this cycle's feature/fix merges pulled transitively into the CLI entrypoints (new providers, combo pipeline strategy #6396, effort/thinking standardization #6241, catalog cache-invalidation #6408). Measured 6534 locally via `check:bundle-size --ratchet` (deterministic gzip, matches CI). Legitimate cycle growth; shrink is separate debt.",
       "_rebaseline_2026_07_19_7808_codeql_alias_resolver_hook": "6534->6762 (+228). PR #7808 (CodeQL js/incomplete-url-substring-sanitization fix): the ESM loader hook source moved out of the inline `HOOK_SOURCE` template literal in bin/aliasResolver.mjs into a real file bin/aliasResolverHook.mjs, loaded via pathToFileURL() instead of a dynamically-built `data:text/javascript,...` URL. The new file is now counted by size-limit as a 5th bin/*.mjs entrypoint. Net +228 = the hook's gzip size (previously hidden inside aliasResolver.mjs because the template literal was compressed away). Security-driven; no shrink opportunity.",
-      "_rebaseline_2026_07_28_v3849_release_preflight": "6762 -> 7666 (+904). Fechamento do ciclo v3.8.49: gzip dos entrypoints bin/*.mjs (size-limit + @size-limit/file) cresceu com o que os merges do ciclo puxam transitivamente para o CLI (novos provedores — 271->290, seletor de protocolo por conexão #8861, catálogos de busca #8814, resiliência). Crescimento legítimo de ciclo, medido localmente com `npm run check:bundle-size` = 7666 (gzip determinístico, bate com o CI). Encolher é dívida separada."
+      "_rebaseline_2026_07_28_v3849_release_preflight": "6762 -> 7666 (+904). Fechamento do ciclo v3.8.49: gzip dos entrypoints bin/*.mjs (size-limit + @size-limit/file) cresceu com o que os merges do ciclo puxam transitivamente para o CLI (novos provedores — 271->290, seletor de protocolo por conexão #8861, catálogos de busca #8814, resiliência). Crescimento legítimo de ciclo, medido localmente com `npm run check:bundle-size` = 7666 (gzip determinístico, bate com o CI). Encolher é dívida separada.",
+      "_rebaseline_2026_08_09_v3850_release_close": "7666 -> 8045 (+379 gzip bytes, +4.9%). Release v3.8.50 close reconciliation measured twice with the real size-limit + @size-limit/file path on tip e0ce95c592. Per-entry measurements remain below their absolute budgets: omniroute.mjs 4380/15000, mcp-server.mjs 1195/5000, nodeRuntimeSupport.mjs 887/8000, reset-password.mjs 1583/6000. The growth accumulated through legitimate CLI/runtime work in this cycle, including global-install ESM alias resolution, Termux cache preparation, and MCP stdio startup hardening; no entrypoint is near its absolute ceiling. The direction:down ratchet stays blocking from this exact measured tip."
     },
     "openapiBreaking": {
       "value": 0,

config/quality/test-discovery-baseline.json

@@ -1,24 +1,10 @@
 {
   "_comment": "Catraca de test-discovery (check-test-discovery.mjs). Cada entrada e um arquivo de teste que NENHUM runner coleta (ele nunca roda) — divida congelada na auditoria 6A.1 (2026-06-09; 195 originais, 135 religados no node runner em 6A.1c). So pode DIMINUIR: religue o teste (ajustando o glob do runner ou movendo o arquivo) e remova a entrada via --update. NAO adicione novos orfaos — corrija o runner.",
-  "_remaining_60": "Categorias: 33 .test.tsx de tests/unit (religaveis via vitest.config root, MAS o experimento 2026-06-09 mostrou 24 arquivos vermelhos — triagem de drift de UI na janela 2026-06-16, junto com os 14 fails do proprio test:vitest:ui atual); 9 open-sse __tests__ + 8 src __tests__ (includes de vitest.config que NENHUM script executa sem filtro); 4 golden-set + 1 benchmarks + 1 live + 1 stress (deliberadamente manuais — decidir runner/gating); 3 integration/services (gated RUN_SERVICES_INT=1, sem runner CI).",
+  "_remaining_13": "13 orfaos restantes: 2 testes de API em settings + 1 snapshot de quota do DB; 4 golden-set + 1 benchmark + 1 teste live + 1 stress (deliberadamente manuais — decidir runner/gating); 3 integration/services (gated RUN_SERVICES_INT=1, sem runner CI).",
   "orphans": [
-    "open-sse/services/__tests__/chatgptTlsClient.test.ts",
-    "open-sse/services/__tests__/claudeTlsClient.test.ts",
-    "open-sse/services/__tests__/grokTlsClient.test.ts",
-    "open-sse/services/__tests__/manifestAdapter.test.ts",
-    "open-sse/services/__tests__/specificityDetector.test.ts",
-    "open-sse/services/__tests__/tierResolver.test.ts",
-    "open-sse/services/__tests__/volumeDetector.test.ts",
-    "open-sse/translator/helpers/__tests__/maxTokensHelper.test.ts",
-    "open-sse/translator/helpers/__tests__/schemaCoercion.test.ts",
     "src/app/api/settings/__tests__/memory.test.ts",
     "src/app/api/settings/__tests__/settings.test.ts",
     "src/lib/db/__tests__/quotaSnapshots.test.ts",
-    "src/lib/memory/__tests__/injection.test.ts",
-    "src/lib/memory/__tests__/qdrant-wiring.test.ts",
-    "src/lib/memory/__tests__/retrieval.test.ts",
-    "src/lib/memory/__tests__/schemas.test.ts",
-    "src/lib/skills/__tests__/integration.test.ts",
     "tests/benchmarks/pipeline-accuracy.test.ts",
     "tests/golden-set/compression-caveman-v2.test.ts",
     "tests/golden-set/compression-quality.test.ts",
@@ -28,36 +14,6 @@
     "tests/integration/services/full-lifecycle.int.test.ts",
     "tests/integration/services/route-guard-services.int.test.ts",
     "tests/live/deepseek-web-live.test.ts",
-    "tests/theoldllm-stress.test.ts",
-    "tests/unit/AutoComboCatalog.test.tsx",
-    "tests/unit/SkillsConceptCard.test.tsx",
-    "tests/unit/agent-skills-page.test.tsx",
-    "tests/unit/dashboard/batch/components/BatchDetailModal.test.tsx",
-    "tests/unit/dashboard/batch/components/ExpirationBadge.test.tsx",
-    "tests/unit/dashboard/batch/components/NewBatchWizard.test.tsx",
-    "tests/unit/dashboard/batch/components/ProgressBarBicolor.test.tsx",
-    "tests/unit/dashboard/batch/components/UploadFileModal.test.tsx",
-    "tests/unit/dashboard/batch/components/useBatchActions.test.tsx",
-    "tests/unit/dashboard/batch/concept-cards.test.tsx",
-    "tests/unit/dashboard/batch/list-regression.test.tsx",
-    "tests/unit/dashboard/batch/sanitization.test.tsx",
-    "tests/unit/omni-skills-page.test.tsx",
-    "tests/unit/shared-clipboard.test.tsx",
-    "tests/unit/shared/components/AutoRoutingBanner.test.tsx",
-    "tests/unit/shared/components/KiroAuthModal.test.tsx",
-    "tests/unit/shared/components/ProxyConfigModal.test.tsx",
-    "tests/unit/translator-friendly-advanced-section.test.tsx",
-    "tests/unit/translator-friendly-compression.test.tsx",
-    "tests/unit/translator-friendly-concept-card.test.tsx",
-    "tests/unit/translator-friendly-integration.test.tsx",
-    "tests/unit/translator-friendly-monitor-tab.test.tsx",
-    "tests/unit/translator-friendly-page-client.test.tsx",
-    "tests/unit/translator-friendly-pipeline-view.test.tsx",
-    "tests/unit/translator-friendly-raw-json-panel.test.tsx",
-    "tests/unit/translator-friendly-result-narrated.test.tsx",
-    "tests/unit/translator-friendly-simple-controls.test.tsx",
-    "tests/unit/translator-friendly-stream-transformer.test.tsx",
-    "tests/unit/translator-friendly-test-bench.test.tsx",
-    "tests/unit/translator-friendly-translate-tab.test.tsx"
+    "tests/theoldllm-stress.test.ts"
   ]
 }

docs/architecture/QUALITY_GATES.md

@@ -186,10 +186,10 @@ Runs on pull requests only.
 
 Runs after `build`. Blocks merge on failure.
 
-| Suite            | Validates                                               | Blocking                                                                   |
-| ---------------- | ------------------------------------------------------- | -------------------------------------------------------------------------- |
-| `test:vitest`    | MCP server (94 tools), autoCombo, cache — vitest runner | Yes                                                                        |
-| `test:vitest:ui` | UI component tests — vitest runner                      | **Advisory** (`continue-on-error: true`) — failing until Fase 6A UI triage |
+| Suite            | Validates                                               | Blocking                                                                                                      |
+| ---------------- | ------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------- |
+| `test:vitest`    | MCP server (94 tools), autoCombo, cache — vitest runner | Yes                                                                                                           |
+| `test:vitest:ui` | UI component tests — vitest runner                      | **Blocking** — pre-existing failures are explicitly excluded in `vitest.config.ts`; new failures fail the job |
 
 ### Nightly workflows (scheduled, advisory)
 
@@ -401,7 +401,7 @@ several "obvious" merges turned out to hide debt and are **not** clean drop-ins.
 
 - `check:openapi-security-tiers` (advisory) — ❌ **NOT cleanly flippable.** It exits 0 but warns that several `traffic-inspector` routes under `LOCAL_ONLY_API_PREFIXES` lack the `x-loopback-only: true` annotation. Enforcing it requires adding those annotations to `openapi.yaml` first.
 - `typecheck:noimplicit:core` (advisory) — largely subsumed by the blocking `check:type-coverage` ratchet. Flip to a ratchet or drop the redundant second `tsc` pass.
-- `test:vitest:ui` (advisory, 14 parked fails) — fix-and-block or delete; don't leave rotting.
+- `test:vitest:ui` (now **blocking**) — pre-existing failures are explicitly excluded in `vitest.config.ts` with `// #8618` tracking comments; new failures fail the job.
 - `check:secrets` (gitleaks, blocking ratchet frozen at 3 documented false-positives) — allowlist the 3 to reach 0, or demote to advisory. Overlaps GitHub native secret-scanning + `check:public-creds`.
 - `check:pr-evidence` (blocking, greps PR-body prose) — high false-positive risk; weakens Hard Rule #18 enforcement if dropped, so this is a genuine policy call.
 - `semgrep` (advisory standalone) — overlaps CodeQL for the OWASP families; wire its baseline to a ratchet or drop.

docs/getting-started/WEB-COOKIE-GUIDE.md

@@ -18,7 +18,7 @@ Unlike API-key providers, Web Cookie providers authenticate using the credential
 
 Many authentication issues are caused by copying cookies from the wrong place.
 
-##  Do NOT copy from Cookie Storage
+## Do NOT copy from Cookie Storage
 
 Most browsers expose stored cookies through:
 
@@ -36,7 +36,7 @@ Although these cookies look correct, they may be:
 
 Using these values may cause authentication failures even if they appear valid.
 
-##  Copy from a Live Request
+## Copy from a Live Request
 
 Instead, use the cookies from a successful request:
 
@@ -80,14 +80,14 @@ The exact credentials required depend on the provider.
 
 Different websites store authentication differently. Some require only cookies, while others may require additional headers or tokens.
 
-| Provider | Credential Format | Provider Guide |
-|----------|-------------------|----------------|
-| Claude Web | Full Cookie request header | `docs/providers/CLAUDE_WEB.md` |
-| ChatGPT Web | _(verify)_ | |
-| Gemini Web | _(verify)_ | |
-| Copilot Web | _(verify)_ | |
-| Grok Web | _(verify)_ | |
-| ... | ... | ... |
+| Provider    | Credential Format                                              | Provider Guide                  |
+| ----------- | -------------------------------------------------------------- | ------------------------------- |
+| Claude Web  | Full Cookie request header                                     | `docs/providers/CLAUDE_WEB.md`  |
+| ChatGPT Web | Full Cookie header or `__Secure-next-auth.session-token` value | `docs/providers/CHATGPT_WEB.md` |
+| Gemini Web  | _(verify)_                                                     |                                 |
+| Copilot Web | _(verify)_                                                     |                                 |
+| Grok Web    | _(verify)_                                                     |                                 |
+| ...         | ...                                                            | ...                             |
 
 > Update this table as new Web Cookie providers are added or existing providers change their authentication requirements.
 

docs/guides/DOCKER_GUIDE.md

@@ -147,11 +147,11 @@ The prod stack runs in parallel with the dev compose (different container names,
 
 The repository ships a multi-stage Dockerfile (`Dockerfile`). Three stages are exposed; pick the right `target` for your use case.
 
-| Stage         | Base image                 | Purpose                                                                                                                                                            |
-| ------------- | -------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
-| `builder`     | `node:24.15.0-trixie-slim` | Installs deps (`npm ci --legacy-peer-deps`) and runs `npm run build -- --webpack`                                                                                  |
-| `runner-base` | `node:24.15.0-trixie-slim` | Production runtime with the Next.js standalone output. **No provider CLIs bundled.**                                                                               |
-| `runner-cli`  | `runner-base`              | Adds `git`, `docker.io`, `docker-compose` and global CLIs: `@openai/codex`, `@anthropic-ai/claude-code`, `droid`, `openclaw`. **Pick this for agentic workflows.** |
+| Stage         | Base image            | Purpose                                                                                                                                                            |
+| ------------- | --------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
+| `builder`     | `node:26-trixie-slim` | Installs deps (`npm ci --legacy-peer-deps`) and runs `npm run build` (Turbopack by default — see Build-time resources below)                                       |
+| `runner-base` | `node:26-trixie-slim` | Production runtime with the Next.js standalone output. **No provider CLIs bundled.**                                                                               |
+| `runner-cli`  | `runner-base`         | Adds `git`, `docker.io`, `docker-compose` and global CLIs: `@openai/codex`, `@anthropic-ai/claude-code`, `droid`, `openclaw`. **Pick this for agentic workflows.** |
 
 Build a specific target manually:
 
@@ -160,14 +160,50 @@ docker build --target runner-base -t omniroute:base .
 docker build --target runner-cli  -t omniroute:cli  .
 ```
 
-Defaults exported by `runner-base`: `PORT=20128`, `HOSTNAME=0.0.0.0`, `NODE_OPTIONS=--max-old-space-size=512`, `DATA_DIR=/app/data`, `OMNIROUTE_MIGRATIONS_DIR=/app/migrations`.
+### Build-time resources
+
+Two build args control what the `builder` stage costs. They are build-time only —
+`OMNIROUTE_MEMORY_MB` (below) is a separate, runtime knob.
+
+| Build arg                   | Default | Effect                                                                 |
+| --------------------------- | ------- | ---------------------------------------------------------------------- |
+| `OMNIROUTE_USE_TURBOPACK`   | `1`     | `0` builds with webpack instead. Lower peak memory, slower.            |
+| `OMNIROUTE_BUILD_MEMORY_MB` | `4096`  | V8 heap ceiling (`--max-old-space-size`) for the spawned `next build`. |
+
+Turbopack compiles in native Rust memory that lives **outside** the V8 heap, so
+`OMNIROUTE_BUILD_MEMORY_MB` does not bound it. On a host with a memory ceiling the
+build is then SIGKILLed by the OOM killer with no error text at all — it simply
+stops mid-`Creating an optimized production build`, which reads like a hang rather
+than an out-of-memory. If the build host is constrained, switch bundlers:
+
+```bash
+docker build --target runner-base \
+  --build-arg OMNIROUTE_USE_TURBOPACK=0 \
+  -t omniroute:base .
+```
+
+`webpackBuildWorker` is enabled, so `next build` runs a parent **and** a worker
+process and each honours `OMNIROUTE_BUILD_MEMORY_MB` separately. Size the container
+ceiling above roughly twice that value, not once.
+
+Measured on this tree (`--target runner-base`, `OMNIROUTE_BUILD_MEMORY_MB=6144`):
+
+| Bundler   | Container ceiling | Result                        |
+| --------- | ----------------- | ----------------------------- |
+| Turbopack | 8 GiB / 16 GiB    | OOM-killed at both, silently  |
+| webpack   | 8 GiB             | build worker SIGKILLed        |
+| webpack   | 12 GiB            | succeeded, peaked at 11.1 GiB |
+
+### Runtime defaults
+
+Defaults exported by `runner-base`: `PORT=20128`, `HOSTNAME=0.0.0.0`, `OMNIROUTE_MEMORY_MB=1024`, `NODE_OPTIONS=--max-old-space-size=1024`, `DATA_DIR=/app/data`, `OMNIROUTE_MIGRATIONS_DIR=/app/migrations`.
 
 Memory behavior in Docker:
 
-- `NODE_OPTIONS=--max-old-space-size=512` is baked into the image as a fallback.
+- The image sets `OMNIROUTE_MEMORY_MB=1024` and derives `NODE_OPTIONS=--max-old-space-size=1024` from it.
 - The actual server process is started by the standalone launcher, which reads `OMNIROUTE_MEMORY_MB` and appends `--max-old-space-size=<OMNIROUTE_MEMORY_MB>`.
 - Node uses the last repeated `--max-old-space-size` value, so setting `OMNIROUTE_MEMORY_MB` controls the effective Docker heap limit.
-- If `OMNIROUTE_MEMORY_MB` is unset, the launcher uses `512`.
+- Because the image always sets it, the launcher's own RAM-calibrated fallback never applies under Docker. Raise it explicitly (`-e OMNIROUTE_MEMORY_MB=2048`) on a host with headroom.
 
 ## Critical Environment Variables
 
@@ -180,7 +216,7 @@ Beyond the defaults documented in [ENVIRONMENT.md](../reference/ENVIRONMENT.md),
 | `REDIS_PORT`                  | Host-side port for the bundled Redis container                                                      | `6379`                   |
 | `REDIS_BIND_HOST`             | Host interface the bundled Redis port is published on (loopback unless you add AUTH)                | `127.0.0.1`              |
 | `AUTO_UPDATE_HOST_REPO_DIR`   | Host path mounted into `cli` profile at `/workspace/omniroute` for self-update workflows            | `.` (current directory)  |
-| `OMNIROUTE_MEMORY_MB`         | Runtime Node heap ceiling for the Docker standalone server; overrides the image fallback above      | `512`                    |
+| `OMNIROUTE_MEMORY_MB`         | Runtime Node heap ceiling for the Docker standalone server; overrides the image default above       | `1024`                   |
 | `DASHBOARD_PORT` / `API_PORT` | Override exposed ports for dashboard (20128) and API (20129)                                        | `20128` / `20129`        |
 | `OMNIROUTE_BASE_PATH`         | URL subpath when the app is published behind a reverse proxy (e.g. `/omniroute`)                    | _(empty = root)_         |
 | `NEXT_PUBLIC_BASE_URL`        | Public browser origin including the subpath (e.g. `https://host/omniroute`)                         | unset                    |

docs/guides/TROUBLESHOOTING.md

@@ -523,6 +523,12 @@ exhausts its bounds of `10,000` visited nodes or depth `12`.
 Each process uses a process-local guard to reserve limited heavyweight capacity before retaining
 and parsing a large request body. A heavyweight lease remains held for the lifetime of an SSE
 response.
+
+When capacity is busy, a heavyweight request first waits up to
+`OMNIROUTE_CHAT_ADMISSION_QUEUE_MS` (default `5000`, `0` disables the wait) for a slot to free up
+before answering the retryable `503`. The bounded wait exists so agent-style clients
+(OpenCode, Claude Code, Cursor) that fan out heavy sub-requests concurrently serialize the burst
+instead of burning their whole retry budget on immediate rejections and dying mid-task.
 Current heavyweight lease occupancy is not surfaced in the dashboard.
 Settings → Resilience → Request Queue → Concurrent Requests does not control this; that setting
 governs a separate provider request-queue mechanism.
@@ -530,13 +536,18 @@ governs a separate provider request-queue mechanism.
 **Fix:**
 
 1. Retry first. Clients should honor `Retry-After` and use backoff rather than immediately
-   repeating the request.
+   repeating the request. Note that with the default `OMNIROUTE_CHAT_ADMISSION_QUEUE_MS=5000`
+   a heavy request already waited up to 5 seconds before the `503`, so a client retry loop should
+   back off beyond that instead of hammering.
 2. If normal deployment traffic repeatedly exhausts the guard, you can cautiously raise
    `OMNIROUTE_CHAT_MAX_HEAVY_IN_FLIGHT` from its default of `1`. Increase it one step at a time,
    restart OmniRoute after each change, and observe memory headroom under representative load.
    Every additional heavyweight request can increase concurrent V8 heap use and container or
    host OOM risk. No value is safe for every deployment; validate the setting against your own
    traffic and memory limits rather than assuming that `2` is universally safe.
+3. Prefer widening the wait (`OMNIROUTE_CHAT_ADMISSION_QUEUE_MS`) over raising the in-flight
+   limit when bursts are short: waiting costs latency, while an extra concurrent heavyweight
+   request costs heap residency for the whole request lifetime.
 
 See the [environment-variable reference](../reference/ENVIRONMENT.md#4-security--authentication)
 for the authoritative admission settings. Loosening the heavyweight classification thresholds

docs/openapi.yaml

@@ -5225,6 +5225,102 @@ paths:
         "200":
           description: Sync initialized
 
+  # ─── Background Jobs (local-only administration) ───────────────
+
+  /api/jobs:
+    get:
+      tags: [System]
+      summary: List registered background jobs
+      description: Local-only runtime administration. Returns each registered job and its latest run.
+      x-internal: true
+      responses:
+        "200":
+          description: Registered jobs
+        "500":
+          description: Failed to list jobs
+
+  /api/jobs/{id}/enable:
+    post:
+      tags: [System]
+      summary: Enable a background job
+      description: Local-only runtime administration. Enables the job and restarts its timer.
+      x-internal: true
+      parameters:
+        - name: id
+          in: path
+          required: true
+          schema:
+            type: string
+      responses:
+        "200":
+          description: Job enabled
+        "404":
+          description: Job not found
+        "500":
+          description: Failed to enable job
+
+  /api/jobs/{id}/disable:
+    post:
+      tags: [System]
+      summary: Disable a background job
+      description: Local-only runtime administration. Disables the job and stops its timer.
+      x-internal: true
+      parameters:
+        - name: id
+          in: path
+          required: true
+          schema:
+            type: string
+      responses:
+        "200":
+          description: Job disabled
+        "404":
+          description: Job not found
+        "500":
+          description: Failed to disable job
+
+  /api/jobs/{id}/run-now:
+    post:
+      tags: [System]
+      summary: Trigger a background job
+      description: >-
+        Local-only runtime administration. Starts the job, or waits for an in-flight
+        run before queueing the next one, subject to OMNIROUTE_RUNNOW_TIMEOUT_MS.
+      x-internal: true
+      parameters:
+        - name: id
+          in: path
+          required: true
+          schema:
+            type: string
+      responses:
+        "200":
+          description: Job trigger accepted
+        "404":
+          description: Job not found
+        "500":
+          description: Failed to trigger job
+
+  /api/jobs/{id}/runs:
+    get:
+      tags: [System]
+      summary: Read background-job run history
+      description: Local-only runtime administration. Returns newest-first run history for one job.
+      x-internal: true
+      parameters:
+        - name: id
+          in: path
+          required: true
+          schema:
+            type: string
+      responses:
+        "200":
+          description: Job run history
+        "404":
+          description: Job not found
+        "500":
+          description: Failed to load job runs
+
   # ─── Resilience & Monitoring ────────────────────────────────────
 
   /api/resilience:
@@ -5247,6 +5343,70 @@ paths:
         "200":
           description: Updated resilience configuration
 
+  /api/resilience/connections:
+    get:
+      tags: [System]
+      summary: Inspect connection resilience state
+      description: >-
+        Local-only operational view of per-connection cooldowns, provider circuit
+        breakers, model lockouts, and recent breaker transitions. Credential columns
+        are excluded by an explicit database whitelist.
+      x-internal: true
+      parameters:
+        - name: windowMs
+          in: query
+          schema:
+            type: integer
+            minimum: 0
+            maximum: 86400000
+            default: 3600000
+        - name: provider
+          in: query
+          schema:
+            type: string
+            minLength: 1
+            maxLength: 64
+      responses:
+        "200":
+          description: Connection, breaker, lockout, window, and degradation metadata
+        "400":
+          description: Invalid query parameters
+        "500":
+          description: Failed to collect resilience state
+
+  /api/telegram/update:
+    post:
+      tags: [System]
+      summary: Receive Telegram updates or Mini App messages
+      description: >-
+        Public Telegram integration endpoint. Bot updates are acknowledged after
+        reply dispatch is queued. Mini App requests must include Telegram-signed
+        initData, which is verified with TELEGRAM_BOT_TOKEN before chat proxying.
+      security: []
+      requestBody:
+        required: true
+        content:
+          application/json:
+            schema:
+              type: object
+              additionalProperties: true
+              properties:
+                initData:
+                  type: string
+                message:
+                  type: string
+                update_id:
+                  type: integer
+      responses:
+        "200":
+          description: Update acknowledged or Mini App reply returned
+        "400":
+          description: Invalid JSON, request shape, or missing Mini App message
+        "401":
+          description: Invalid Mini App initData signature
+        "503":
+          description: Telegram integration is not configured
+
   /api/resilience/reset:
     post:
       tags: [System]

docs/proposals/TELEGRAM-MINIAPP.md

@@ -0,0 +1,143 @@
+---
+title: "Feasibility — Telegram Mini App Integration"
+version: 3.8.49
+lastUpdated: 2026-08-08
+---
+
+# Telegram Mini App Integration — Feasibility Analysis
+
+**Status: FEASIBLE with moderate effort (estimated 2–4 dev-days for a working slice)**
+
+## 1. What "Telegram Mini App" means here
+
+A Telegram Mini App is an iframe-hosted web app opened inside Telegram (via
+inline buttons / bot menu buttons) that talks to a bot backend through the
+[Telegram WebApp SDK](https://core.telegram.org/bots/webapps). For OmniRoute
+the natural shape is:
+
+- **Bot backend** (new): receives Telegram updates (webhook), validates the
+  Mini App's `initData` signature, and proxies chat requests to OmniRoute's
+  existing OpenAI-compatible `/v1/chat/completions` surface.
+- **Mini App frontend** (new): a small chat UI served by OmniRoute (Next.js
+  route or `public/` static bundle), using the Telegram WebApp JS SDK.
+
+## 2. Current state of the codebase (verified against `main` @ 918fba5e3)
+
+### Already present — outbound notifications only
+
+| Piece                        | Location                                    | What it does                                                                                    |
+| ---------------------------- | ------------------------------------------- | ----------------------------------------------------------------------------------------------- |
+| Telegram webhook integration | `src/lib/webhooks/integrations/telegram.ts` | Builds `sendMessage` payloads for **outbound** gateway events (model, provider, latency, error) |
+| Webhook dispatcher           | `src/lib/webhookDispatcher.ts`              | Routes by kind; decrypts `botToken` from DB metadata for telegram                               |
+| Webhook kinds                | `src/lib/db/webhooks.ts`                    | `slack \| telegram \| discord \| custom`                                                        |
+| Webhook CRUD + test          | `src/app/api/webhooks/*`                    | Create/update/test; telegram kind skips `url` (uses bot token + chat_id)                        |
+| Bot token validation         | `telegram.ts:18`                            | `BOT_TOKEN_RE = /^\d+:[A-Za-z0-9_-]{35,}$/`                                                     |
+| Encryption requirement       | `webhooks/route.ts:77`                      | Telegram webhooks require DB encryption enabled (bot tokens stored at rest)                     |
+
+### Missing — what a Mini App needs that does not exist yet
+
+| Gap                              | Detail                                                                                                                                                                                      |
+| -------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| **Inbound Bot API listener**     | No `setWebhook` registration, no `/bot<token>/getUpdates` polling, no update handling anywhere. Only the `sendMessage` direction exists.                                                    |
+| **WebApp `initData` validation** | No HMAC-SHA256 check of `initData` against the bot token (`WebAppData` hash validation from the Bot API docs).                                                                              |
+| **Telegram bot library**         | `package.json` has no `telegraf`/`grammy`/`telegram-bot-api` dependency. Would need to add one or hand-roll the (small) HMAC + fetch logic.                                                 |
+| **Mini App hosting surface**     | `public/` exists (static assets) and Next.js routes exist; no `/miniapp` route or static bundle yet.                                                                                        |
+| **Session → API key mapping**    | Mini App users need to authenticate to `/v1/chat/completions`. Two options: per-user generated OmniRoute API keys (via `src/lib/db/apiKeys`) or a bot-side proxy that injects a shared key. |
+
+## 3. Constraints
+
+### 3.1 Architectural
+
+- **No existing inbound-bot layer.** The webhook system is strictly
+  event→outbound. A Mini App needs a _new_ Bot API webhook endpoint
+  (`POST /api/telegram/webhook/<botToken-prefix>` or a dedicated route) plus
+  update dispatch. This is additive — no conflicts with the existing
+  `webhooks/` subsystem, but the two must not share the `botToken` storage
+  semantics blindly (webhooks store bot tokens for _outbound_; the Mini App
+  needs the same token for _inbound_ signature checks — same token, new use).
+- **Public HTTPS required.** Telegram only delivers updates to an HTTPS
+  endpoint with a valid cert. Self-hosted OmniRoute behind Tailscale/ngrok
+  needs a public tunnel or Cloudflare Tunnel for the webhook path
+  (a future webhook-URL setting). The dashboard can render the current
+  public origin (`OMNIROUTE_PUBLIC_BASE_URL`) but no webhook registration
+  helper exists.
+- **Encryption gate.** `webhooks/route.ts:77` already refuses telegram
+  kinds without DB encryption. The Mini App bot token has the same
+  sensitivity (it _is_ the HMAC secret for initData validation) — same gate
+  applies, which is a _good_ constraint (no plaintext tokens).
+
+### 3.2 Telegram platform
+
+- **initData is the only trust anchor.** Mini App auth = verify
+  `hash` field of `initData` using HMAC-SHA256(key = SHA256(bot_token),
+  data = sorted `key=value` pairs minus `hash`). Must be implemented
+  server-side; never trust the client.
+- **No inbound push to arbitrary users.** Telegram bots cannot initiate
+  conversations. The Mini App works for users who _already_ have the bot —
+  or you add a `/start` command handler + deep-link (`t.me/bot?startapp=`).
+- **Rate limits.** Bot API ~30 msg/s per bot, 20 msg/min per chat group.
+  Chat responses via `sendMessage`/`answerWebAppQuery` are fine at gateway
+  scale, but streaming must be emulated (send progressive edits or chunked
+  messages) — no native SSE into Telegram.
+- **WebApp SDK quirks.** `Telegram.WebApp.ready()` must be called; theme
+  params come from the SDK; the mini app is sandboxed iframe (no
+  `window.open` to external, clipboard limited). For a chat UI this is fine.
+
+### 3.3 Security / policy
+
+- **Per-user key issuance is the clean model.** Rather than exposing the
+  admin's own API keys, mint a scoped OmniRoute API key per Telegram user
+  (`apiKeys` table + `isModelAllowedForKey` policy), or proxy with a single
+  gateway key and map `user_id` → account. Recommendation: per-user keys so
+  existing rate-limit / model-allowlist / policy code applies unchanged.
+- **initData expiry.** `auth_date` in initData must be checked (Telegram
+  recommends < 24h; short TTLs for chat flows).
+- **Secret handling.** Bot token must stay in the encrypted DB / env —
+  mirror the existing `isEncryptionEnabled()` gate.
+
+## 4. Required next steps (implementation plan)
+
+### Phase 0 — Spike (½–1 dev-day)
+
+1. Add `grammy` or `telegraf` (or ~60 lines of hand-rolled HMAC + fetch).
+2. Implement `src/lib/telegram/initData.ts` — `verifyInitData(initData, botToken)`.
+3. Stand up a throwaway `POST /api/telegram/miniapp/webhook` route behind
+   a dedicated webhook secret; register via `setWebhook` once, locally.
+
+### Phase 1 — Minimal chat slice (1–2 dev-days)
+
+1. **Webhook endpoint** `POST /api/telegram/bot/update` (or
+   `/api/telegram/miniapp/update`): parse Update, verify initData, dispatch.
+2. **Command handler**: `/start` → reply with deep link
+   `https://t.me/<bot>?startapp=<userKey>`; `startapp` param carries a
+   one-time token that maps to a generated OmniRoute API key.
+3. **Chat proxy**: map `initData.user.id` → API key → call
+   `handleChat` (same path as `/v1/chat/completions`) → reply via
+   `sendMessage` (non-stream) or chunked edits (fake streaming).
+4. **Mini App page**: `src/app/(dashboard)/miniapp/page.tsx` (or static
+   bundle in `public/miniapp/`) — Telegram WebApp SDK init + minimal chat
+   UI posting to the bot webhook.
+5. **Config**: `TELEGRAM_BOT_TOKEN` env (or reuse webhook metadata),
+   `OMNIROUTE_PUBLIC_BASE_URL` for webhook URL display; doc in
+   `.env.example` + `ENVIRONMENT.md` (env-doc-sync check).
+
+### Phase 2 — Production hardening (1 dev-day)
+
+- Streaming emulation (message edits), error/backpressure mapping to Bot API
+  limits, per-user key revocation (`/logout` command → revoke API key),
+  usage/rate-limit surfacing (reuse `enforceApiKeyPolicy`), webhook
+  registration helper in dashboard settings, i18n for the mini app UI.
+
+## 5. Verdict
+
+**Feasible.** The gateway already exposes the exact API a Mini App chat
+needs (`/v1/chat/completions` with per-key policy), and the outbound
+Telegram webhook shows the team already handles bot tokens safely
+(encryption gate + token format validation). The genuinely new surface is
+small: an inbound update webhook + initData HMAC verification + a thin
+chat proxy + a static Mini App page. No changes to the core SSE/relay
+pipeline are required.
+
+**Primary risks:** (1) public HTTPS requirement for the webhook (tunnel
+needed on self-hosted installs), (2) no native streaming to Telegram
+(UX tradeoff), (3) initData trust must be strictly server-side.

docs/providers/CHATGPT_WEB.md

@@ -0,0 +1,125 @@
+---
+title: "Providers — ChatGPT Web (session credentials via Cookie Editor)"
+version: 3.8.50
+lastUpdated: 2026-08-08
+---
+
+# Providers — ChatGPT Web (Plus/Pro session credentials)
+
+`chatgpt-web` (alias `cgpt-web`, display name **ChatGPT Web (Plus/Pro)**) sends OpenAI-format chat requests through an authenticated `chatgpt.com` browser session. It authenticates with the `__Secure-next-auth.session-token` cookie — **no API key required**.
+
+> **New to Web Cookie providers?**
+>
+> Read **`docs/getting-started/WEB-COOKIE-GUIDE.md`** for the general setup process, limitations, and troubleshooting before following this provider-specific guide.
+
+---
+
+## 1. What credential does OmniRoute need?
+
+Defined in `src/shared/constants/providers/web-cookie.ts` + `src/shared/providers/webSessionCredentials.ts`:
+
+| Field                      | Value                                                                         |
+| -------------------------- | ----------------------------------------------------------------------------- |
+| Provider id                | `chatgpt-web`                                                                 |
+| Credential name            | `__Secure-next-auth.session-token`                                            |
+| Accepts full Cookie header | ✅ yes                                                                        |
+| Accepted storage keys      | `cookie`, `sessionToken`, `session-token`, `__Secure-next-auth.session-token` |
+
+Two paste formats both work:
+
+- **Bare value** — just the token contents: `eyJhbGciOi...`
+- **Full Cookie header** — `__Secure-next-auth.session-token=eyJhbGciOi...; cf_clearance=...` (preferred — carries rotation/anti-bot cookies the executor needs)
+
+---
+
+## 2. Copy the cookie header with Cookie Editor
+
+Cookie Editor can copy the cookies for the active `chatgpt.com` tab as an HTTP header string.
+Always compare the exported value with a live authenticated request as described in section 3.
+
+### 2.1 Install and pin
+
+1. Install **[Cookie-Editor](https://chromewebstore.google.com/detail/cookie-editor/hlkenndednhfkekhgcdicdfddnkalmdm)** (Moustachauve) in Chrome/Edge, or the Firefox equivalent.
+2. Pin it to the toolbar if you use it regularly.
+
+### 2.2 Copy the credential
+
+1. Go to **https://chatgpt.com** and make sure you're **signed in with the Plus/Pro account** you want OmniRoute to use.
+2. Open a conversation and send at least one message (forces the session token to be live/refreshed).
+3. Click the **Cookie Editor** icon to open its side panel for the active tab.
+4. Find `__Secure-next-auth.session-token`. If it's split into chunks (`__Secure-next-auth.session-token.0`, `.1`, …), select **all** of them — OmniRoute's `nextAuthCookie.ts` merges rotated chunk families.
+5. Click **Copy**, choose **Header string**, and copy the resulting `name=value; name=value` text.
+
+> **If the token is missing:** confirm that you are signed in, send a message to refresh the session, and inspect the live request in section 3.
+
+---
+
+## 3. Verify the required data (before pasting)
+
+The repo's `WEB-COOKIE-GUIDE.md` mandates a live-request check. Do it once per session:
+
+1. With chatgpt.com open, press **F12** → **Network** tab.
+2. Refresh the page, then send a chat message.
+3. Click the conversation request (e.g. `/backend-api/conversation` or the SSE stream) → **Headers** → **Request Headers** → **Cookie**.
+4. Confirm it contains `__Secure-next-auth.session-token=...` — **not** just `cf_clearance` or `__cf_bm`.
+
+The value you copied in step 2.3 must match what the live request sends. If they differ, re-copy from Cookie Editor.
+
+---
+
+## 4. Add / update the credential in OmniRoute
+
+### Dashboard (typical user path)
+
+1. Open the OmniRoute dashboard → **Providers** → **Add Provider**.
+2. Search **ChatGPT Web (Plus/Pro)** (id `chatgpt-web`).
+3. Paste the copied cookie header into the credential field.
+4. Click **Test Connection**.
+5. Save.
+
+If requests later return 401 or 403, re-copy the header from a fresh live session. The executor merges `Set-Cookie` rotations while the connection is active, but it cannot recover a credential that is no longer accepted upstream.
+
+### Bulk / session pools (many accounts)
+
+For multiple ChatGPT sessions, use the bulk web-session import or session-pool endpoints:
+
+- `POST /api/providers/bulk-web-session` — import many cookie credentials at once
+- `GET /api/session-pools` + `/api/session-pools/[provider]` — pool rotation across accounts
+
+Each credential blob must carry the `__Secure-next-auth.session-token` value under one of the accepted storage keys (`cookie`, `sessionToken`, `session-token`, or the cookie's exact name).
+
+### Renewing when the session expires
+
+Web sessions can stop working after sign-out or server-side rotation. Re-run steps 2.2 through 4 whenever requests start failing with 401/403.
+
+---
+
+## 5. Contributing updates
+
+If you changed the credential contract (new storage key, new cookie name, changed hint) or are filling the docs gap, contribute it:
+
+1. Update `src/shared/providers/webSessionCredentials.ts` (credential name / placeholder / storage keys) or `src/shared/constants/providers/web-cookie.ts` (`authHint`).
+2. Update this guide (`docs/providers/CHATGPT_WEB.md`) and the provider table in `docs/getting-started/WEB-COOKIE-GUIDE.md`.
+3. Update `.env.example` + `docs/reference/ENVIRONMENT.md` if you touched env vars, then run:
+   ```bash
+   node scripts/check/check-env-doc-sync.mjs   # must pass
+   ```
+4. Run the provider/unit tests:
+   ```bash
+   npm run test:unit
+   # targeted: tests/unit/chatgpt-web.test.ts (stealth path)
+   ```
+5. Follow `CONTRIBUTING.md`, branch from the current active release tip, use a Conventional Commit message, and open the PR against that active release branch.
+
+> ⚠️ **Never commit a real cookie value.** All examples above are placeholders. If a test fixture needs a token, use a fake `eyJhbGciOi...` string.
+
+---
+
+## Troubleshooting
+
+| Symptom                          | Likely cause                                 | Fix                                                       |
+| -------------------------------- | -------------------------------------------- | --------------------------------------------------------- |
+| Cookie not in Cookie Editor      | Signed out / not HttpOnly-visible            | Sign in; enable HttpOnly display in options               |
+| Token missing from live request  | Request is not authenticated                  | Sign in and send a chat message first                     |
+| 401 after Test Connection passed | Expired or rotated session                    | Re-copy from a fresh live request                         |
+| Chunked token fails              | Only one chunk pasted                        | Select all `__Secure-next-auth.session-token.*` chunks    |

docs/reference/ENVIRONMENT.md

@@ -736,12 +736,12 @@ The logging system writes to both stdout and rotated log files. All configuratio
 | `CALL_LOGS_TABLE_MAX_ROWS`                | `100000`                   | Max rows in the `call_logs` SQLite table before pruning.                          |
 | `ENABLE_REQUEST_LOGS`                     | _(unset)_                  | Force detailed request logging on or off, overriding the dashboard setting.       |
 | `MAX_PENDING_REQUEST_AGE_MS`              | `3600000` (1 hour)         | Max age for orphaned active request log entries before in-memory cleanup.         |
-| `CALL_LOG_PIPELINE_CAPTURE_STREAM_CHUNKS` | `true`                     | Store stream chunks in pipeline artifacts when `call_log_pipeline_enabled=true`.  |
+| `CALL_LOG_PIPELINE_CAPTURE_STREAM_CHUNKS` | `false`                    | Store stream chunks in pipeline artifacts when `call_log_pipeline_enabled=true`. Opt-in (`true`) — off by default to save disk. |
 | `CALL_LOG_PIPELINE_MAX_SIZE_KB`           | `512`                      | Max pipeline call log artifact size in KB when `call_log_pipeline_enabled=true`.  |
 | `PROXY_LOGS_TABLE_MAX_ROWS`               | `100000`                   | Max rows in the `proxy_logs` SQLite table before pruning.                         |
 | `APP_LOG_ROTATION_CHECK_INTERVAL_MS`      | `60000` (1 min)            | How often `src/lib/logRotation.ts` re-checks the active log file size.            |
 | `CHAT_LOG_TEXT_LIMIT`                     | `65536`                    | Max string length retained in chat log artifacts (default 64 KB).                 |
-| `CHAT_LOG_ARRAY_TAIL_ITEMS`               | `24`                       | Number of array items retained from the tail when truncating chat log payloads.   |
+| `CHAT_LOG_ARRAY_TAIL_ITEMS`               | `128`                      | Number of array items retained from the tail when truncating chat log payloads.   |
 | `CHAT_LOG_MAX_DEPTH`                      | `6`                        | Max nesting depth before chat log payloads are truncated.                         |
 | `CHAT_LOG_MAX_OBJECT_KEYS`                | `80`                       | Max object keys retained in chat log payloads (0 = unlimited).                    |
 | `CHAT_DEBUG_FILE`                         | `false`                    | When true, `serializeArtifactForStorage` skips size-based truncation. Debug only. |
@@ -976,7 +976,7 @@ changing them requires a code edit, not an env var:
 | `CURSOR_AGENT_CLI_VERSION`       | _(detect / pin)_    | `open-sse/utils/cursorAgentCliVersion.ts`  | Agent CLI build id (`YYYY.MM.DD-<hash>`) for `x-cursor-client-version: cli-…` on Agent Run.   |
 | `CURSOR_DATA_DIR`                | _(probed)_          | `open-sse/utils/cursorAgentCliVersion.ts`  | Override Cursor Agent CLI data dir (`…/versions/<id>`); same var the official agent uses.    |
 | `CURSOR_TOKEN`                   | _(unset)_           | `scripts/ad-hoc/cursor-tap.cjs`            | Direct Cursor bearer token used by developer tooling.                                        |
-| `OMNIROUTE_LOG_REQUEST_SHAPE`    | enabled (`!== "0"`) | `src/app/api/v1/chat/completions/route.ts` | Log content-type/length markers for large chat payloads. Set `"0"` to silence.               |
+| `OMNIROUTE_LOG_REQUEST_SHAPE`    | disabled (opt-in via `"1"`) | `src/app/api/v1/chat/completions/route.ts` | Log content-type/length markers for large chat payloads when `"1"` is set. Off by default to reduce log noise. |
 | `DEBUG_RESPONSES_SSE_TO_JSON`    | _(unset)_           | `open-sse/handlers/responseTranslator.ts`  | Set `true` to log Responses API SSE→JSON translation details.                                |
 | `NEXT_PUBLIC_OMNIROUTE_E2E_MODE` | _(unset)_           | E2E test harness                           | Set `true` to enable E2E test mode (relaxed auth, test hooks).                               |
 
@@ -1389,3 +1389,31 @@ Used by `src/lib/vncSession/manifest.ts` to configure Docker-based headless Chro
 | `REDIS_BIND_HOST` | `127.0.0.1` | Bind address for the embedded Redis service. |
 | `REDIS_PORT` | `6379` | Port for the embedded Redis service. |
 | `OMNIROUTE_REDIS_BIND_HOST` | – | OmniRoute-scoped override for the embedded Redis bind address. |
+
+---
+
+## 24. Release v3.8.50 additions
+
+These settings were introduced after the previous environment-contract snapshot.
+
+| Variable | Default | Source File | Description |
+| --- | --- | --- | --- |
+| `OMNIROUTE_CHAT_ADMISSION_QUEUE_MS` | `5000` | `src/shared/middleware/chatBodyAdmission.ts` | Maximum wait for a heavyweight chat admission slot before a retryable `503`; `0` restores immediate rejection. |
+| `OMNIROUTE_RUNNOW_TIMEOUT_MS` | `30000` | `src/app/api/jobs/[id]/run-now/route.ts` | Bounds how long a run-now call waits for an in-flight job before starting the queued run. |
+| `CHAT_LOG_MAX_BODY_KB` | `1024` | `src/lib/logEnv.ts` | Maximum request or response body size before log summarization, in KiB. |
+| `ADOBE_FIREFLY_BROWSER_REFRESH` | enabled | `open-sse/services/adobeFireflySession.ts` | Keeps IMS and browser-risk state fresh through account-scoped Chrome CDP sessions; set `0` to disable. |
+| `ADOBE_FIREFLY_SESSION_DISK` | enabled | `open-sse/services/adobeFireflySession.ts` | Persists repaired Adobe sessions under `DATA_DIR`; set `0` for memory-only state. |
+| `ADOBE_FIREFLY_MIN_SUBMIT_GAP_MS` | `12000` | `open-sse/services/adobeFireflySession.ts` | Minimum spacing between Adobe Firefly generate submissions. |
+| `ADOBE_FIREFLY_BATCH_EXTRA_GAP_MS` | `15000` | `open-sse/services/adobeFireflySession.ts` | Extra quiet period after every third successful Adobe submission. |
+| `ADOBE_FIREFLY_CHROME_CDP_PORT` | `9334` | `open-sse/services/adobeFireflyChromeRuntime.ts` | CDP port for the account-scoped Chrome runtime. |
+| `ADOBE_FIREFLY_CHROME_VISIBLE` | `0` | `open-sse/services/adobeFireflyChromeRuntime.ts` | Set `1` to keep the Adobe renewal browser visible; the default parks a headed window off-screen. |
+| `ADOBE_FIREFLY_CHROME_HEADLESS` | `0` | `open-sse/services/adobeFireflyChromeRuntime.ts` | Debug-only true-headless mode; Adobe colligo normally rejects the resulting risk session. |
+| `ADOBE_FIREFLY_CHROME_FORCE_RESTART` | `0` | `open-sse/services/adobeFireflyChromeRuntime.ts` | Set `1` to restart the account-scoped Chrome runtime before renewal. |
+| `ADOBE_FIREFLY_CHROME_PING` | automatic | `open-sse/services/adobeFireflyChromeRuntime.ts` | `1` forces, and `0` disables, the in-page generate probe used to prove the renewed ARP session. |
+| `ADOBE_FIREFLY_LOGIN_WAIT_MS` | context-dependent | `open-sse/services/adobeFireflyChromeRuntime.ts` | Interactive-login wait budget: `0` on background renewal and `300000` on the explicit login flow unless overridden. |
+| `ADOBE_FIREFLY_FORTER_WAIT_MS` | `45000` | `open-sse/services/adobeFireflyChromeRuntime.ts` | Maximum wait for a fresh Forter token during session renewal. |
+| `CHROME_PATH` | auto-detect | `open-sse/services/adobeFireflyChromeRuntime.ts` | Optional absolute Chrome executable used when platform auto-detection is insufficient. |
+| `TELEGRAM_BOT_TOKEN` | _(unset)_ | `src/lib/telegram/config.ts` | BotFather token that enables the inbound webhook and signs Mini App `initData`. |
+| `TELEGRAM_DEFAULT_MODEL` | `auto/chat` | `src/lib/telegram/chatProxy.ts` | Model used for Telegram chat replies. |
+| `TELEGRAM_BOT_API_BASE` | `https://api.telegram.org` | `src/lib/telegram/config.ts` | Bot API base URL override for proxies or self-hosted Bot API servers. |
+| `TELEGRAM_WEBHOOK_TIMEOUT_MS` | `60000` | `src/lib/telegram/config.ts` | Timeout in milliseconds for outbound Bot API calls. |

docs/security/GUARDRAILS.md

@@ -1,13 +1,13 @@
 ---
 title: "Guardrails"
 version: 3.8.50
-lastUpdated: 2026-08-07
+lastUpdated: 2026-08-08
 ---
 
 # Guardrails
 
 > **Source of truth:** `src/lib/guardrails/`
-> **Last updated:** 2026-08-07 — v3.8.50 (Modality Bridge PR-1: mode selector, task-aware prompt, describe cache, transparency header + stats)
+> **Last updated:** 2026-08-08 — v3.8.50 (Modality Bridge PR-2: dashboard settings page, live stats, test action, and media-provider shortcuts)
 
 Guardrails enforce safety, policy, and content transformations at the boundary
 between OmniRoute and upstream providers. Each guardrail can inspect (and
@@ -134,6 +134,22 @@ swap is already visible in the response body's `model` field.
 PR-3-reserved `audio`). Counters reset on process restart by design
 (telemetry, not accounting).
 
+#### Dashboard configuration
+
+The dedicated dashboard page is
+`/dashboard/settings/modality-bridge`. Its URL-addressable `Vision`, `Audio`,
+and `Video` tabs preserve query parameters while switching the `tab` value.
+The Vision tab is live: it exposes enablement, mode, model selection (including
+the automatic default), task-aware prompting, advanced timeout/image/cache
+limits, runtime counters, and a guarded sample request. Audio and Video are
+explicit placeholders: Audio is reserved for PR-3, while Video remains tracked
+in issue `#9760`.
+
+The former Vision Bridge card under AI settings is a compatibility link to the
+new page; it no longer owns a second copy of the form. Media Providers also
+links Image-to-Text and Speech-to-Text workflows to the corresponding Modality
+Bridge tabs without removing the existing Speech-to-Text playground.
+
 **Self-loop admission bypass:** when the describe call routes through OmniRoute's
 own `/v1` self-loop (non-standard provider model), the sub-request sends
 `x-omniroute-admission-bypass: internal` and is authenticated with the resolved
@@ -355,9 +371,16 @@ Environment variables read by the built-in guardrails:
 | `PII_RESPONSE_SANITIZATION` / `_MODE` | `pii-masker` (downstream) | Controls response-side masker behavior.                                                             |
 
 The Vision Bridge reads runtime config from the DB-backed settings store
-(`getSettings()`), not env vars: `visionBridgeEnabled`, `visionBridgeModel`,
-`visionBridgePrompt`, `visionBridgeTimeout`, `visionBridgeMaxImages`. Defaults
-live in `src/shared/constants/visionBridgeDefaults.ts`.
+(`getSettings()`), not env vars. The primary keys are
+`modalityBridgeVisionEnabled`, `modalityBridgeVisionMode`,
+`modalityBridgeVisionModel`, `modalityBridgeVisionTaskAware`,
+`modalityBridgeVisionPrompt`, `modalityBridgeVisionTimeout`,
+`modalityBridgeVisionMaxImages`, `modalityBridgeCacheEnabled`,
+`modalityBridgeCacheTtlMinutes`, and `modalityBridgeCacheMaxEntries`. The legacy
+`visionBridge*` keys are accepted only as the documented one-cycle read
+fallback; dashboard writes use the primary keys. Defaults and the fallback
+resolver live in `src/shared/constants/modalityBridgeDefaults.ts`, with legacy
+constants retained in `src/shared/constants/visionBridgeDefaults.ts`.
 
 ## Custom Guardrails
 

docs/video-preset-generation.md

@@ -0,0 +1,113 @@
+# Video Generation Through Preset Jobs
+
+Custom provider nodes whose `/videos` surface is an **async submit → poll → fetch-result API** (instead of a synchronous generation endpoint) can be wired into the `/api/v1/videos/generations` route without any new provider code. The model row carries a `generationConfig.preset`, and the dispatcher routes the request through a single job executor that is configured entirely by declarative preset data.
+
+## How dispatch works
+
+1. The route parses `model` as `provider/model` and resolves the provider node's credentials (`POST /api/v1/videos/generations`).
+2. `handleVideoGeneration` (in `open-sse/handlers/videoGeneration.ts`) checks whether the provider is a **custom provider node** (no entry in the static video registry).
+3. For custom nodes it reads the custom model row via `getCustomModelVideoPreset(provider, model)`:
+   - The model row has `generationConfig.preset` set (e.g. `"agnes-video-job"`) → dispatch through the **job executor** (`open-sse/handlers/videoGeneration/job.ts`).
+   - The preset name does not match any known preset → **502** `Unknown video job preset: <preset>` (server-side misconfiguration).
+   - No preset configured → fall back to the generic OpenAI-compatible sync handler, mirroring the images route.
+4. The job executor runs the preset pipeline: **submit** the job, **poll** for terminal status, **read** the finished video URL, and return the standard OpenAI-compatible response shape.
+
+The executor is one handler family; every provider-specific detail (paths, auth, body shape, status/result fields, poll cadence) is data in the preset definition.
+
+## Response contract
+
+Both the sync and job paths return the same shape:
+
+```json
+{
+  "created": 1234567890,
+  "data": [{ "url": "https://…", "format": "mp4" }]
+}
+```
+
+This is the shape the media-generation consumer reads (`data.data[0].url`), so preset-job providers are drop-in replacements for sync providers.
+
+## Presets
+
+Presets live in `open-sse/handlers/videoGeneration/job.ts` (`VIDEO_JOB_PRESETS`). Each preset declares:
+
+| Field                                        | Meaning                                                                                                                                                                                                |
+| -------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
+| `authHeaderName` / `authScheme`              | `x-api-key` with `raw` value (Agnes, muapi) or `Authorization` with `Bearer` prefix (Sora). Missing credentials → request goes out without an auth header.                                             |
+| `baseUrlFallback`                            | Default base URL. Overridden by the provider connection's `providerSpecificData.baseUrl` (or top-level `baseUrl`), which wins when set.                                                                |
+| `submit.path` / `submit.buildBody`           | Where and how the job is submitted. `{model}` in the path is substituted with the encoded model id; the body is built from `model`/`prompt`/`duration` plus pass-through of every other request field. |
+| `taskIdPath`                                 | Dot path into the submit response identifying the job (e.g. `task_id`, `request_id`, `id`). Missing job id → **502**.                                                                                  |
+| `poll.pathTemplate`                          | Poll URL template; `{taskId}` is substituted.                                                                                                                                                          |
+| `statusPath` / `statusDone` / `statusFailed` | Where the job status lives and which values are terminal.                                                                                                                                              |
+| `resultPath`                                 | Dot path into the poll response holding the finished video URL: a string, a string array, or an array of `{ url }` objects are all accepted. Completed job with no readable URL → **502**.             |
+| `maxPolls` / `pollIntervalMs`                | Poll budget (default 60 polls × 2000 ms). Exhausted → **504** `Video job timed out`.                                                                                                                   |
+
+### `agnes-video-job` — Agnes Video V2.0
+
+- Auth: `x-api-key: <key>` (raw).
+- Base URL fallback: `https://apihub.agnes-ai.com`.
+- Submit: `POST /v1/videos` with `{ model, prompt, ...extras }` — image, mode, `num_frames`, `frame_rate` and other provider knobs pass through untouched.
+- Job id: `task_id` from the submit response.
+- Poll: `GET /v1/videos/{taskId}`; status at `status` (`completed` / `failed`).
+- Result: `metadata.url` — the completed video URL is returned as JSON metadata, not a binary body.
+
+### `muapi-video-job` — muapi.ai
+
+- Auth: `x-api-key: <key>` (raw).
+- Base URL fallback: `https://api.muapi.ai`.
+- Submit: `POST /api/v1/{model}` with `{ prompt, duration?, ...extras }`.
+- Job id: `request_id` from the submit response.
+- Poll: `GET /api/v1/predictions/{taskId}/result`; status at `status` (`completed` / `failed`).
+- Result: `outputs` — an array of video URLs.
+
+### `sora-job` — OpenAI Sora
+
+- Auth: `Authorization: Bearer <key>`.
+- Base URL fallback: `https://api.openai.com`.
+- Submit: `POST /v1/videos` with `{ model, prompt, seconds?, ...extras }`. `seconds` is a **string** enum (`"4" | "8" | "12"`) in the Sora API, so a numeric `duration` is stringified; size mapping is intentionally not forced.
+- Job id: `id` from the submit response.
+- Poll: `GET /v1/videos/{taskId}`; status at `status` (`completed` / `failed`).
+- Result: `data` — an array whose entries are either a URL string or `{ url: "…" }`.
+
+## Setup
+
+1. **Register the provider node** as an OpenAI-compatible custom provider (`providerSpecificData.baseUrl` optional — the preset's `baseUrlFallback` is used when absent).
+2. **Register a custom model** tagged with the `videos` endpoint and a `generationConfig`:
+
+   ```json
+   {
+     "id": "super-video-v1",
+     "name": "Super Video v1",
+     "source": "manual",
+     "apiFormat": "chat-completions",
+     "supportedEndpoints": ["videos"],
+     "generationConfig": { "preset": "agnes-video-job" }
+   }
+   ```
+
+   `addCustomModel` (in `src/lib/db/models.ts`) accepts `generationConfig?: { preset: string }` as its final parameter and persists it on the model row; `updateCustomModel` forwards it the same way. The provider-models API accepts `generationConfig` on create and update.
+
+3. **Call the route** as usual:
+
+   ```bash
+   curl -X POST http://localhost:8787/api/v1/videos/generations \
+     -H "Content-Type: application/json" \
+     -H "Authorization: Bearer $API_KEY" \
+     -d '{
+       "model": "my-custom-provider/super-video-v1",
+       "prompt": "a cat playing piano",
+       "duration": 5
+     }'
+   ```
+
+## Troubleshooting
+
+| Symptom                                                                        | Cause                                                                                            |
+| ------------------------------------------------------------------------------ | ------------------------------------------------------------------------------------------------ |
+| `400 Unknown video provider: …`                                                | Non-custom provider not in the static registry; preset jobs only apply to custom provider nodes. |
+| `502 Unknown video job preset: …`                                              | `generationConfig.preset` does not match any preset in `VIDEO_JOB_PRESETS`. Fix the model row.   |
+| `502 Video provider did not return a job id (…)`                               | Submit succeeded but the response had no readable value at `taskIdPath`.                         |
+| `502 Video job failed (…)` / `Video job completed but no result URL found (…)` | Poll reached a terminal `statusFailed` state, or `resultPath` held no readable URL.              |
+| `504 Video job timed out after 60 polls (…)`                                   | Job never reached a terminal status within the poll budget.                                      |
+| Upstream 4xx/5xx passthrough                                                   | `fetchJson` returns the upstream status when the submit/poll request itself is not OK.           |
+| Requests go out without auth                                                   | No `apiKey`/`accessToken` on the provider connection; the executor sends `Content-Type` only.    |

electron/package-lock.json

@@ -55,9 +55,9 @@
       "license": "MIT"
     },
     "node_modules/@electron/asar/node_modules/brace-expansion": {
-      "version": "1.1.16",
-      "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-1.1.16.tgz",
-      "integrity": "sha512-IDw48K2/2kRkg9LdJxurvq3lV3aBgq0REY89duEqFRthjlPdXHKMj7EnQOXVckxzgisinf3nHfrcE2FufFLXMw==",
+      "version": "1.1.18",
+      "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-1.1.18.tgz",
+      "integrity": "sha512-Edep/X9fGqVNmzKBVsDYIOtD+z1tuezV70LBjdCst9Tqu76lsnvRiZ6oTic1n+/BIwX6QDGAO94PN4N2SADvtw==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
@@ -257,9 +257,9 @@
       "license": "MIT"
     },
     "node_modules/@electron/universal/node_modules/brace-expansion": {
-      "version": "2.1.2",
-      "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-2.1.2.tgz",
-      "integrity": "sha512-w5JZcKgdhDOgOwm8H+KgbosopHMuGcl6qbulwjtz3SM7I7P3yW1eAjzMPLrIE+NQ9vjgANKHWeMHnrT0OXW1oA==",
+      "version": "2.1.4",
+      "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-2.1.4.tgz",
+      "integrity": "sha512-hGfVzPxthbf3+2yjg/RBs60cB0FhqBS/zvdV/4wn4/BmN0bNMMHPc4V/BbFieqf1TKAGGAHnY4eSjajCl0f2Xg==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
@@ -297,6 +297,45 @@
         "url": "https://github.com/sponsors/isaacs"
       }
     },
+    "node_modules/@electron/windows-sign": {
+      "version": "1.2.2",
+      "resolved": "https://registry.npmjs.org/@electron/windows-sign/-/windows-sign-1.2.2.tgz",
+      "integrity": "sha512-dfZeox66AvdPtb2lD8OsIIQh12Tp0GNCRUDfBHIKGpbmopZto2/A8nSpYYLoedPIHpqkeblZ/k8OV0Gy7PYuyQ==",
+      "dev": true,
+      "license": "BSD-2-Clause",
+      "optional": true,
+      "peer": true,
+      "dependencies": {
+        "cross-dirname": "^0.1.0",
+        "debug": "^4.3.4",
+        "fs-extra": "^11.1.1",
+        "minimist": "^1.2.8",
+        "postject": "^1.0.0-alpha.6"
+      },
+      "bin": {
+        "electron-windows-sign": "bin/electron-windows-sign.js"
+      },
+      "engines": {
+        "node": ">=14.14"
+      }
+    },
+    "node_modules/@electron/windows-sign/node_modules/fs-extra": {
+      "version": "11.4.0",
+      "resolved": "https://registry.npmjs.org/fs-extra/-/fs-extra-11.4.0.tgz",
+      "integrity": "sha512-EQsFzMUJkCKGr1ePqlYADkIUmHW1s3ZXr5Yqy6wbGrfUCphpl2maM/kyOIRA2HpP3AaFQTZXD4ldjek+nccddA==",
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "peer": true,
+      "dependencies": {
+        "graceful-fs": "^4.2.0",
+        "jsonfile": "^6.0.1",
+        "universalify": "^2.0.0"
+      },
+      "engines": {
+        "node": ">=14.14"
+      }
+    },
     "node_modules/@isaacs/fs-minipass": {
       "version": "4.0.1",
       "resolved": "https://registry.npmjs.org/@isaacs/fs-minipass/-/fs-minipass-4.0.1.tgz",
@@ -835,16 +874,16 @@
       "optional": true
     },
     "node_modules/brace-expansion": {
-      "version": "5.0.7",
-      "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-5.0.7.tgz",
-      "integrity": "sha512-7oFy703dxfY3/NLxC1fh2SUCQ0H9rmAY+5EpDVfXjUTTs+HEwR2nYaqLv+GWcTsumwxPfiz6CzCNkwXwBUwqCA==",
+      "version": "5.0.9",
+      "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-5.0.9.tgz",
+      "integrity": "sha512-ScQ4IuvIEF1TMlP7Zt+vjJ//9zlPb2SDcxWxM3bk8s6t6GGdJ7KO1dCcTidOPJKePW30LE/2cT7wCyPho9/Wxg==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
         "balanced-match": "^4.0.2"
       },
       "engines": {
-        "node": "18 || 20 || >=22"
+        "node": "20 || >=22"
       }
     },
     "node_modules/buffer-from": {
@@ -1091,6 +1130,15 @@
       "dev": true,
       "license": "MIT"
     },
+    "node_modules/cross-dirname": {
+      "version": "0.1.0",
+      "resolved": "https://registry.npmjs.org/cross-dirname/-/cross-dirname-0.1.0.tgz",
+      "integrity": "sha512-+R08/oI0nl3vfPcqftZRpytksBXDzOUveBq/NBVx0sUp1axwzPQrKinNx5yd5sxPu8j1wIy8AfnVQ+5eFdha6Q==",
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "peer": true
+    },
     "node_modules/cross-spawn": {
       "version": "7.0.6",
       "resolved": "https://registry.npmjs.org/cross-spawn/-/cross-spawn-7.0.6.tgz",
@@ -1260,9 +1308,9 @@
       "license": "MIT"
     },
     "node_modules/dir-compare/node_modules/brace-expansion": {
-      "version": "1.1.16",
-      "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-1.1.16.tgz",
-      "integrity": "sha512-IDw48K2/2kRkg9LdJxurvq3lV3aBgq0REY89duEqFRthjlPdXHKMj7EnQOXVckxzgisinf3nHfrcE2FufFLXMw==",
+      "version": "1.1.18",
+      "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-1.1.18.tgz",
+      "integrity": "sha512-Edep/X9fGqVNmzKBVsDYIOtD+z1tuezV70LBjdCst9Tqu76lsnvRiZ6oTic1n+/BIwX6QDGAO94PN4N2SADvtw==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
@@ -1411,6 +1459,19 @@
         "node": ">=14.0.0"
       }
     },
+    "node_modules/electron-builder-squirrel-windows": {
+      "version": "26.15.3",
+      "resolved": "https://registry.npmjs.org/electron-builder-squirrel-windows/-/electron-builder-squirrel-windows-26.15.3.tgz",
+      "integrity": "sha512-Jc19XPV9y9+2bAdZPkXuVNGNIEFBq9poHC61l8Kv6FdK7DRG3+Ic0rerC0DXOaeHNz8yW0fg/JnF8GQROOF5MA==",
+      "dev": true,
+      "license": "MIT",
+      "peer": true,
+      "dependencies": {
+        "app-builder-lib": "26.15.3",
+        "builder-util": "26.15.3",
+        "electron-winstaller": "5.4.0"
+      }
+    },
     "node_modules/electron-publish": {
       "version": "26.15.3",
       "resolved": "https://registry.npmjs.org/electron-publish/-/electron-publish-26.15.3.tgz",
@@ -1445,6 +1506,66 @@
         "tiny-typed-emitter": "^2.1.0"
       }
     },
+    "node_modules/electron-winstaller": {
+      "version": "5.4.0",
+      "resolved": "https://registry.npmjs.org/electron-winstaller/-/electron-winstaller-5.4.0.tgz",
+      "integrity": "sha512-bO3y10YikuUwUuDUQRM4KfwNkKhnpVO7IPdbsrejwN9/AABJzzTQ4GeHwyzNSrVO+tEH3/Np255a3sVZpZDjvg==",
+      "dev": true,
+      "hasInstallScript": true,
+      "license": "MIT",
+      "peer": true,
+      "dependencies": {
+        "@electron/asar": "^3.2.1",
+        "debug": "^4.1.1",
+        "fs-extra": "^7.0.1",
+        "lodash": "^4.17.21",
+        "temp": "^0.9.0"
+      },
+      "engines": {
+        "node": ">=8.0.0"
+      },
+      "optionalDependencies": {
+        "@electron/windows-sign": "^1.1.2"
+      }
+    },
+    "node_modules/electron-winstaller/node_modules/fs-extra": {
+      "version": "7.0.1",
+      "resolved": "https://registry.npmjs.org/fs-extra/-/fs-extra-7.0.1.tgz",
+      "integrity": "sha512-YJDaCJZEnBmcbw13fvdAM9AwNOJwOzrE4pqMqBq5nFiEqXUqHwlK4B+3pUw6JNvfSPtX05xFHtYy/1ni01eGCw==",
+      "dev": true,
+      "license": "MIT",
+      "peer": true,
+      "dependencies": {
+        "graceful-fs": "^4.1.2",
+        "jsonfile": "^4.0.0",
+        "universalify": "^0.1.0"
+      },
+      "engines": {
+        "node": ">=6 <7 || >=8"
+      }
+    },
+    "node_modules/electron-winstaller/node_modules/jsonfile": {
+      "version": "4.0.0",
+      "resolved": "https://registry.npmjs.org/jsonfile/-/jsonfile-4.0.0.tgz",
+      "integrity": "sha512-m6F1R3z8jjlf2imQHS2Qez5sjKWQzbuuhuJ/FKYFRZvPE3PuHcSMVZzfsLhGVOkfd20obL5SWEBew5ShlquNxg==",
+      "dev": true,
+      "license": "MIT",
+      "peer": true,
+      "optionalDependencies": {
+        "graceful-fs": "^4.1.6"
+      }
+    },
+    "node_modules/electron-winstaller/node_modules/universalify": {
+      "version": "0.1.2",
+      "resolved": "https://registry.npmjs.org/universalify/-/universalify-0.1.2.tgz",
+      "integrity": "sha512-rBJeI5CXAlmy1pV+617WB9J63U6XcazHHF2f2dbJix4XzpUF0RS3Zbj0FGIOCAva5P/d/GBOYaACQ1w+0azUkg==",
+      "dev": true,
+      "license": "MIT",
+      "peer": true,
+      "engines": {
+        "node": ">= 4.0.0"
+      }
+    },
     "node_modules/emoji-regex": {
       "version": "8.0.0",
       "resolved": "https://registry.npmjs.org/emoji-regex/-/emoji-regex-8.0.0.tgz",
@@ -1627,9 +1748,9 @@
       "license": "MIT"
     },
     "node_modules/filelist/node_modules/brace-expansion": {
-      "version": "2.1.2",
-      "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-2.1.2.tgz",
-      "integrity": "sha512-w5JZcKgdhDOgOwm8H+KgbosopHMuGcl6qbulwjtz3SM7I7P3yW1eAjzMPLrIE+NQ9vjgANKHWeMHnrT0OXW1oA==",
+      "version": "2.1.4",
+      "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-2.1.4.tgz",
+      "integrity": "sha512-hGfVzPxthbf3+2yjg/RBs60cB0FhqBS/zvdV/4wn4/BmN0bNMMHPc4V/BbFieqf1TKAGGAHnY4eSjajCl0f2Xg==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
@@ -1792,9 +1913,9 @@
       "license": "MIT"
     },
     "node_modules/glob/node_modules/brace-expansion": {
-      "version": "1.1.16",
-      "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-1.1.16.tgz",
-      "integrity": "sha512-IDw48K2/2kRkg9LdJxurvq3lV3aBgq0REY89duEqFRthjlPdXHKMj7EnQOXVckxzgisinf3nHfrcE2FufFLXMw==",
+      "version": "1.1.18",
+      "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-1.1.18.tgz",
+      "integrity": "sha512-Edep/X9fGqVNmzKBVsDYIOtD+z1tuezV70LBjdCst9Tqu76lsnvRiZ6oTic1n+/BIwX6QDGAO94PN4N2SADvtw==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
@@ -2113,9 +2234,9 @@
       }
     },
     "node_modules/js-yaml": {
-      "version": "4.3.0",
-      "resolved": "https://registry.npmjs.org/js-yaml/-/js-yaml-4.3.0.tgz",
-      "integrity": "sha512-1td788aAnnZ5qs7V2QIRl1owjtYpbKt749Y3xauqQgwIIGF/xXWz1wMTEBx5O3LK3lXLVuqXPdPxj2BoFHaW9Q==",
+      "version": "4.3.1",
+      "resolved": "https://registry.npmjs.org/js-yaml/-/js-yaml-4.3.1.tgz",
+      "integrity": "sha512-CY6crGq313MX8GkwvB7tzgp99vjQxY1++5y10/BKN/GUfHqWaOGQMNZkBvqSzsZKWk/ijwHlWzzkLulsGHhjWQ==",
       "funding": [
         {
           "type": "github",
@@ -2359,6 +2480,20 @@
         "node": ">= 18"
       }
     },
+    "node_modules/mkdirp": {
+      "version": "0.5.6",
+      "resolved": "https://registry.npmjs.org/mkdirp/-/mkdirp-0.5.6.tgz",
+      "integrity": "sha512-FP+p8RB8OWpF3YZBCrP5gtADmtXApB5AMLn+vdyA+PyxCjrCs00mjyUozssO33cwDeT3wNGdLxJ5M//YqtHAJw==",
+      "dev": true,
+      "license": "MIT",
+      "peer": true,
+      "dependencies": {
+        "minimist": "^1.2.6"
+      },
+      "bin": {
+        "mkdirp": "bin/cmd.js"
+      }
+    },
     "node_modules/ms": {
       "version": "2.1.3",
       "resolved": "https://registry.npmjs.org/ms/-/ms-2.1.3.tgz",
@@ -2622,6 +2757,36 @@
         "node": ">=18"
       }
     },
+    "node_modules/postject": {
+      "version": "1.0.0-alpha.6",
+      "resolved": "https://registry.npmjs.org/postject/-/postject-1.0.0-alpha.6.tgz",
+      "integrity": "sha512-b9Eb8h2eVqNE8edvKdwqkrY6O7kAwmI8kcnBv1NScolYJbo59XUF0noFq+lxbC1yN20bmC0WBEbDC5H/7ASb0A==",
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "peer": true,
+      "dependencies": {
+        "commander": "^9.4.0"
+      },
+      "bin": {
+        "postject": "dist/cli.js"
+      },
+      "engines": {
+        "node": ">=14.0.0"
+      }
+    },
+    "node_modules/postject/node_modules/commander": {
+      "version": "9.5.0",
+      "resolved": "https://registry.npmjs.org/commander/-/commander-9.5.0.tgz",
+      "integrity": "sha512-KRs7WVDKg86PWiuAqhDrAQnTXZKraVcCc6vFdL14qrZ/DcWwuRo7VoiYXalXO7S5GKpqYiVEwCbgFDfxNHKJBQ==",
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "peer": true,
+      "engines": {
+        "node": "^12.20.0 || >=14"
+      }
+    },
     "node_modules/proc-log": {
       "version": "6.1.0",
       "resolved": "https://registry.npmjs.org/proc-log/-/proc-log-6.1.0.tgz",
@@ -2816,6 +2981,21 @@
         "node": ">= 4"
       }
     },
+    "node_modules/rimraf": {
+      "version": "2.6.3",
+      "resolved": "https://registry.npmjs.org/rimraf/-/rimraf-2.6.3.tgz",
+      "integrity": "sha512-mwqeW5XsA2qAejG46gYdENaxXjx9onRNCfn7L0duuP4hCuTIi/QO7PDK07KJfp1d+izWPrzEJDcSqBa0OZQriA==",
+      "deprecated": "Rimraf versions prior to v4 are no longer supported",
+      "dev": true,
+      "license": "ISC",
+      "peer": true,
+      "dependencies": {
+        "glob": "^7.1.3"
+      },
+      "bin": {
+        "rimraf": "bin.js"
+      }
+    },
     "node_modules/roarr": {
       "version": "2.15.4",
       "resolved": "https://registry.npmjs.org/roarr/-/roarr-2.15.4.tgz",
@@ -3045,9 +3225,9 @@
       }
     },
     "node_modules/tar": {
-      "version": "7.5.20",
-      "resolved": "https://registry.npmjs.org/tar/-/tar-7.5.20.tgz",
-      "integrity": "sha512-9FcyK4PA6+WbzlTM9WhQm6vB5W7cP7dUiPsv1g7YDwEQnQ1CGpK3MGlKk/ITVWMk05kHZuBhmVhiv8LZoy/PFQ==",
+      "version": "7.5.22",
+      "resolved": "https://registry.npmjs.org/tar/-/tar-7.5.22.tgz",
+      "integrity": "sha512-MFO/QzvtAOmJbkhOaCTvbGcFN9L9b+JunIsDwaKljSOdcLMea3NJ1k9Usz/rjdfSXTq4dfzfeS7W4p4YOAAHeA==",
       "dev": true,
       "license": "BlueOak-1.0.0",
       "dependencies": {
@@ -3071,6 +3251,21 @@
         "node": ">=18"
       }
     },
+    "node_modules/temp": {
+      "version": "0.9.4",
+      "resolved": "https://registry.npmjs.org/temp/-/temp-0.9.4.tgz",
+      "integrity": "sha512-yYrrsWnrXMcdsnu/7YMYAofM1ktpL5By7vZhf15CrXijWWrEYZks5AXBudalfSWJLlnen/QUJUB5aoB0kqZUGA==",
+      "dev": true,
+      "license": "MIT",
+      "peer": true,
+      "dependencies": {
+        "mkdirp": "^0.5.1",
+        "rimraf": "~2.6.2"
+      },
+      "engines": {
+        "node": ">=6.0.0"
+      }
+    },
     "node_modules/temp-file": {
       "version": "3.4.0",
       "resolved": "https://registry.npmjs.org/temp-file/-/temp-file-3.4.0.tgz",

open-sse/config/audioRegistry.ts

@@ -247,6 +247,17 @@ export const AUDIO_TRANSCRIPTION_PROVIDERS: Record<string, AudioProvider> = {
     format: "speechmatics",
     models: [{ id: "enhanced", name: "Enhanced" }],
   },
+
+  nanogpt: {
+    id: "nanogpt",
+    baseUrl: "https://nano-gpt.com/api/v1/audio/transcriptions",
+    authType: "apikey",
+    authHeader: "bearer",
+    models: [
+      { id: "whisper-1", name: "Whisper 1" },
+      { id: "gpt-4o-transcription", name: "GPT-4o Transcription" },
+    ],
+  },
 };
 
 /**
@@ -570,6 +581,17 @@ export const AUDIO_SPEECH_PROVIDERS: Record<string, AudioProvider> = {
       { id: "mimo-v2.5-tts-voiceclone", name: "MiMo V2.5 Voice Clone" },
     ],
   },
+
+  nanogpt: {
+    id: "nanogpt",
+    baseUrl: "https://nano-gpt.com/api/v1/audio/speech",
+    authType: "apikey",
+    authHeader: "bearer",
+    models: [
+      { id: "tts-1-hd", name: "TTS 1 HD" },
+      { id: "tts-1", name: "TTS 1" },
+    ],
+  },
 };
 
 /**

open-sse/config/constants.ts

@@ -1,4 +1,5 @@
 import { getUpstreamTimeoutConfig } from "@/shared/utils/runtimeTimeouts";
+import { resolvePublicCred } from "../utils/publicCreds.ts";
 import type { LegacyProvider } from "./providerRegistry.ts";
 import { loadProviderCredentials } from "./credentialLoader.ts";
 import { generateLegacyProviders } from "./providerRegistry.ts";
@@ -18,6 +19,15 @@ export const FETCH_TIMEOUT_MS = upstreamTimeouts.fetchTimeoutMs;
 // idle for this duration. Override with STREAM_IDLE_TIMEOUT_MS env var.
 export const STREAM_IDLE_TIMEOUT_MS = upstreamTimeouts.streamIdleTimeoutMs;
 
+// Grace period (ms) a client-disconnect finalization waits for the stream's own
+// completion bookkeeping to land before persisting a 499. See #9653 — a client
+// that closes right after reading a fully-completed SSE stream can otherwise
+// race OmniRoute's own completion callback, resulting in a false 499 with zero
+// token usage for a request that actually delivered its full response. Set
+// STREAM_DISCONNECT_GRACE_PERIOD_MS=0 to disable and restore the old
+// immediate-fail behavior.
+export const STREAM_DISCONNECT_GRACE_PERIOD_MS = upstreamTimeouts.streamDisconnectGracePeriodMs;
+
 // Timeout for the first non-ping SSE event. Inherits REQUEST_TIMEOUT_MS when
 // set, unless STREAM_READINESS_TIMEOUT_MS is specified directly. This must stay
 // conservative for large prompts and slow first-byte reasoning providers.
@@ -65,27 +75,27 @@ export const PROVIDERS: Record<string, LegacyProvider> = new Proxy(
   {} as Record<string, LegacyProvider>,
   {
     get(_, prop) {
-      if (typeof prop === 'symbol') return undefined;
+      if (typeof prop === "symbol") return undefined;
       return Reflect.get(initProviders(), prop, _providers);
     },
     has(_, prop) {
-      if (typeof prop === 'symbol') return false;
+      if (typeof prop === "symbol") return false;
       return Reflect.has(initProviders(), prop);
     },
     ownKeys() {
       return Reflect.ownKeys(initProviders());
     },
     getOwnPropertyDescriptor(_, prop) {
-      if (typeof prop === 'symbol') return undefined;
+      if (typeof prop === "symbol") return undefined;
       return Object.getOwnPropertyDescriptor(initProviders(), prop);
     },
     set(_, prop, value) {
-      if (typeof prop === 'symbol') return false;
+      if (typeof prop === "symbol") return false;
       (initProviders() as Record<string, LegacyProvider>)[prop] = value;
       return true;
     },
     deleteProperty(_, prop) {
-      if (typeof prop === 'symbol') return false;
+      if (typeof prop === "symbol") return false;
       return Reflect.deleteProperty(initProviders(), prop);
     },
   }
@@ -124,6 +134,11 @@ export const OAUTH_ENDPOINTS = {
     auth: "https://github.com/login/oauth/authorize",
     deviceCode: "https://github.com/login/device/code",
   },
+  openference: {
+    token: "https://openference.com/oauth/token",
+    auth: "https://openference.com/app/oauth/authorize",
+    clientId: resolvePublicCred("openference_id"),
+  },
 };
 
 // Cache TTLs (seconds)

open-sse/config/embeddingRegistry.ts

@@ -409,6 +409,25 @@ export const EMBEDDING_PROVIDERS: Record<string, EmbeddingProvider> = {
       },
     ],
   },
+
+  nanogpt: {
+    id: "nanogpt",
+    baseUrl: "https://nano-gpt.com/v1/embeddings",
+    authType: "apikey",
+    authHeader: "bearer",
+    models: [
+      {
+        id: "text-embedding-3-small",
+        name: "Text Embedding 3 Small",
+        dimensions: 1536,
+      },
+      {
+        id: "text-embedding-3-large",
+        name: "Text Embedding 3 Large",
+        dimensions: 3072,
+      },
+    ],
+  },
 };
 
 const EMBEDDING_PROVIDER_ALIASES: Record<string, string> = {

open-sse/config/freeModelCatalog.data.ts

@@ -311,7 +311,6 @@ export const FREE_MODEL_BUDGETS: FreeModelBudget[] = [
   { provider: "nscale", modelId: "openai/gpt-oss-20b", displayName: "openai/gpt-oss-20b", monthlyTokens: 0, creditTokens: 5000000, freeType: "one-time-initial", poolKey: "nscale", tos: "caution" },
   { provider: "nscale", modelId: "meta-llama/Llama-4-Scout-17B-16E-Instruct", displayName: "meta-llama/Llama-4-Scout-17B-16E-Instruct", monthlyTokens: 0, creditTokens: 5000000, freeType: "one-time-initial", poolKey: "nscale", tos: "caution" },
   { provider: "nscale", modelId: "meta-llama/Llama-3.3-70B-Instruct", displayName: "meta-llama/Llama-3.3-70B-Instruct", monthlyTokens: 0, creditTokens: 5000000, freeType: "one-time-initial", poolKey: "nscale", tos: "caution" },
-  { provider: "nvidia", modelId: "z-ai/glm-5.1", displayName: "GLM 5.1", monthlyTokens: 0, creditTokens: 0, freeType: "one-time-initial", poolKey: "nvidia", tos: "caution" },
   { provider: "nvidia", modelId: "z-ai/glm-5.2", displayName: "GLM 5.2", monthlyTokens: 0, creditTokens: 0, freeType: "one-time-initial", poolKey: "nvidia", tos: "caution" },
   { provider: "nvidia", modelId: "minimaxai/minimax-m2.7", displayName: "MiniMax M2.7", monthlyTokens: 0, creditTokens: 0, freeType: "one-time-initial", poolKey: "nvidia", tos: "caution" },
   { provider: "nvidia", modelId: "google/gemma-4-31b-it", displayName: "Gemma 4 31B", monthlyTokens: 0, creditTokens: 0, freeType: "one-time-initial", poolKey: "nvidia", tos: "caution" },
@@ -321,7 +320,6 @@ export const FREE_MODEL_BUDGETS: FreeModelBudget[] = [
   { provider: "nvidia", modelId: "qwen/qwen3.5-397b-a17b", displayName: "Qwen3.5-397B-A17B", monthlyTokens: 0, creditTokens: 0, freeType: "one-time-initial", poolKey: "nvidia", tos: "caution" },
   { provider: "nvidia", modelId: "qwen/qwen3.5-122b-a10b", displayName: "Qwen3.5-122B-A10B", monthlyTokens: 0, creditTokens: 0, freeType: "one-time-initial", poolKey: "nvidia", tos: "caution" },
   { provider: "nvidia", modelId: "stepfun-ai/step-3.5-flash", displayName: "Step 3.5 Flash", monthlyTokens: 0, creditTokens: 0, freeType: "one-time-initial", poolKey: "nvidia", tos: "caution" },
-  { provider: "nvidia", modelId: "deepseek-ai/deepseek-v4-pro", displayName: "DeepSeek V4 Pro", monthlyTokens: 0, creditTokens: 0, freeType: "one-time-initial", poolKey: "nvidia", tos: "caution" },
   { provider: "nvidia", modelId: "openai/gpt-oss-120b", displayName: "GPT OSS 120B", monthlyTokens: 0, creditTokens: 0, freeType: "one-time-initial", poolKey: "nvidia", tos: "caution" },
   { provider: "nvidia", modelId: "openai/gpt-oss-20b", displayName: "GPT OSS 20B", monthlyTokens: 0, creditTokens: 0, freeType: "one-time-initial", poolKey: "nvidia", tos: "caution" },
   { provider: "nvidia", modelId: "nvidia/nemotron-3-super-120b-a12b", displayName: "Nemotron 3 Super 120B A12B", monthlyTokens: 0, creditTokens: 0, freeType: "one-time-initial", poolKey: "nvidia", tos: "caution" },

open-sse/config/nvidiaHostedModels.snapshot.json

@@ -1,5 +1,4 @@
 [
-  "deepseek-ai/deepseek-v4-pro",
   "google/gemma-4-31b-it",
   "minimaxai/minimax-m2.7",
   "mistralai/devstral-2-123b-instruct-2512",
@@ -13,6 +12,5 @@
   "qwen/qwen3.5-397b-a17b",
   "stepfun-ai/step-3.5-flash",
   "thinkingmachines/inkling",
-  "z-ai/glm-5.1",
   "z-ai/glm-5.2"
 ]

open-sse/config/providerModels.ts

@@ -90,10 +90,65 @@ export function getDefaultModel(aliasOrId: string): string | null {
   return models?.[0]?.id || null;
 }
 
+/** Score a registry entry by how many capability flags it defines. */
+function modelRichness(m: RegistryModel): number {
+  let score = 0;
+  if (m.supportsXHighEffort !== undefined) score += 10; // critical for effort routing
+  if (m.supportsReasoning !== undefined) score += 5;
+  if (m.contextLength !== undefined) score += 3;
+  if (m.maxOutputTokens !== undefined) score += 2;
+  if (m.supportsVision !== undefined) score += 2;
+  if (m.toolCalling !== undefined) score += 2;
+  if (m.interleavedField !== undefined) score += 1;
+  if (m.unsupportedParams !== undefined) score += 1;
+  return score;
+}
+
+function getGlobalModel(modelId: string): RegistryModel | undefined {
+  // 1. Exact match — collect all, pick the richest
+  let candidates: RegistryModel[] = [];
+  for (const models of Object.values(PROVIDER_MODELS)) {
+    const found = models.find((m) => m.id === modelId);
+    if (found) candidates.push(found);
+  }
+  if (candidates.length > 0) {
+    return candidates.sort((a, b) => modelRichness(b) - modelRichness(a))[0];
+  }
+
+  // 2. Strip provider prefix (e.g. moonshotai/kimi-k3-free -> kimi-k3-free)
+  const basename = modelId.split("/").pop() || modelId;
+  candidates = [];
+  for (const models of Object.values(PROVIDER_MODELS)) {
+    const found = models.find((m) => m.id === basename);
+    if (found) candidates.push(found);
+  }
+  if (candidates.length > 0) {
+    return candidates.sort((a, b) => modelRichness(b) - modelRichness(a))[0];
+  }
+
+  // 3. Substring match for base model name (e.g. kimi-k3-free -> kimi-k3)
+  // Finds the longest matching base model ID; on ties, prefers the richer entry.
+  let bestMatch: RegistryModel | undefined;
+  for (const models of Object.values(PROVIDER_MODELS)) {
+    for (const m of models) {
+      if (basename.startsWith(m.id)) {
+        if (
+          !bestMatch ||
+          m.id.length > bestMatch.id.length ||
+          (m.id.length === bestMatch.id.length && modelRichness(m) > modelRichness(bestMatch))
+        ) {
+          bestMatch = m;
+        }
+      }
+    }
+  }
+  return bestMatch;
+}
+
 export function getProviderModel(aliasOrId: string, modelId: string): RegistryModel | undefined {
   const models = PROVIDER_MODELS[aliasOrId];
-  if (!models) return undefined;
-  return models.find((model) => model.id === modelId);
+  if (!models) return getGlobalModel(modelId);
+  return models.find((model) => model.id === modelId) || getGlobalModel(modelId);
 }
 
 export function isValidModel(
@@ -103,26 +158,20 @@ export function isValidModel(
 ): boolean {
   if (passthroughProviders.has(aliasOrId)) return true;
   const models = PROVIDER_MODELS[aliasOrId];
-  if (!models) return false;
-  return models.some((m) => m.id === modelId);
+  if (!models) return !!getGlobalModel(modelId);
+  return models.some((m) => m.id === modelId) || !!getGlobalModel(modelId);
 }
 
 export function findModelName(aliasOrId: string, modelId: string): string {
   const models = PROVIDER_MODELS[aliasOrId];
-  if (!models) return modelId;
-  const found = models.find((m) => m.id === modelId);
+  if (!models) return getGlobalModel(modelId)?.name || modelId;
+  const found = models.find((m) => m.id === modelId) || getGlobalModel(modelId);
   return found?.name || modelId;
 }
 
 export function getModelTargetFormat(aliasOrId: string, modelId: string): string | null {
   const models = PROVIDER_MODELS[aliasOrId];
-  // Strip provider prefix if present: "openai/gpt-5.6-luna" → "gpt-5.6-luna"
-  const prefix = aliasOrId + "/";
-  const bareModelId =
-    typeof modelId === "string" && modelId.startsWith(prefix)
-      ? modelId.slice(prefix.length)
-      : modelId;
-  const found = models?.find((m) => m.id === bareModelId);
+  const found = models?.find((m) => m.id === modelId) || getGlobalModel(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
@@ -130,14 +179,17 @@ export function getModelTargetFormat(aliasOrId: string, modelId: string): string
   // 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$/i.test(bareModelId)) return "openai-responses";
+  if (aliasOrId === "openai" && /-pro$/i.test(modelId)) return "openai-responses";
   return null;
 }
 
 export function getModelStripTypes(aliasOrId: string, modelId: string): string[] {
   const models = PROVIDER_MODELS[aliasOrId];
-  if (!models) return [];
-  const found = models.find((m) => m.id === modelId);
+  if (!models)
+    return Array.isArray(getGlobalModel(modelId)?.strip)
+      ? [...getGlobalModel(modelId)!.strip!]
+      : [];
+  const found = models.find((m) => m.id === modelId) || getGlobalModel(modelId);
   return Array.isArray(found?.strip) ? [...found.strip] : [];
 }
 
@@ -262,7 +314,7 @@ function resolveProviderModelList(aliasOrId: string): {
 
 export function supportsXHighEffort(aliasOrId: string, modelId: string): boolean {
   const { models: providerModels } = resolveProviderModelList(aliasOrId);
-  const model = providerModels?.find((entry) => entry.id === modelId);
+  const model = providerModels?.find((entry) => entry.id === modelId) || getGlobalModel(modelId);
   if (model?.supportsXHighEffort !== undefined) {
     return model.supportsXHighEffort !== false;
   }

open-sse/config/providers/index.ts

@@ -121,6 +121,8 @@ import { chatgpt_webProvider } from "./registry/chatgpt-web/index.ts";
 import { openrouterProvider } from "./registry/openrouter/index.ts";
 import { cheaperinferenceProvider } from "./registry/cheaperinference/index.ts";
 import { openvectaProvider } from "./registry/openvecta/index.ts";
+import { openferenceProvider } from "./registry/openference/index.ts";
+import { openference_apiProvider } from "./registry/openference-api/index.ts";
 import { orcarouterProvider } from "./registry/orcarouter/index.ts";
 import { copilot_webProvider } from "./registry/copilot-web/index.ts";
 import { copilot_m365_webProvider } from "./registry/copilot-m365-web/index.ts";
@@ -345,6 +347,8 @@ export const REGISTRY: Record<string, RegistryEntry> = {
   openrouter: openrouterProvider,
   cheaperinference: cheaperinferenceProvider,
   openvecta: openvectaProvider,
+  openference: openferenceProvider,
+  "openference-api": openference_apiProvider,
   orcarouter: orcarouterProvider,
   "copilot-web": copilot_webProvider,
   "copilot-m365-web": copilot_m365_webProvider,

open-sse/config/providers/registry/nanogpt/index.ts

@@ -8,6 +8,7 @@ export const nanogptProvider: RegistryEntry = {
   executor: "default",
   baseUrl: "https://nano-gpt.com/api/v1/chat/completions",
   modelsUrl: "https://nano-gpt.com/api/v1/models",
+  responsesBaseUrl: "https://nano-gpt.com/api/v1/responses",
   authType: "apikey",
   authHeader: "bearer",
   models: CHAT_OPENAI_COMPAT_MODELS.nanogpt,

open-sse/config/providers/registry/nvidia/index.ts

@@ -32,8 +32,6 @@ export const nvidiaProvider: RegistryEntry = {
     { id: "qwen/qwen3.5-122b-a10b", name: "Qwen3.5-122B-A10B" },
     { id: "stepfun-ai/step-3.5-flash", name: "Step 3.5 Flash" },
     { id: "stepfun-ai/step-3.7-flash", name: "Step 3.7 Flash" },
-    { id: "deepseek-ai/deepseek-v4-pro", name: "DeepSeek V4 Pro", supportsReasoning: true },
-    { id: "deepseek-ai/deepseek-v4-flash", name: "DeepSeek V4 Flash", supportsReasoning: true },
     // Sweep 2026-06-19: verified present in the live NVIDIA NIM /v1/models catalog.
     { id: "moonshotai/kimi-k2.6", name: "Kimi K2.6" },
     { id: "openai/gpt-oss-120b", name: "GPT OSS 120B", toolCalling: false },

open-sse/config/providers/registry/openference-api/index.ts

@@ -0,0 +1,18 @@
+import type { RegistryEntry } from "../../shared.ts";
+import { buildOpenAiCompatibleRegistryEntry } from "../../shared.ts";
+
+/**
+ * Openference API key — OpenAI-compatible gateway (https://openference.com/).
+ *
+ * Bearer API keys (`sk-…`) hit the same api.openference.com/v1/* surface as OAuth
+ * JWTs. Live model discovery uses NAMED_OPENAI_STYLE_PROVIDERS; the seed below is
+ * the offline fallback when the live fetch fails.
+ */
+export const openference_apiProvider: RegistryEntry = buildOpenAiCompatibleRegistryEntry({
+  id: "openference-api",
+  alias: "ofa",
+  baseUrl: "https://api.openference.com/v1/chat/completions",
+  responsesBaseUrl: "https://api.openference.com/v1/responses",
+  passthroughModels: true,
+  models: [{ id: "GLM-5.2", name: "GLM 5.2", contextLength: 850000 }],
+});

open-sse/config/providers/registry/openference/index.ts

@@ -0,0 +1,25 @@
+import { resolvePublicCred, type RegistryEntry } from "../../shared.ts";
+
+/**
+ * Openference — OpenAI-compatible AI inference gateway (https://openference.com/).
+ *
+ * OAuth access tokens are ES256 JWTs accepted as Bearer credentials on
+ * api.openference.com/v1/*. Live model discovery uses NAMED_OPENAI_STYLE_PROVIDERS;
+ * seed models below are the offline fallback when the live fetch fails.
+ */
+export const openferenceProvider: RegistryEntry = {
+  id: "openference",
+  alias: "of",
+  format: "openai",
+  executor: "default",
+  baseUrl: "https://api.openference.com/v1/chat/completions",
+  responsesBaseUrl: "https://api.openference.com/v1/responses",
+  authType: "oauth",
+  authHeader: "bearer",
+  passthroughModels: true,
+  oauth: {
+    clientIdDefault: resolvePublicCred("openference_id"),
+    tokenUrl: "https://openference.com/oauth/token",
+  },
+  models: [{ id: "GLM-5.2", name: "GLM 5.2", contextLength: 850000 }],
+};

open-sse/config/videoRegistry.ts

@@ -339,6 +339,15 @@ export const VIDEO_PROVIDERS: Record<string, VideoProvider> = {
     format: "adobe-firefly-video",
     models: toRegistryVideoModels(),
   },
+
+  nanogpt: {
+    id: "nanogpt",
+    baseUrl: "https://nano-gpt.com/api/v1/video/generations",
+    authType: "apikey",
+    authHeader: "bearer",
+    format: "openai",
+    models: [{ id: "default", name: "NanoGPT Video" }],
+  },
 };
 
 /**

open-sse/executors/azure-ai.ts

@@ -0,0 +1,35 @@
+import { DefaultExecutor } from "./default.ts";
+import type { ProviderCredentials } from "./base.ts";
+import { applyAzureParamRules } from "./azureParamRules.ts";
+
+/**
+ * Azure AI Foundry (`azure-ai`).
+ *
+ * URL building, auth headers and the `responses` vs `chat` apiType switch all
+ * live in `DefaultExecutor`, keyed on the `azure-ai` provider id — this subclass
+ * inherits them unchanged and adds only the Azure request-param rules.
+ *
+ * Before this existed, `azure-ai` fell through to the bare `DefaultExecutor`
+ * while `azure-openai` had the rules inline, so the same Azure deployment
+ * behaved differently depending on which connection served it: `azure-openai`
+ * succeeded and `azure-ai` returned HTTP 400 for `max_tokens` /
+ * `reasoning_effort`.
+ */
+export class AzureAiExecutor extends DefaultExecutor {
+  constructor() {
+    super("azure-ai");
+  }
+
+  override transformRequest(
+    model: string,
+    body: unknown,
+    stream: boolean,
+    credentials: ProviderCredentials
+  ): unknown {
+    return applyAzureParamRules(
+      model,
+      body,
+      super.transformRequest(model, body, stream, credentials)
+    );
+  }
+}

open-sse/executors/azure-openai.ts

@@ -1,9 +1,9 @@
 import { DefaultExecutor } from "./default.ts";
 import type { ProviderCredentials } from "./base.ts";
 import { stripTrailingSlashes } from "../utils/urlSanitize.ts";
+import { applyAzureParamRules } from "./azureParamRules.ts";
 
 const DEFAULT_API_VERSION = "2024-12-01-preview";
-const GPT5_OR_REASONING_DEPLOYMENT = /(?:^|[/_-])(?:gpt-5|o(?:1|3|4))(?:[._-]|$)/i;
 
 function normalizeAzureBaseUrl(rawBaseUrl?: string | null): string {
   const normalized = stripTrailingSlashes((rawBaseUrl || "").trim());
@@ -57,37 +57,10 @@ export class AzureOpenAIExecutor extends DefaultExecutor {
     stream: boolean,
     credentials: ProviderCredentials
   ): unknown {
-    const transformed = super.transformRequest(model, body, stream, credentials);
-    if (!GPT5_OR_REASONING_DEPLOYMENT.test(model)) return transformed;
-    if (!transformed || typeof transformed !== "object" || Array.isArray(transformed)) {
-      return transformed;
-    }
-
-    const original =
-      body && typeof body === "object" && !Array.isArray(body)
-        ? (body as Record<string, unknown>)
-        : null;
-    const normalized = { ...(transformed as Record<string, unknown>) };
-
-    if (original?.max_completion_tokens !== undefined) {
-      normalized.max_completion_tokens = original.max_completion_tokens;
-    } else if (
-      normalized.max_completion_tokens === undefined &&
-      original?.max_tokens !== undefined
-    ) {
-      normalized.max_completion_tokens = original.max_tokens;
-    }
-    delete normalized.max_tokens;
-
-    if (normalized.temperature !== undefined && normalized.temperature !== 1) {
-      delete normalized.temperature;
-    }
-
-    const hasTools = Array.isArray(normalized.tools) && normalized.tools.length > 0;
-    if (hasTools || normalized.reasoning_effort === "none") {
-      delete normalized.reasoning_effort;
-    }
-
-    return normalized;
+    return applyAzureParamRules(
+      model,
+      body,
+      super.transformRequest(model, body, stream, credentials)
+    );
   }
 }

open-sse/executors/azureParamRules.ts

@@ -0,0 +1,76 @@
+/**
+ * Azure Chat Completions param rules, shared by every Azure wire path.
+ *
+ * Azure's newer deployments reject a handful of stock OpenAI Chat Completions
+ * params and return HTTP 400 rather than ignoring them:
+ *
+ *   - `max_tokens`      -> "Unsupported parameter: 'max_tokens' is not supported
+ *                          with this model. Use 'max_completion_tokens' instead."
+ *   - `temperature`     -> only the default (1) is accepted.
+ *   - `reasoning_effort` -> "Function tools with reasoning_effort are not
+ *                          supported ... Please use /v1/responses instead."
+ *
+ * This logic previously lived inline in `AzureOpenAIExecutor`, so it only
+ * covered the `azure-openai` provider. `azure-ai` (Azure AI Foundry) routes
+ * through `DefaultExecutor` and inherited none of it, which meant an identical
+ * deployment 400'd on one connection and succeeded on the other. Extracted here
+ * so both executors apply exactly the same rules.
+ */
+
+/**
+ * Deployments that require `max_completion_tokens` instead of `max_tokens`.
+ *
+ * Matches the GPT-5 family and the o1/o3/o4 reasoning series at a token
+ * boundary, so a deployment named `my-gpt-5-prod` matches while an unrelated
+ * `piston-o4-legacy`-style name does not match by accident. `gpt-chat-latest`
+ * is listed explicitly: it is a moving alias that currently resolves to a
+ * GPT-5-era model and rejects `max_tokens`, but carries no version number for
+ * the boundary pattern to key on.
+ */
+export const AZURE_COMPLETION_TOKEN_DEPLOYMENT =
+  /(?:^|[/_-])(?:gpt-5|o(?:1|3|4))(?:[._-]|$)|^gpt-chat-latest$/i;
+
+/**
+ * Apply the Azure param rules to an already-translated Chat Completions body.
+ *
+ * `originalBody` is the pre-translation request, consulted only to recover a
+ * caller-supplied token budget that translation may have moved or dropped.
+ * Returns `transformed` untouched when the deployment is unaffected or the body
+ * is not a plain object, and never mutates either input.
+ */
+export function applyAzureParamRules(
+  model: string,
+  originalBody: unknown,
+  transformed: unknown
+): unknown {
+  if (!AZURE_COMPLETION_TOKEN_DEPLOYMENT.test(model)) return transformed;
+  if (!transformed || typeof transformed !== "object" || Array.isArray(transformed)) {
+    return transformed;
+  }
+
+  const original =
+    originalBody && typeof originalBody === "object" && !Array.isArray(originalBody)
+      ? (originalBody as Record<string, unknown>)
+      : null;
+  const normalized = { ...(transformed as Record<string, unknown>) };
+
+  if (original?.max_completion_tokens !== undefined) {
+    normalized.max_completion_tokens = original.max_completion_tokens;
+  } else if (normalized.max_completion_tokens === undefined && original?.max_tokens !== undefined) {
+    normalized.max_completion_tokens = original.max_tokens;
+  }
+  delete normalized.max_tokens;
+
+  if (normalized.temperature !== undefined && normalized.temperature !== 1) {
+    delete normalized.temperature;
+  }
+
+  // Azure 400s on reasoning_effort as soon as tools are present, which is every
+  // agentic client (Claude Code, Cursor agent) on every turn.
+  const hasTools = Array.isArray(normalized.tools) && normalized.tools.length > 0;
+  if (hasTools || normalized.reasoning_effort === "none") {
+    delete normalized.reasoning_effort;
+  }
+
+  return normalized;
+}

open-sse/executors/base/reasoningEffort.ts

@@ -2,16 +2,19 @@
 // Extracted verbatim from base.ts. Deps are config/services only (no host import → no cycle).
 import { PROVIDER_CLAUDE } from "../../services/systemTransforms.ts";
 import { isClaudeCodeCompatible } from "../../services/provider.ts";
-import { supportsClaudeMaxEffort, supportsXHighEffort } from "../../config/providerModels.ts";
+import {
+  supportsClaudeMaxEffort,
+  supportsXHighEffort,
+  getProviderModel,
+} from "../../config/providerModels.ts";
 
 /**
  * Sanitize reasoning_effort for providers that don't accept all values.
  *
- * The claude→openai translator passes output_config.effort through verbatim
- * (including max) and only performs form conversion; provider-aware effort
- * policy is owned here. Combined with runtime alias remapping (e.g.
- * claude-opus-4-6 → mimo/mimo-v2.5-pro), this routes a client's effort value
- * to OpenAI-shape providers that don't accept it:
+ * The claude→openai translator may emit reasoning_effort=max/xhigh when the
+ * client sends output_config.effort=max on a Claude-shape request. Combined with
+ * runtime alias remapping (e.g. claude-opus-4-6 → mimo/mimo-v2.5-pro), this
+ * routes xhigh to OpenAI-shape providers that don't accept the value:
  *
  *   xiaomi-mimo : low|medium|high only — 400 literal_error on xhigh
  *   mistral     : devstral models reject reasoning_effort entirely
@@ -140,9 +143,11 @@ export function mapNvidiaGlm52ReasoningParams(
 }
 
 export function supportsMaxEffortForProvider(provider: string, model: string): boolean {
+  const resolvedModelId = getProviderModel(provider, model)?.id || model;
+
   const isClaude =
     (provider === PROVIDER_CLAUDE || isClaudeCodeCompatible(provider)) &&
-    supportsClaudeMaxEffort(model);
+    supportsClaudeMaxEffort(resolvedModelId);
   // opencode-go proxies DeepSeek with the native DeepSeek API contract, which
   // accepts {high, max} literally. Without this opt-in, max would be
   // normalized to xhigh (the OmniRoute-internal top tier) and rejected by the
@@ -151,11 +156,12 @@ export function supportsMaxEffortForProvider(provider: string, model: string): b
   // Ollama Cloud also accepts literal max (for example GLM 5.2 supports
   // low|medium|high|max|none) and rejects xhigh.
   const isOpencodeGoDeepSeek =
-    (provider === "opencode-go" || provider === "opencode-zen") &&
-    model.toLowerCase().includes("deepseek");
+    provider === "opencode-go" && resolvedModelId.toLowerCase().includes("deepseek");
   const isOllamaCloud = provider === "ollama-cloud";
-  const isMoonshotK3 =
-    (provider === "moonshot" || provider === "kimi") && /^kimi-k3(?:$|-)/i.test(model);
+  // Kimi K3 only accepts literal max and rejects xhigh natively. Apply this mapping
+  // regardless of provider so that OpenAI-compatible proxies (e.g. TokenRouter)
+  // correctly pass max instead of the internal xhigh top tier.
+  const isMoonshotK3 = /^kimi-k3(?:$|-)/i.test(resolvedModelId);
   return isClaude || isOpencodeGoDeepSeek || isOllamaCloud || isMoonshotK3;
 }
 
@@ -253,16 +259,6 @@ export function sanitizeReasoningEffortForProvider(
   const effortStr = typeof c.effort === "string" ? c.effort.toLowerCase() : "";
   const modelStr = model || "";
 
-  // Oh My Pi exposes `minimal`, while Codex's Responses API starts at `low`.
-  // Normalize every carrier before the Codex executor sends the upstream request.
-  if (provider === "codex" && effortStr === "minimal") {
-    log?.info?.(
-      "REASONING_SANITIZE",
-      `${provider}/${modelStr}: normalized reasoning_effort minimal → low`
-    );
-    return writeEffortValue(b, "low", c);
-  }
-
   const githubOptIn =
     provider === "github" && GITHUB_REASONING_EFFORT_OPT_IN_PATTERN.test(modelStr);
   const rejecting =
@@ -298,27 +294,48 @@ export function sanitizeReasoningEffortForProvider(
   }
 
   const supportsXHigh = supportsXHighEffort(provider, modelStr);
-  const shouldDowngradeXHigh = effortStr === "xhigh" && !supportsXHigh;
-  const supportsXHighForMax = supportsXHigh;
   const supportsMax = supportsMaxEffortForProvider(provider, modelStr);
-  const shouldNormalizeMaxToXHigh = effortStr === "max" && !supportsMax && supportsXHighForMax;
-  const shouldDowngradeMax = effortStr === "max" && !supportsMax && !supportsXHighForMax;
 
-  if (shouldNormalizeMaxToXHigh) {
+  // ── xhigh handling ──────────────────────────────────────────────────────
+  // xhigh is OmniRoute-internal. Map it to the best effort the model accepts.
+  if (effortStr === "xhigh") {
+    if (supportsXHigh) return body; // model accepts xhigh natively
+    if (supportsMax) {
+      log?.info?.(
+        "REASONING_SANITIZE",
+        `${provider}/${modelStr}: mapped reasoning_effort xhigh → max`
+      );
+      return writeEffortValue(b, "max", c);
+    }
+    // Model explicitly rejects xhigh — gracefully degrade to high (its highest standard tier)
     log?.info?.(
       "REASONING_SANITIZE",
-      `${provider}/${modelStr}: normalized reasoning_effort max → xhigh`
-    );
-    return writeEffortValue(b, "xhigh", c);
-  }
-
-  if (shouldDowngradeXHigh || shouldDowngradeMax) {
-    log?.info?.(
-      "REASONING_SANITIZE",
-      `${provider}/${modelStr}: downgraded reasoning_effort ${effortStr} → high`
+      `${provider}/${modelStr}: downgraded reasoning_effort xhigh → high`
     );
     return writeEffortValue(b, "high", c);
   }
 
+  // ── max handling ────────────────────────────────────────────────────────
+  // NEW DEFAULT: pass max through unchanged. Most reasoning-capable APIs
+  // accept max natively. Only degrade when we KNOW the model rejects it
+  // (registry has supportsXHighEffort explicitly set to false AND it's not
+  // in the supportsMax whitelist). Unknown models pass through — trust the
+  // upstream, and if it 400s the user gets a clear signal. This prevents
+  // new models from being unusable for weeks until they're whitelisted (#8057).
+  if (effortStr === "max") {
+    if (supportsMax) return body; // explicitly known to accept max
+    if (!supportsXHigh) {
+      // Model is explicitly flagged as rejecting xhigh (and not in supportsMax) —
+      // it likely only accepts standard tiers. Degrade to its highest: high.
+      log?.info?.(
+        "REASONING_SANITIZE",
+        `${provider}/${modelStr}: downgraded reasoning_effort max → high (model rejects max/xhigh)`
+      );
+      return writeEffortValue(b, "high", c);
+    }
+    // Default: pass max through unchanged — trust the upstream
+    return body;
+  }
+
   return body;
 }

open-sse/executors/chatgpt-web.ts

@@ -2822,8 +2822,7 @@ export class ChatGptWebExecutor extends BaseExecutor {
     const modelSlug = resolvedModel.slug;
     const { hasTools, requestedTools, effectiveMessages } = prepareToolMessages(
       (body || {}) as Record<string, unknown>,
-      messages as Array<{ role: string; content: unknown }>,
-      { hardened: isThinkingCapableModel(model, modelSlug) }
+      messages as Array<{ role: string; content: unknown }>
     );
 
     if (!credentials.apiKey) {

open-sse/executors/cliproxyapi.ts

@@ -408,12 +408,13 @@ export class CliproxyapiExecutor extends BaseExecutor {
 
     input.log?.info?.("CPA", `CLIProxyAPI → ${url} (model: ${input.model}, shape: ${shape})`);
 
-    // _toolNameMap is an in-memory channel to chatCore for response-side
-    // tool name restoration; never send it over the wire.
+    // _toolNameMap and _namespaceToolIdentityMap are in-memory channels to
+    // chatCore for response-side tool name restoration; never send them over
+    // the wire.
     const wireBody =
       transformedBody && typeof transformedBody === "object"
         ? JSON.stringify(transformedBody, (key, value) =>
-            key === "_toolNameMap" ? undefined : value
+            key === "_toolNameMap" || key === "_namespaceToolIdentityMap" ? undefined : value
           )
         : JSON.stringify(transformedBody);
 

open-sse/executors/codebuddy-cn.ts

@@ -1,5 +1,82 @@
 import { DefaultExecutor } from "./default.ts";
-import type { ProviderCredentials } from "./base.ts";
+import type { ExecuteInput, ExecutorExecuteResult, ProviderCredentials } from "./base.ts";
+
+const SENSITIVE_CONTENT_REJECTION =
+  "抱歉，系统检测到您当前输入的信息存在敏感内容，我无法响应您的请求，请检查后重新输入";
+const LARGE_TOOL_METADATA_BYTES = 64 * 1024;
+
+function responseFromResult(result: ExecutorExecuteResult): Response {
+  return result instanceof Response ? result : result.response;
+}
+
+function credentialsFromResult(
+  result: ExecutorExecuteResult,
+  fallback: ProviderCredentials
+): ProviderCredentials {
+  if (result instanceof Response || !result.headers) return fallback;
+
+  const authorization = Object.entries(result.headers).find(
+    ([name]) => name.toLowerCase() === "authorization"
+  )?.[1];
+  if (!authorization?.startsWith("Bearer ")) return fallback;
+
+  return {
+    ...fallback,
+    accessToken: authorization.slice("Bearer ".length),
+    expiresAt: undefined,
+  };
+}
+
+function compactToolDescriptions(body: unknown): unknown | null {
+  if (!body || typeof body !== "object" || Array.isArray(body)) return null;
+
+  const request = body as Record<string, unknown>;
+  if (!Array.isArray(request.tools) || request.tools.length === 0) return null;
+
+  const originalTools = request.tools;
+  try {
+    const serializedTools = JSON.stringify(originalTools);
+    if (new TextEncoder().encode(serializedTools).byteLength < LARGE_TOOL_METADATA_BYTES) {
+      return null;
+    }
+  } catch {
+    return null;
+  }
+
+  let tools: unknown[] | null = null;
+  originalTools.forEach((tool, index) => {
+    if (!tool || typeof tool !== "object" || Array.isArray(tool)) return;
+
+    const declaration = tool as Record<string, unknown>;
+    if (
+      declaration.type !== "function" ||
+      !declaration.function ||
+      typeof declaration.function !== "object" ||
+      Array.isArray(declaration.function)
+    ) {
+      return;
+    }
+
+    const toolFunction = declaration.function as Record<string, unknown>;
+    if (!Object.prototype.hasOwnProperty.call(toolFunction, "description")) return;
+
+    const compactFunction = { ...toolFunction };
+    delete compactFunction.description;
+    tools ??= originalTools.slice();
+    tools[index] = { ...declaration, function: compactFunction };
+  });
+
+  return tools ? { ...request, tools } : null;
+}
+
+async function isSensitiveContentRejection(response: Response): Promise<boolean> {
+  if (response.status !== 400) return false;
+  const responseText = await response
+    .clone()
+    .text()
+    .catch(() => "");
+  return responseText.includes(SENSITIVE_CONTENT_REJECTION);
+}
 
 /**
  * CodeBuddyCnExecutor — talks to https://copilot.tencent.com/v2/chat/completions
@@ -21,6 +98,26 @@ export class CodeBuddyCnExecutor extends DefaultExecutor {
     super("codebuddy-cn");
   }
 
+  async execute(input: ExecuteInput): Promise<ExecutorExecuteResult> {
+    const result = await super.execute(input);
+    if (!(await isSensitiveContentRejection(responseFromResult(result)))) {
+      return result;
+    }
+
+    const compactBody = compactToolDescriptions(input.body);
+    if (!compactBody) return result;
+
+    input.log?.debug?.(
+      "CODEBUDDY_CN",
+      "Upstream rejected an oversized tool request as sensitive content; retrying with compact tool descriptions"
+    );
+    return super.execute({
+      ...input,
+      body: compactBody,
+      credentials: credentialsFromResult(result, input.credentials),
+    });
+  }
+
   transformRequest(
     model: string,
     body: unknown,

open-sse/executors/codex/tools.ts

@@ -30,6 +30,114 @@ export function isCodexFreePlan(providerSpecificData: unknown): boolean {
   return typeof plan === "string" && plan.trim().toLowerCase() === "free";
 }
 
+type JsonRecord = Record<string, unknown>;
+
+const REDUNDANT_ONEOF_OBJECT_MAP_FIELDS = [
+  "properties",
+  "patternProperties",
+  "$defs",
+  "definitions",
+] as const;
+
+const REDUNDANT_ONEOF_ARRAY_SCHEMA_FIELDS = ["prefixItems", "oneOf", "anyOf", "allOf"] as const;
+
+const REDUNDANT_ONEOF_SINGLE_SCHEMA_FIELDS = [
+  "items",
+  "additionalProperties",
+  "not",
+  "if",
+  "then",
+  "else",
+] as const;
+
+const REDUNDANT_ONEOF_ANNOTATION_KEYS = new Set(["const", "description", "title", "$comment"]);
+
+/**
+ * Remove a redundant `oneOf` when it is fully covered by a sibling `enum`.
+ *
+ * The Codex private Responses endpoint (`chatgpt.com/backend-api/codex/responses`)
+ * intermittently returns a 502 `upstream_empty_response` when a tool parameter
+ * carries the JSON-Schema pattern `oneOf: [{const, ...annotations}]` together
+ * with a sibling `enum` whose value set exactly matches the `const` set. In that
+ * case `oneOf` adds no constraint beyond `enum`, so dropping it is semantically
+ * safe and eliminates the trigger.
+ *
+ * Only the exact-match redundant case is stripped. Bare `oneOf[const]` without
+ * a sibling `enum`, narrowing const sets, non-matching enums, type-discriminated
+ * `oneOf`, and `anyOf`/`allOf` are all preserved.
+ */
+export function stripRedundantOneOfConstEnum(schema: unknown): unknown {
+  if (Array.isArray(schema)) {
+    return schema.map((entry) => stripRedundantOneOfConstEnum(entry));
+  }
+  if (!isPlainObject(schema)) return schema;
+
+  const result: JsonRecord = { ...schema };
+
+  maybeStripRedundantOneOf(result);
+
+  for (const field of REDUNDANT_ONEOF_OBJECT_MAP_FIELDS) {
+    const map = result[field];
+    if (isPlainObject(map)) {
+      result[field] = Object.fromEntries(
+        Object.entries(map).map(([key, value]) => [key, stripRedundantOneOfConstEnum(value)])
+      );
+    }
+  }
+
+  for (const field of REDUNDANT_ONEOF_ARRAY_SCHEMA_FIELDS) {
+    if (Array.isArray(result[field])) {
+      result[field] = (result[field] as unknown[]).map((entry) =>
+        stripRedundantOneOfConstEnum(entry)
+      );
+    }
+  }
+
+  for (const field of REDUNDANT_ONEOF_SINGLE_SCHEMA_FIELDS) {
+    if (result[field] !== undefined) {
+      result[field] = stripRedundantOneOfConstEnum(result[field]);
+    }
+  }
+
+  return result;
+}
+
+function maybeStripRedundantOneOf(node: JsonRecord): void {
+  const branches = node.oneOf;
+  if (!Array.isArray(branches) || branches.length === 0) return;
+
+  const enumValues = Array.isArray(node.enum) ? node.enum : null;
+  if (!enumValues || enumValues.length === 0) return;
+
+  // Every branch must be {const, ...annotations only}.
+  const constValues: unknown[] = [];
+  for (const branch of branches) {
+    if (!isPlainObject(branch)) return;
+    const branchKeys = Object.keys(branch);
+    if (!branchKeys.includes("const")) return;
+    if (!branchKeys.every((key) => REDUNDANT_ONEOF_ANNOTATION_KEYS.has(key))) return;
+    constValues.push((branch as JsonRecord).const);
+  }
+
+  // Restrict to string consts and string enums (confirmed production shape).
+  if (!constValues.every((value) => typeof value === "string")) return;
+  if (!enumValues.every((value) => typeof value === "string")) return;
+
+  // All const values must be unique.
+  if (new Set(constValues).size !== constValues.length) return;
+
+  // The const set must exactly match the enum set.
+  const enumSet = new Set(enumValues);
+  if (enumSet.size !== constValues.length) return;
+  if (!constValues.every((value) => enumSet.has(value))) return;
+
+  delete node.oneOf;
+}
+
+function isPlainObject(value: unknown): value is JsonRecord {
+  return typeof value === "object" && value !== null && !Array.isArray(value);
+}
+
 export function normalizeCodexTools(
   body: Record<string, unknown>,
   options?: { dropImageGeneration?: boolean; preserveCustomTools?: boolean }
@@ -138,7 +246,9 @@ export function normalizeCodexTools(
     // Codex/OpenAI Responses API rejects `pattern` fields using regex lookaround
     // (e.g. `^(?=.*@).+$`) with a 400 "regex lookaround is not supported" error.
     // Strip those before the schema reaches upstream (9router#1556).
-    const sanitizedParameters = stripUnsupportedRegexPatterns(parameters);
+    const sanitizedParameters = stripRedundantOneOfConstEnum(
+      stripUnsupportedRegexPatterns(parameters)
+    );
 
     // Rewrite in-place to Responses format
     for (const key of Object.keys(tool)) {

open-sse/executors/commandCode.ts

@@ -48,6 +48,21 @@ function recordOrEmpty(value: unknown): JsonRecord {
   return {};
 }
 
+/**
+ * Build the `arguments` field for an assistant tool-call part that Command
+ * Code's /alpha/generate schema REQUIRES (rejects a missing field with
+ * `missing required field 'arguments'`). Valid source values round-trip:
+ *   - object arguments  -> JSON string of the object
+ *   - string arguments  -> the string as-is (already valid JSON)
+ *   - missing / empty / invalid JSON -> "{}" (a valid empty-object string)
+ */
+function toolCallArgumentsString(value: unknown): string {
+  const parsed = recordOrEmpty(value);
+  if (isRecord(value)) return JSON.stringify(parsed);
+  if (typeof value === "string" && value.trim()) return value;
+  return JSON.stringify(parsed);
+}
+
 function normalizeContentText(content: unknown): string {
   if (typeof content === "string") return content;
   return asRecordArray(content)
@@ -244,11 +259,15 @@ function convertMessages(
         const id = stringValue(call.id) || "";
         if (!id || !pairedToolCallIds.has(id)) continue;
         const fn = isRecord(call.function) ? call.function : {};
+        const parsedInput = recordOrEmpty(fn.arguments);
         parts.push({
           type: "tool-call",
           toolCallId: id,
           toolName: stringValue(fn.name) || "",
-          input: recordOrEmpty(fn.arguments),
+          input: parsedInput,
+          // /alpha/generate requires this field on assistant tool-call parts;
+          // a missing one is rejected with `missing required field 'arguments'`.
+          arguments: toolCallArgumentsString(fn.arguments),
         });
       }
 
@@ -420,7 +439,61 @@ type AggregateState = {
   usage: JsonRecord | null;
 };
 
+function firstRecord(record: JsonRecord, keys: readonly string[]): JsonRecord {
+  for (const key of keys) {
+    const value = record[key];
+    if (isRecord(value)) return value;
+  }
+  return {};
+}
+
+function firstNumber(record: JsonRecord, keys: readonly string[]): number | undefined {
+  for (const key of keys) {
+    const value = numberValue(record[key]);
+    if (value !== undefined) return value;
+  }
+  return undefined;
+}
+
+/** Keep earlier finish-step usage when the terminal finish event omits it. */
+function mergeCommandCodeUsage(previous: JsonRecord | null, next: unknown): JsonRecord | null {
+  if (!isRecord(next)) return previous;
+
+  const merged: JsonRecord = { ...(previous || {}), ...next };
+  for (const key of [
+    "inputTokenDetails",
+    "input_token_details",
+    "input_tokens_details",
+    "prompt_tokens_details",
+    "outputTokenDetails",
+    "output_token_details",
+    "output_tokens_details",
+    "completion_tokens_details",
+    "reasoningTokenDetails",
+    "reasoning_token_details",
+  ]) {
+    const before = isRecord(previous?.[key]) ? previous[key] : {};
+    const after = isRecord(next[key]) ? next[key] : {};
+    if (Object.keys(before).length > 0 || Object.keys(after).length > 0) {
+      merged[key] = { ...before, ...after };
+    }
+  }
+  return merged;
+}
+
+function rememberCommandCodeUsage(state: AggregateState, event: JsonRecord): void {
+  const usage =
+    event.type === "finish-step"
+      ? (event.usage ?? event.totalUsage)
+      : (event.totalUsage ?? event.usage);
+  state.usage = mergeCommandCodeUsage(state.usage, usage);
+}
+
 function applyEventToAggregate(event: JsonRecord, state: AggregateState): void {
+  // Some Command Code protocol revisions attach usage to the terminal payload
+  // without preserving the event type. Capture it before event-specific handling.
+  rememberCommandCodeUsage(state, event);
+
   switch (event.type) {
     case "text-delta":
       state.content += stringValue(event.text) || "";
@@ -440,9 +513,10 @@ function applyEventToAggregate(event: JsonRecord, state: AggregateState): void {
       });
       break;
     }
+    case "finish-step":
+      break;
     case "finish":
       state.finishReason = mapFinishReason(event.finishReason);
-      state.usage = isRecord(event.totalUsage) ? event.totalUsage : null;
       break;
   }
 }
@@ -460,30 +534,72 @@ function applyEventToAggregateOrThrow(event: JsonRecord, state: AggregateState):
 
 function usageFromCommandCode(usage: JsonRecord | null) {
   if (!usage) return undefined;
-  const details = isRecord(usage.inputTokenDetails) ? usage.inputTokenDetails : {};
-  const cacheRead = numberValue(details.cacheReadTokens) || 0;
-  const noCache = numberValue(details.noCacheTokens) || 0;
+  const inputDetails = firstRecord(usage, [
+    "inputTokenDetails",
+    "input_token_details",
+    "input_tokens_details",
+    "prompt_tokens_details",
+  ]);
+  const outputDetails = firstRecord(usage, [
+    "outputTokenDetails",
+    "output_token_details",
+    "output_tokens_details",
+    "completion_tokens_details",
+  ]);
+  const reasoningDetails = firstRecord(usage, [
+    "reasoningTokenDetails",
+    "reasoning_token_details",
+    "reasoning_tokens_details",
+  ]);
+  const cacheRead =
+    firstNumber(usage, [
+      "cachedInputTokens",
+      "cached_input_tokens",
+      "cacheReadInputTokens",
+      "cache_read_input_tokens",
+      "cacheReadTokens",
+      "cache_read_tokens",
+      "cached_tokens",
+    ]) ??
+    firstNumber(inputDetails, [
+      "cachedTokens",
+      "cached_tokens",
+      "cacheReadTokens",
+      "cache_read_tokens",
+    ]);
+  const noCache = firstNumber(inputDetails, ["noCacheTokens", "no_cache_tokens"]);
   // Command Code's totalUsage.inputTokens is the FULL prompt total and already
   // includes the cached portion (noCacheTokens + cacheReadTokens = inputTokens),
   // so we must NOT add cacheRead back — that would double-count. There is no
   // cache-write field in the upstream payload, so cache creation stays unset.
-  const inputTokens = numberValue(usage.inputTokens) || 0;
-  const prompt = inputTokens;
-  const completion = numberValue(usage.outputTokens) || 0;
+  const prompt =
+    firstNumber(usage, ["inputTokens", "input_tokens", "promptTokens", "prompt_tokens"]) ??
+    (noCache ?? 0) + (cacheRead ?? 0);
+  const reasoning =
+    firstNumber(usage, ["reasoningTokens", "reasoning_tokens"]) ??
+    firstNumber(outputDetails, ["reasoningTokens", "reasoning_tokens"]) ??
+    firstNumber(reasoningDetails, ["reasoningTokens", "reasoning_tokens"]);
+  const textOutput = firstNumber(outputDetails, ["textTokens", "text_tokens"]);
+  const completion =
+    firstNumber(usage, [
+      "outputTokens",
+      "output_tokens",
+      "completionTokens",
+      "completion_tokens",
+    ]) ?? (textOutput ?? 0) + (reasoning ?? 0);
+  const total = firstNumber(usage, ["totalTokens", "total_tokens"]) ?? prompt + completion;
   const result: JsonRecord = {
     prompt_tokens: prompt,
+    prompt_tokens_details: { cached_tokens: cacheRead ?? 0 },
     completion_tokens: completion,
-    total_tokens: prompt + completion,
+    completion_tokens_details: { reasoning_tokens: reasoning ?? 0 },
+    total_tokens: total,
   };
   // Surface the cache breakdown as informational fields so logUsage prints
   // `| cache_read=X | no_cache=Y` and appendRequestLog persists them. These are
   // NOT added to prompt_tokens (already included) — metering stays accurate.
-  if (cacheRead > 0) result.cache_read_input_tokens = cacheRead;
-  if (noCache > 0) result.no_cache_tokens = noCache;
-  // Keep reasoning_token_details (reasoningTokens) when present so stream.ts's
-  // extractUsage can surface it as reasoning_tokens.
-  const reasoningDetails = isRecord(usage.reasoningTokenDetails) ? usage.reasoningTokenDetails : {};
-  const reasoning = numberValue(reasoningDetails.reasoningTokens);
+  if (cacheRead !== undefined && cacheRead > 0) result.cache_read_input_tokens = cacheRead;
+  if (noCache !== undefined && noCache > 0) result.no_cache_tokens = noCache;
   if (reasoning !== undefined && reasoning > 0) result.reasoning_tokens = reasoning;
   return result;
 }
@@ -523,6 +639,7 @@ function createStreamResponse(
 
       const emitEvent = (event: unknown) => {
         if (!isRecord(event) || closed) return;
+        rememberCommandCodeUsage(state, event);
         if (!sentRole) {
           sentRole = true;
           controller.enqueue(sse(chatCompletionChunk(id, model, { role: "assistant" })));
@@ -562,9 +679,10 @@ function createStreamResponse(
           }
           case "reasoning-end":
             break;
+          case "finish-step":
+            break;
           case "finish": {
             state.finishReason = mapFinishReason(event.finishReason);
-            state.usage = isRecord(event.totalUsage) ? event.totalUsage : null;
             controller.enqueue(sse(chatCompletionChunk(id, model, {}, state.finishReason)));
             // Emit a standards-compliant usage-only chunk (choices: []) before
             // [DONE] when upstream reported usage. stream.ts's extractUsage

open-sse/executors/deepseek-web.ts

@@ -515,7 +515,6 @@ export function messagesToPrompt(
   historyWindow = 0
 ): string {
   if (messages.length === 0) return "";
-
   const systemParts: string[] = [];
   const conversation: Array<{ role: string; text: string }> = [];
   const callNameById = new Map<string, string>();
@@ -527,8 +526,9 @@ export function messagesToPrompt(
     } else if (m.role === "user" || m.role === "assistant") {
       if (text) conversation.push({ role: m.role, text });
       if (m.role === "user") lastUserContent = text;
-      const calls = Array.isArray((m as { tool_calls?: unknown }).tool_calls)
-        ? (m as { tool_calls: Array<{ id?: string; function?: { name?: string } }> }).tool_calls
+      const toolCalls = (m as { tool_calls?: unknown }).tool_calls;
+      const calls = Array.isArray(toolCalls)
+        ? (toolCalls as Array<{ id?: string; function?: { name?: string } }>)
         : [];
       for (const c of calls) {
         if (c?.id && typeof c.function?.name === "string") callNameById.set(c.id, c.function.name);

open-sse/executors/default.ts

@@ -61,12 +61,11 @@ import {
 } from "@/lib/providers/validation/urlHelpers";
 import { forwardOpencodeClientHeaders } from "../utils/opencodeHeaders.ts";
 import { resolveZaiUrl } from "./default/zaiFormatOverride.ts";
+import { normalizePoolConfig } from "./default/poolConfig.ts";
 import { acquireNvidiaConcurrencySlot } from "./default/nvidiaConcurrencyGate.ts";
 import { resolveAlibabaProviderBaseUrl } from "@/shared/constants/alibabaProviderRegions";
 import { usesCcWireImage } from "../services/ccWireImageBuiltins.ts";
 
-import type { PoolConfig } from "../services/sessionPool/types.ts";
-
 const NVIDIA_TOOL_CALL_ID_PATTERN = /^[A-Za-z0-9]{9}$/;
 
 function normalizeNvidiaToolCallId(id: unknown): unknown {
@@ -146,7 +145,7 @@ export class DefaultExecutor extends BaseExecutor {
     super(provider, PROVIDERS[provider] || PROVIDERS.openai);
     const registryEntry = getRegistryEntry(provider);
     if (registryEntry?.poolConfig) {
-      this.poolConfig = registryEntry.poolConfig as PoolConfig;
+      this.poolConfig = normalizePoolConfig(registryEntry.poolConfig) ?? undefined;
     }
   }
 

open-sse/executors/default/poolConfig.ts

@@ -0,0 +1,33 @@
+import type { PoolConfig } from "../../services/sessionPool/types.ts";
+
+export function normalizePoolConfig(value: Record<string, unknown>): PoolConfig | null {
+  const {
+    minSessions,
+    maxSessions,
+    cooldownBase,
+    cooldownMax,
+    cooldownJitter,
+    requestTimeout,
+    requestJitter,
+  } = value;
+  if (
+    typeof minSessions !== "number" ||
+    typeof maxSessions !== "number" ||
+    typeof cooldownBase !== "number" ||
+    typeof cooldownMax !== "number" ||
+    typeof cooldownJitter !== "number" ||
+    typeof requestTimeout !== "number" ||
+    typeof requestJitter !== "number"
+  ) {
+    return null;
+  }
+  return {
+    minSessions,
+    maxSessions,
+    cooldownBase,
+    cooldownMax,
+    cooldownJitter,
+    requestTimeout,
+    requestJitter,
+  };
+}

open-sse/executors/duckduckgo-web.ts

@@ -137,8 +137,23 @@ interface DuckDuckGoModelCapabilities {
   reasoningEffort: string | null;
 }
 
+type DuckDuckGoRequestMessage = Record<string, unknown> & {
+  role: string;
+  content: unknown;
+};
+
 let durablePublicKey: JsonWebKey | null = null;
 
+export function normalizeDuckDuckGoMessages(value: unknown): DuckDuckGoRequestMessage[] {
+  if (!Array.isArray(value)) return [];
+  return value.flatMap((message) => {
+    if (!message || typeof message !== "object" || Array.isArray(message)) return [];
+    const record = message as Record<string, unknown>;
+    if (typeof record.role !== "string") return [];
+    return [{ ...record, role: record.role, content: record.content }];
+  });
+}
+
 function extractDuckDuckGoContent(data: unknown): string {
   if (!data || typeof data !== "object") return "";
   const record = data as Record<string, unknown>;
@@ -251,11 +266,14 @@ export function normalizeDuckDuckGoModel(model: string | undefined): string {
 }
 
 function getDuckDuckGoModelCapabilities(model: string): DuckDuckGoModelCapabilities {
-  // Per duckchat/v1/models (2026-07-22): claude-haiku-4-5 and gpt-oss-120b take a "low"
-  // reasoningEffort on the free tier; the others omit it (duck.ai applies its own default).
+  // `reasoningEffort` is REQUIRED on every duckchat/v1/chat request. Omitting it
+  // returns 400 ERR_BAD_REQUEST — A/B verified live against duck.ai with an
+  // otherwise byte-identical payload (200 with the field, 400 without, repeated).
+  // The live duck.ai bundle always sends one, so there is no "let the server
+  // pick a default" path any more.
   if (model === "claude-haiku-4-5") return { reasoningEffort: "low" };
   if (model === "tinfoil/gpt-oss-120b") return { reasoningEffort: "low" };
-  return { reasoningEffort: null };
+  return { reasoningEffort: "none" };
 }
 
 function extractDuckDuckGoFeVersion(html: string): string | null {
@@ -353,7 +371,6 @@ export class DuckDuckGoWebExecutor extends BaseExecutor {
   }
 
   private warmed = false;
-  private seeded = false;
   private feVersion = DEFAULT_FE_VERSION;
   private pendingVqdHash1: string | null = null;
   private readonly cookieJar = new Map<string, string>();
@@ -440,14 +457,12 @@ export class DuckDuckGoWebExecutor extends BaseExecutor {
     const { model, body, stream, signal, upstreamExtraHeaders } = input;
     const upstreamModel = normalizeDuckDuckGoModel(model);
     const bodyObj = (body || {}) as Record<string, unknown>;
-    const rawMessages = Array.isArray((body as { messages?: unknown[] } | null)?.messages)
-      ? ((body as { messages: unknown[] }).messages as Array<Record<string, unknown>>)
-      : [];
+    const rawMessages = normalizeDuckDuckGoMessages(bodyObj.messages);
     const { hasTools, requestedTools, effectiveMessages } = prepareToolMessages(
       bodyObj,
       rawMessages
     );
-    const messages = effectiveMessages as Array<Record<string, unknown>>;
+    const messages = effectiveMessages;
     const isStreaming = stream !== false;
     const upstreamHeaders = upstreamExtraHeaders || {};
 
@@ -561,7 +576,12 @@ export class DuckDuckGoWebExecutor extends BaseExecutor {
       }
 
       await this.warmSession(mergedSignal);
-      await this.seedChallengeChain(upstreamModel, mergedSignal);
+      // NOTE: the throwaway "seed" chat POST that used to run here has been removed.
+      // It existed to coax a usable challenge out of the upstream while the solver
+      // was broken; now that the solver reproduces a real browser's probe vectors
+      // exactly, the first real request succeeds on its own. Keeping it only doubled
+      // the chat calls per user request against an IP-rate-limited endpoint, which
+      // showed up as spurious 429 ERR_RATE_LIMIT.
       const vqdHeaders = await this.acquireAuthHeaders(mergedSignal);
       if (!vqdHeaders.vqd4 && !vqdHeaders.vqdHash1) {
         clearTimeout(timeout);
@@ -770,41 +790,6 @@ export class DuckDuckGoWebExecutor extends BaseExecutor {
     );
   }
 
-  private async seedChallengeChain(model: string, signal: AbortSignal): Promise<void> {
-    if (this.seeded || signal.aborted) return;
-    this.seeded = true;
-    const seedMessages = [{ role: "user", content: "hi" }];
-    const previousPending = this.pendingVqdHash1;
-    try {
-      const vqdHeaders = await this.acquireAuthHeaders(signal);
-      if (!vqdHeaders.vqd4 && !vqdHeaders.vqdHash1) {
-        this.pendingVqdHash1 = previousPending;
-        return;
-      }
-      const response = await fetch(CHAT_URL, {
-        method: "POST",
-        headers: mergeHeadersCaseInsensitive(this.buildRequestHeaders(), {
-          Accept: "text/event-stream",
-          "Content-Type": "application/json",
-          "x-ddg-journey-id": randomUUID().replaceAll("-", ""),
-          "x-fe-signals": makeDuckDuckGoFeSignals(),
-          "x-fe-version": this.feVersion,
-          ...(vqdHeaders.vqd4 ? { "x-vqd-4": vqdHeaders.vqd4 } : {}),
-          ...(vqdHeaders.vqdHash1 ? { "x-vqd-hash-1": vqdHeaders.vqdHash1 } : {}),
-        }),
-        body: JSON.stringify(buildDuckDuckGoPayload(model, seedMessages, false)),
-        signal,
-      });
-      this.rememberResponseCookies(response);
-      if (response.ok) this.rememberChallengeHeader(response);
-      else this.pendingVqdHash1 = previousPending;
-      await response.body?.cancel().catch(() => {});
-    } catch (error) {
-      void error;
-      this.pendingVqdHash1 = previousPending;
-    }
-  }
-
   private async processResponse(
     response: Response,
     streaming: boolean,

open-sse/executors/duckduckgo-web/challenge.ts

@@ -5,12 +5,38 @@ import { createHash } from "node:crypto";
 import vm from "node:vm";
 import { parseFragment, serialize } from "parse5";
 
+// WARNING: the contents of this template literal are NOT TypeScript — they are plain
+// script-mode JavaScript executed via `vm.runInContext`. `vm.runInContext` compiles in
+// script (non-module) mode, so an `export` keyword anywhere in here is a hard
+// SyntaxError that kills the whole solver. A refactor that mass-added `export` to the
+// five `function` declarations below silently broke every DuckDuckGo chat request
+// (solve threw -> unsolved challenge sent -> HTTP 418 ERR_CHALLENGE). Do not add
+// `export`/`import` to this string; `duckduckgo-challenge-split.test.ts` guards this.
 export const CHALLENGE_STUBS = String.raw`
 var __ua = __DDG_REAL_UA__;
 var __HTML_LOOKUP = __DDG_HTML_LOOKUP__;
-export function __makeHtmlElement(tag) {
+// Browser-fidelity shims for the DDG "am I a real browser" probes.
+// In a browser every built-in stringifies as native code; under a plain vm
+// context the user-land re-declarations below would otherwise leak their source.
+function __nativeFn(fn, name){
+  Object.defineProperty(fn, 'name', { value: name, configurable: true });
+  fn.toString = function(){ return 'function ' + name + '() { [native code] }'; };
+  return fn;
+}
+__nativeFn(parseInt, 'parseInt');
+__nativeFn(parseFloat, 'parseFloat');
+__nativeFn(isNaN, 'isNaN');
+__nativeFn(encodeURIComponent, 'encodeURIComponent');
+__nativeFn(decodeURIComponent, 'decodeURIComponent');
+// NOTE: do NOT seal Math. Real Chromium reports Object.isSealed(Math) === false,
+// and at least one challenge variant probes exactly that; sealing it here made
+// the vector differ from the browser by one and failed the challenge.
+function __makeHtmlElement(tag) {
   var state = { _innerHTML: '', _qsaCount: 0, _cssText: '' };
-  var el = {
+  // Instantiate against the real per-tag constructor so
+  // document.createElement('div') instanceof HTMLDivElement holds.
+  var el = Object.create(__ctorForTag(tag).prototype);
+  Object.assign(el, {
     tagName: String(tag).toUpperCase(), nodeName: String(tag).toUpperCase(), nodeType: 1,
     children: [], childNodes: [], classList: [], dataset: {},
     offsetWidth: 1, offsetHeight: 1, clientWidth: 1, clientHeight: 1, scrollHeight: 1, scrollWidth: 1,
@@ -19,9 +45,9 @@ export function __makeHtmlElement(tag) {
     getAttribute: function(a){ if(a==='srcdoc') return state._srcdoc||''; return null; },
     hasAttribute: function(){ return false; }, appendChild: function(c){ return c; }, removeChild: function(c){ return c; },
     addEventListener: function(){}, removeEventListener: function(){}, querySelector: function(){ return null; },
-    querySelectorAll: function(s){ if (s === '*') { var arr = []; arr.length = state._qsaCount; return arr; } return []; },
+    querySelectorAll: function(s){ if (s === '*') { return __makeNodeList(state._qsaCount); } return __makeNodeList(0); },
     cloneNode: function(){ return __makeHtmlElement(tag); }
-  };
+  });
   Object.defineProperty(el, 'style', { value: new Proxy({}, { set: function(t, k, v){ t[k] = v; if (k === 'cssText') state._cssText = String(v); return true; }, get: function(t, k){ if (k === 'cssText') return state._cssText; return t[k] || ''; } }), enumerable: true, configurable: true });
   Object.defineProperty(el, 'innerHTML', { get: function(){ return state._innerHTML; }, set: function(v){ var key = String(v); var entry = __HTML_LOOKUP && __HTML_LOOKUP[key]; if (entry) { state._innerHTML = String(entry.html); state._qsaCount = entry.count|0; } else { state._innerHTML = key; state._qsaCount = 0; } }, enumerable: true, configurable: true });
   Object.defineProperty(el, 'outerHTML', { get: function(){ return '<' + tag + '>' + state._innerHTML + '</' + tag + '>'; }, enumerable: true });
@@ -30,7 +56,7 @@ export function __makeHtmlElement(tag) {
   Object.defineProperty(el, 'contentDocument', { get: function(){ return __ifDoc; }, enumerable: true });
   return el;
 }
-export function __mkObj(name, base) {
+function __mkObj(name, base) {
   base = base || {};
   return new Proxy(base, {
     get: function(t, k) {
@@ -54,18 +80,105 @@ export function __mkObj(name, base) {
     has: function(t, k){ return k in t; }, set: function(t, k, v){ t[k] = v; return true; }
   });
 }
-export function __parseCssDisplay(cssText){ if(!cssText) return ''; var m = String(cssText).match(/(?:^|;)\\s*display\\s*:\\s*([^;]+)/i); return m ? String(m[1]).trim() : ''; }
-export function __getComputedStyle(el){ var cssText = el && el.style && el.style.cssText || ''; var display = __parseCssDisplay(cssText); return { getPropertyValue: function(name){ if(String(name).toLowerCase()==='display') return display; return ''; }, cssText: cssText, display: display }; }
+function __parseCssDisplay(cssText){ if(!cssText) return ''; var m = String(cssText).match(/(?:^|;)\s*display\s*:\s*([^;]+)/i); return m ? String(m[1]).trim() : ''; }
+function __getComputedStyle(el){ var cssText = el && el.style && el.style.cssText || ''; var display = __parseCssDisplay(cssText); return { getPropertyValue: function(name){ if(String(name).toLowerCase()==='display') return display; return ''; }, cssText: cssText, display: display }; }
 var __ifMeta = __mkObj('meta', { getAttribute: function(a){ return a==='content' ? "default-src 'none'; script-src 'unsafe-inline';" : null; }, hasAttribute: function(a){ return a==='content'; }, tagName: 'META', nodeName: 'META' });
 var __ifDoc = __mkObj('iframeDoc', { querySelector: function(s){ if (s && s.indexOf('Content-Security-Policy') !== -1) return __ifMeta; if (s === 'meta') return __ifMeta; return null; }, querySelectorAll: function(s){ if (s && s.indexOf('Content-Security-Policy') !== -1) return [__ifMeta]; if (s === 'meta') return [__ifMeta]; return []; }, getElementsByTagName: function(t){ return t && t.toLowerCase()==='meta' ? [__ifMeta] : []; }, body: __mkObj('iframeBody'), head: __mkObj('iframeHead'), documentElement: __mkObj('iframeRoot'), createElement: function(){ return __mkObj('elem', {setAttribute:function(){}, appendChild:function(){}, removeChild:function(){}, getAttribute:function(){return null;}, hasAttribute:function(){return false;}}); }, cookie: '', readyState: 'complete' });
 var __iframeEl = __mkObj('iframe', { contentDocument: __ifDoc, contentWindow: __mkObj('iframeWin', { document: __ifDoc, top: undefined, parent: undefined }), document: __ifDoc, getAttribute: function(a){ if (a==='sandbox') return 'allow-scripts allow-same-origin'; if (a==='srcdoc') return ''; if (a==='id') return 'jsa'; return null; }, hasAttribute: function(a){ return a==='sandbox'||a==='id'; }, tagName: 'IFRAME', nodeName: 'IFRAME', id: 'jsa' });
-var document = __mkObj('document', { querySelector: function(s){ if (s === '#jsa') return __iframeEl; if (s && s.indexOf('Content-Security-Policy') !== -1) return __ifMeta; return null; }, querySelectorAll: function(s){ if (s === '#jsa') return [__iframeEl]; if (s && s.indexOf('Content-Security-Policy') !== -1) return [__ifMeta]; return []; }, getElementById: function(id){ return id==='jsa' ? __iframeEl : null; }, getElementsByTagName: function(t){ if(t&&t.toLowerCase()==='iframe') return [__iframeEl]; return []; }, getElementsByClassName: function(){ return []; }, body: __mkObj('body', {appendChild:function(){}, removeChild:function(){}, querySelector:function(s){return s==='#jsa'?__iframeEl:null;}, querySelectorAll:function(s){return s==='#jsa'?[__iframeEl]:[];}}), head: __mkObj('head'), documentElement: __mkObj('root'), createElement: function(tag){ return __makeHtmlElement(tag||'div'); }, createTextNode: function(t){ return {nodeType:3, nodeValue:String(t||''), textContent:String(t||'')}; }, cookie: '', readyState: 'complete', title: '', addEventListener: function(){}, removeEventListener: function(){} });
+// document.body keeps a LIVE children collection: challenges append a node and
+// assert body.children.length grew by exactly 1, then remove it again.
+var __bodyKids = [];
+Object.defineProperty(__bodyKids, 'constructor', { value: HTMLCollection, enumerable: false, configurable: true });
+var __body = __mkObj('body', {
+  appendChild: function(c){ __bodyKids.push(c); return c; },
+  removeChild: function(c){ var i = __bodyKids.indexOf(c); if (i !== -1) __bodyKids.splice(i, 1); return c; },
+  contains: function(c){ return __bodyKids.indexOf(c) !== -1; },
+  querySelector: function(s){ return s === '#jsa' ? __iframeEl : null; },
+  querySelectorAll: function(s){ return s === '#jsa' ? [__iframeEl] : __makeNodeList(0); },
+  children: __bodyKids, childNodes: __bodyKids,
+  tagName: 'BODY', nodeName: 'BODY', nodeType: 1
+});
+var document = __mkObj('document', { querySelector: function(s){ if (s === '#jsa') return __iframeEl; if (s && s.indexOf('Content-Security-Policy') !== -1) return __ifMeta; return null; }, querySelectorAll: function(s){ if (s === '#jsa') return [__iframeEl]; if (s && s.indexOf('Content-Security-Policy') !== -1) return [__ifMeta]; return __makeNodeList(__bodyKids.length + 3); }, getElementById: function(id){ return id==='jsa' ? __iframeEl : null; }, getElementsByTagName: function(t){ if(t&&t.toLowerCase()==='iframe') return [__iframeEl]; return []; }, getElementsByClassName: function(){ return []; }, body: __body, head: __mkObj('head'), documentElement: __mkObj('root'), createElement: function(tag){ return __makeHtmlElement(tag||'div'); }, createTextNode: function(t){ return {nodeType:3, nodeValue:String(t||''), textContent:String(t||'')}; }, cookie: '', readyState: 'complete', title: '', addEventListener: function(){}, removeEventListener: function(){} });
   var window = __mkObj('window', { document: document, __DDG_BE_VERSION__: 1, __DDG_FE_CHAT_HASH__: 1, navigator: __mkObj('navigator', { userAgent: __ua, webdriver: false, language: 'en-US', languages: ['en-US','en'], platform: 'Linux x86_64', vendor: 'Google Inc.', appVersion: '5.0 (X11)', cookieEnabled: true, onLine: true, hardwareConcurrency: 8, deviceMemory: 8 }), innerWidth: 1280, innerHeight: 800, outerWidth: 1280, outerHeight: 800, devicePixelRatio: 1, screen: __mkObj('screen', { width:1920, height:1080, availWidth:1920, availHeight:1080, colorDepth:24, pixelDepth:24 }), location: __mkObj('location', { href:'https://duck.ai/', origin:'https://duck.ai', host:'duck.ai', hostname:'duck.ai', protocol:'https:', pathname:'/' }), performance: __mkObj('perf', { now: function(){ return 0; }, timeOrigin: 0 }), history: __mkObj('history', { length: 1, state: null }), addEventListener: function(){}, removeEventListener: function(){}, dispatchEvent: function(){return true;}, setTimeout: function(fn){ try{fn();}catch(e){} return 0; }, clearTimeout: function(){}, hasOwnProperty: function(k){ if (k==='__DDG_BE_VERSION__'||k==='__DDG_FE_CHAT_HASH__') return true; return Object.prototype.hasOwnProperty.call(this,k); } });
 window.top = window; window.self = window; window.window = window; window.parent = window; window.globalThis = window;
+// Object.prototype.toString.call(window) must be "[object Window]".
+try { window[Symbol.toStringTag] = 'Window'; } catch (e) {}
+// In a browser a sloppy-mode function called with no receiver gets the global
+// object, and challenges assert (function(){return this;})() === window.
+// In a vm context that is the context's own global, so alias it to window.
+try {
+  var __g = (function(){ return this; })();
+  if (__g && __g !== window) {
+    Object.defineProperty(__g, Symbol.toStringTag, { value: 'Window', configurable: true });
+    // Copy by VALUE, not via accessors. Two reasons:
+    //  1) the var top/self/navigator/... declarations further down are hoisted,
+    //     so those names already exist on the vm global and an "in" guard would
+    //     skip them, leaving window.navigator undefined;
+    //  2) accessors closing over the window binding would recurse once it is
+    //     rebound to __g below.
+    // The stub window is static, so a value copy is equivalent.
+    var __winStub = window;
+    for (var __k in __winStub) {
+      try { __g[__k] = __winStub[__k]; } catch (e) {}
+    }
+    // hasOwnProperty is probed for the __DDG_* markers; keep the stub's version.
+    try { __g.hasOwnProperty = function(k){ return __winStub.hasOwnProperty(k); }; } catch (e) {}
+    window = __g;
+    window.top = window; window.self = window; window.window = window; window.parent = window; window.globalThis = window;
+  }
+} catch (e) {}
 var top = window, self = window, parent = window, navigator = window.navigator, location = window.location, screen = window.screen, performance = window.performance, history = window.history;
 var __R = null, __E = null;
-export function __HTMLClass(name){ var c = function(){}; c.prototype = __mkObj(name+'.proto'); return c; }
-var HTMLElement = __HTMLClass('HTMLElement'), HTMLDivElement = __HTMLClass('HTMLDivElement'), HTMLIFrameElement = __HTMLClass('HTMLIFrameElement'), HTMLDocument = __HTMLClass('HTMLDocument'), Document = __HTMLClass('Document'), Element = __HTMLClass('Element'), Node = __HTMLClass('Node'), Window = __HTMLClass('Window'), Event = __HTMLClass('Event'), MouseEvent = __HTMLClass('MouseEvent'), KeyboardEvent = __HTMLClass('KeyboardEvent'), TouchEvent = __HTMLClass('TouchEvent'), XMLHttpRequest = __HTMLClass('XMLHttpRequest'), WebSocket = __HTMLClass('WebSocket'), Image = __HTMLClass('Image'), FormData = __HTMLClass('FormData'), Blob = __HTMLClass('Blob'), File = __HTMLClass('File'), FileReader = __HTMLClass('FileReader'), URL = __HTMLClass('URL'), URLSearchParams = __HTMLClass('URLSearchParams'), Headers = __HTMLClass('Headers'), Request = __HTMLClass('Request'), Response = __HTMLClass('Response');
+// Real DOM constructor chain. Some DDG challenge variants assert
+// HTMLDivElement.prototype instanceof HTMLElement and
+// HTMLElement.prototype instanceof Element, so these cannot be flat
+// unrelated stubs — the prototype links have to be real.
+function __DomClass(name, parent){
+  var c = function(){};
+  if (parent) c.prototype = Object.create(parent.prototype);
+  c.prototype.constructor = c;
+  Object.defineProperty(c, 'name', { value: name, configurable: true });
+  c.toString = function(){ return 'function ' + name + '() { [native code] }'; };
+  return c;
+}
+var EventTarget = __DomClass('EventTarget', null);
+var Node = __DomClass('Node', EventTarget);
+var Element = __DomClass('Element', Node);
+var HTMLElement = __DomClass('HTMLElement', Element);
+var HTMLDivElement = __DomClass('HTMLDivElement', HTMLElement);
+var HTMLIFrameElement = __DomClass('HTMLIFrameElement', HTMLElement);
+var HTMLLIElement = __DomClass('HTMLLIElement', HTMLElement);
+var HTMLUnknownElement = __DomClass('HTMLUnknownElement', HTMLElement);
+var Document = __DomClass('Document', Node);
+var HTMLDocument = __DomClass('HTMLDocument', Document);
+var NodeList = __DomClass('NodeList', null);
+var HTMLCollection = __DomClass('HTMLCollection', null);
+// Map a tag name to the constructor a browser would use, so
+// document.createElement('div') instanceof HTMLDivElement holds.
+function __ctorForTag(tag){
+  var t = String(tag||'div').toLowerCase();
+  if (t === 'div') return HTMLDivElement;
+  if (t === 'iframe') return HTMLIFrameElement;
+  if (t === 'li') return HTMLLIElement;
+  return HTMLElement;
+}
+// A NodeList-like: array-shaped but NOT a real Array, with .constructor.name
+// === 'NodeList' — challenges check both !Array.isArray(x) and the ctor name.
+function __makeNodeList(length){
+  var nl = Object.create(NodeList.prototype);
+  var n = length|0;
+  for (var i = 0; i < n; i++) nl[i] = __makeHtmlElement('div');
+  Object.defineProperty(nl, 'length', { value: n, enumerable: false, configurable: true });
+  nl.item = function(i){ return this[i] || null; };
+  nl.forEach = function(fn, thisArg){ for (var i = 0; i < n; i++) fn.call(thisArg, this[i], i, this); };
+  nl[Symbol.iterator] = function(){ var i = 0, self = this; return { next: function(){ return i < n ? { value: self[i++], done: false } : { value: undefined, done: true }; } }; };
+  return nl;
+}
+function __HTMLClass(name){ var c = function(){}; c.prototype = __mkObj(name+'.proto'); return c; }
+// NOTE: HTMLElement / HTMLDivElement / HTMLIFrameElement / Element / Node /
+// Document / HTMLDocument / NodeList are defined above via __DomClass with a
+// REAL prototype chain — do not redeclare them here or the instanceof probes break.
+var Window = __HTMLClass('Window'), Event = __HTMLClass('Event'), MouseEvent = __HTMLClass('MouseEvent'), KeyboardEvent = __HTMLClass('KeyboardEvent'), TouchEvent = __HTMLClass('TouchEvent'), XMLHttpRequest = __HTMLClass('XMLHttpRequest'), WebSocket = __HTMLClass('WebSocket'), Image = __HTMLClass('Image'), FormData = __HTMLClass('FormData'), Blob = __HTMLClass('Blob'), File = __HTMLClass('File'), FileReader = __HTMLClass('FileReader'), URL = __HTMLClass('URL'), URLSearchParams = __HTMLClass('URLSearchParams'), Headers = __HTMLClass('Headers'), Request = __HTMLClass('Request'), Response = __HTMLClass('Response');
 var fetch = function(){ return Promise.resolve(__mkObj('resp', {ok:true, status:200, json:function(){return Promise.resolve({});}, text:function(){return Promise.resolve('');}})); };
 var getComputedStyle = __getComputedStyle;
 `;
@@ -90,9 +203,16 @@ export function buildHtmlLookup(js: string): Record<string, { html: string; coun
     if (seen.has(html)) continue;
     seen.add(html);
     const fragment = parseFragment(html);
+    // `count` backs `element.querySelectorAll('*').length` for an element whose
+    // innerHTML is `html`. `querySelectorAll('*')` on a container returns its
+    // DESCENDANTS, and `countHtmlElements` already excludes the `#document-fragment`
+    // root, so the fragment's element count IS the descendant count. The former
+    // `- 1` undercounted by one (verified against a real browser: for
+    // `<li><div></li><li></div` Chromium reports 3, this returned 2), which
+    // corrupted every probe that multiplies by that length.
     lookup[html] = {
       html: serialize(fragment),
-      count: Math.max(0, countHtmlElements(fragment) - 1),
+      count: countHtmlElements(fragment),
     };
   }
   return lookup;
@@ -102,14 +222,33 @@ export function sha256Base64(value: string): string {
   return createHash("sha256").update(value, "utf8").digest("base64");
 }
 
+// Shape of the object a DDG challenge program resolves to.
 type DuckDuckGoChallengeResult = {
   client_hashes?: unknown;
+  meta?: unknown;
   [key: string]: unknown;
 };
 
+/**
+ * Origin the solved challenge claims to come from. The duck.ai frontend stamps
+ * `meta.origin` with its own origin and the upstream cross-checks it.
+ */
+export const DUCKDUCKGO_CHALLENGE_ORIGIN = "https://duck.ai";
+
+/**
+ * `meta.stack` mimics the frontend's captured Error stack. The upstream only
+ * requires a plausible stack that points at the duck.ai bundle — verified by
+ * ablation: a generic bundle path is accepted, omitting the field is not.
+ */
+function buildChallengeStack(origin: string, bundlePath: string): string {
+  const url = `${origin}${bundlePath}`;
+  return `Error\nat l (${url}:2:1695625)\nat async ${url}:2:1519117`;
+}
+
 export async function solveDuckDuckGoChallenge(
   challenge: string,
-  userAgent: string
+  userAgent: string,
+  options: { origin?: string; bundlePath?: string } = {}
 ): Promise<string> {
   // SECURITY NOTE: This function executes base64-decoded JavaScript from duck.ai via vm.runInContext.
   // The challenge code is upstream-supplied (supply-chain surface). It is sandboxed with a 5s timeout
@@ -121,14 +260,31 @@ export async function solveDuckDuckGoChallenge(
   );
   const context = vm.createContext({});
   vm.runInContext(stubs, context, { timeout: 5000 });
+  const startedAt = Date.now();
   const result = (await vm.runInContext(js, context, {
     timeout: 5000,
   })) as DuckDuckGoChallengeResult;
+  const elapsedMs = Date.now() - startedAt;
   const clientHashes = Array.isArray(result.client_hashes) ? result.client_hashes : [];
   if (clientHashes.length === 0)
     throw new Error("DuckDuckGo challenge returned empty client_hashes");
   clientHashes[0] = userAgent;
   result.client_hashes = clientHashes.map((hash) => sha256Base64(String(hash)));
+
+  // The real frontend augments the challenge's own `meta` with origin / stack /
+  // duration before sending it back. Omitting them yields 418 ERR_CHALLENGE even
+  // when every client_hash is correct (confirmed by capturing a real browser's
+  // x-vqd-hash-1 header, which always carries all three).
+  const origin = options.origin ?? DUCKDUCKGO_CHALLENGE_ORIGIN;
+  const bundlePath = options.bundlePath ?? "/dist/duckai-dist/entry.duckai.js";
+  const meta = (result.meta ?? {}) as Record<string, unknown>;
+  result.meta = {
+    ...meta,
+    origin,
+    stack: buildChallengeStack(origin, bundlePath),
+    duration: String(elapsedMs),
+  };
+
   return Buffer.from(JSON.stringify(result), "utf8").toString("base64");
 }
 

open-sse/executors/gemini-business.ts

@@ -80,16 +80,7 @@ export class GeminiBusinessExecutor extends BaseExecutor {
     // Extract cookies from credentials — check apiKey/cookie first, then
     // try each __Secure-1PSID* key in providerSpecificData individually.
     // A user with only __Secure-1PSID (no PSIDTS) is still valid.
-    const directCookie =
-      readCredentialString(credentials?.apiKey) || readCredentialString(credentials?.cookie);
-    const psid = readProviderSpecificString(credentials?.providerSpecificData, [
-      "__Secure-1PSID",
-      "cookie",
-    ]);
-    const psidts = readProviderSpecificString(credentials?.providerSpecificData, [
-      "__Secure-1PSIDTS",
-    ]);
-    const cookie = directCookie || [psid, psidts].filter(Boolean).join("; ");
+    const cookie = resolveGeminiBusinessCookie(credentials);
 
     if (!cookie) {
       return makeErrorResult(
@@ -380,6 +371,15 @@ function readProviderSpecificString(providerSpecificData: unknown, keys: string[
   return "";
 }
 
+export function resolveGeminiBusinessCookie(credentials: unknown): string {
+  if (!credentials || typeof credentials !== "object") return "";
+  const data = credentials as Record<string, unknown>;
+  const directCookie = readCredentialString(data.apiKey) || readCredentialString(data.cookie);
+  const psid = readProviderSpecificString(data.providerSpecificData, ["__Secure-1PSID", "cookie"]);
+  const psidts = readProviderSpecificString(data.providerSpecificData, ["__Secure-1PSIDTS"]);
+  return directCookie || [psid, psidts].filter(Boolean).join("; ");
+}
+
 function extractTextContent(content: unknown): string {
   if (typeof content === "string") return content.trim();
   if (Array.isArray(content)) {

open-sse/executors/index.ts

@@ -25,6 +25,7 @@ import { ChatGptWebExecutor } from "./chatgpt-web.ts";
 import { BlackboxWebExecutor } from "./blackbox-web.ts";
 import { MuseSparkWebExecutor } from "./muse-spark-web.ts";
 import { AzureOpenAIExecutor } from "./azure-openai.ts";
+import { AzureAiExecutor } from "./azure-ai.ts";
 import { CommandCodeExecutor } from "./commandCode.ts";
 import { GitlabExecutor } from "./gitlab.ts";
 import { NlpCloudExecutor } from "./nlpcloud.ts";
@@ -89,6 +90,7 @@ const executors = {
   glmt: new GlmExecutor("glmt"),
   cu: new CursorExecutor(), // Alias for cursor
   "azure-openai": new AzureOpenAIExecutor(),
+  "azure-ai": new AzureAiExecutor(),
   "command-code": new CommandCodeExecutor(),
   cmd: new CommandCodeExecutor(), // Alias
   gitlab: new GitlabExecutor(),
@@ -263,6 +265,7 @@ export { ChatGptWebExecutor } from "./chatgpt-web.ts";
 export { BlackboxWebExecutor } from "./blackbox-web.ts";
 export { MuseSparkWebExecutor } from "./muse-spark-web.ts";
 export { AzureOpenAIExecutor } from "./azure-openai.ts";
+export { AzureAiExecutor } from "./azure-ai.ts";
 export { CommandCodeExecutor } from "./commandCode.ts";
 export { GitlabExecutor } from "./gitlab.ts";
 export { NlpCloudExecutor } from "./nlpcloud.ts";

open-sse/executors/theoldllm.ts

@@ -108,13 +108,9 @@ export function mapModel(model: string): string {
 const TOKEN_SEED = "oldllm-client-2026";
 const UA_PREFIX = CHROME_UA.slice(0, 20); // "Mozilla/5.0 (Windows"
 
-type TheOldLlmProxy = {
-  type?: string;
-  host: string;
-  port: number;
-  username?: string | null;
-  password?: string | null;
-} | null;
+type TheOldLlmProxy = Awaited<
+  ReturnType<typeof import("../../src/lib/db/proxies").resolveProxyForProvider>
+>;
 
 interface TheOldLlmFetchDependencies {
   resolveProxy: () => Promise<TheOldLlmProxy>;

open-sse/handlers/chatCore.ts

@@ -142,6 +142,8 @@ import {
   getExplicitModelOutputCap,
   resolveInputTokenCapForGate,
 } from "@/lib/modelCapabilities.ts";
+import { checkRequestCapabilityFit, deriveRequestCapabilityRequirements, buildCapabilityMismatchMessage } from "@/shared/constants/capabilities/capabilityFilter.ts";
+import { isFeatureFlagEnabled } from "@/shared/utils/featureFlags.ts";
 import { toPositiveInteger } from "../services/reasoningTokenBuffer.ts";
 import { normalizeThinkingForModel } from "@/shared/constants/modelSpecs.ts";
 import {
@@ -171,6 +173,7 @@ import {
   ANTIGRAVITY_PRE_RESPONSE_TIMEOUT_CODE,
   STREAM_RECOVERY,
   DEFAULT_MAX_TOKENS,
+  STREAM_DISCONNECT_GRACE_PERIOD_MS,
 } from "../config/constants.ts";
 import { createRecoverableStream, makeContinuationBody } from "../services/streamRecovery.ts";
 import {
@@ -520,7 +523,6 @@ export async function handleChatCore({
       `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
   // provider/credentials once and delegate so the existing call sites stay byte-identical.
@@ -549,7 +551,6 @@ export async function handleChatCore({
       })
     ).catch(() => {});
   };
-
   // Key-health updater extracted to chatCore/keyHealth.ts (#3501); bind the per-request log once
   // and delegate so the existing call sites stay byte-identical.
   const recordKeyHealthStatus = (
@@ -557,11 +558,9 @@ export async function handleChatCore({
     creds: Record<string, unknown> | null | undefined,
     transport?: string
   ): void => recordKeyHealthStatusFor(status, creds, log, transport);
-
   const persistCodexQuotaState = async (headers: Record<string, string> | null, status = 0) => {
     const currentConnectionId = getCurrentConnectionId();
     if (provider !== "codex" || !currentConnectionId || !headers) return;
-
     try {
       const existingProviderData =
         credentials?.providerSpecificData && typeof credentials.providerSpecificData === "object"
@@ -576,28 +575,23 @@ export async function handleChatCore({
         status,
       });
       if (!built) return;
-
       if (built.exhaustionLog) {
         log?.debug?.("CODEX", built.exhaustionLog);
       }
-
       // Invalidate the preflight cache for this connection so the next
       // isModelAvailable check fetches fresh quota data.
       if (status === 429) {
         invalidateCodexQuotaCache(currentConnectionId);
       }
-
       await updateProviderConnection(currentConnectionId, {
         providerSpecificData: built.nextProviderData,
       });
-
       credentials.providerSpecificData = built.nextProviderData;
     } catch (err) {
       const errMessage = err instanceof Error ? err.message : String(err);
       log?.debug?.("CODEX", `Failed to persist codex quota state: ${errMessage}`);
     }
   };
-
   // ── Phase 9.2: Idempotency check ──
   // Resolve the idempotency key once here and reuse it at the Phase 9.2 save site below,
   // rather than re-deriving it. (#3821-review LEDGER-6)
@@ -616,13 +610,11 @@ export async function handleChatCore({
   if (idempotencyHit) {
     return idempotencyHit;
   }
-
   // T07: Inject connectionId into credentials so executors can rotate API keys
   // using providerSpecificData.extraApiKeys (API Key Round-Robin feature)
   if (connectionId && credentials && !credentials.connectionId) {
     credentials.connectionId = connectionId;
   }
-
   // Endpoint/format resolution extracted to chatCore/requestFormat.ts (#3501); pure derivation
   // from the inbound request, destructured so every downstream use stays byte-identical.
   const {
@@ -2265,8 +2257,19 @@ export async function handleChatCore({
   // the latter is a Kiro/Claude passthrough alias channel with string values,
   // while namespace identities carry `{namespace, name}` for the #7936 response
   // seam. Extract first because Kiro merge may reuse `_toolNameMap` below.
+  //
+  // #9780 — prefer the dedicated channel: on a pivot the openai->claude/gemini
+  // step publishes its own alias map on `_toolNameMap`, so that property alone
+  // yields aliases here. The `_toolNameMap` read stays as the fallback for the
+  // non-pivot producers (executors/base.ts, cliproxyapi.ts, antigravity).
+  const namespaceIdentityMap = translatedBody._namespaceToolIdentityMap;
   const requestToolIdentityMap =
-    translatedBody._toolNameMap instanceof Map ? translatedBody._toolNameMap : null;
+    namespaceIdentityMap instanceof Map
+      ? namespaceIdentityMap
+      : translatedBody._toolNameMap instanceof Map
+        ? translatedBody._toolNameMap
+        : null;
+  delete translatedBody._namespaceToolIdentityMap;
   delete translatedBody._toolNameMap;
 
   // Kiro: sanitize tool schemas before dispatch. Kiro returns 400 "Improperly
@@ -2637,7 +2640,16 @@ export async function handleChatCore({
     }
   }
   // === /Quota Share enforcement PRE-hook ===
-
+  if (isFeatureFlagEnabled("CAPABILITY_FILTER_ENABLED")) {
+    const fit = checkRequestCapabilityFit(getResolvedModelCapabilities({ provider, model: effectiveModel }),
+      deriveRequestCapabilityRequirements(body as Record<string, unknown>), provider);
+    if (!fit.compatible) {
+      const msg = buildCapabilityMismatchMessage(fit.terminalReason!, provider, effectiveModel);
+      log?.warn?.("CAPABILITY", msg);
+      trackPendingRequest(model, provider, connectionId, false);
+      return createErrorResult(400, msg, null, fit.terminalReason, "invalid_request_error");
+    }
+  }
   // Get executor for this provider (with optional upstream proxy routing)
   const executor = await resolveExecutorWithProxy(provider);
   const getExecutionCredentials = () =>
@@ -4334,9 +4346,14 @@ export async function handleChatCore({
     try {
       const firstChoice = translatedResponse?.choices?.[0];
       const msg = firstChoice?.message;
+      // The response being cached now will be replayed as history on the *next*
+      // turn, where the read side (translator/index.ts) keys the lookup by the
+      // message's real position in that future `messages` array — i.e. right
+      // after everything the client sent this turn.
+      const bodyMessages = (body as { messages?: unknown[] } | null | undefined)?.messages;
       cacheReasoningFromAssistantMessage(msg, provider, model, {
         requestId: skillRequestId,
-        messageIndex: 0,
+        messageIndex: Array.isArray(bodyMessages) ? bodyMessages.length : 0,
       });
     } catch {
       // Cache capture is non-critical — never block the response
@@ -4761,12 +4778,15 @@ export async function handleChatCore({
     // with tool_calls so it can be replayed on subsequent turns (DeepSeek V4, Kimi K2, etc.)
     if (normalizedStreamStatus === 200 && streamResponseBody) {
       try {
-        const body = streamResponseBody as Record<string, unknown>;
-        const choices = body.choices as { message?: Record<string, unknown> }[] | undefined;
+        const streamBody = streamResponseBody as Record<string, unknown>;
+        const choices = streamBody.choices as { message?: Record<string, unknown> }[] | undefined;
         const msg = choices?.[0]?.message;
+        // See the non-streaming capture above: messageIndex must match the
+        // position this message will occupy in the *next* turn's history.
+        const bodyMessages = (body as { messages?: unknown[] } | null | undefined)?.messages;
         cacheReasoningFromAssistantMessage(msg, provider, model, {
           requestId: skillRequestId,
-          messageIndex: 0,
+          messageIndex: Array.isArray(bodyMessages) ? bodyMessages.length : 0,
         });
       } catch {
         // Cache capture is non-critical — never block the stream
@@ -4909,13 +4929,20 @@ export async function handleChatCore({
   });
   const handleStreamFailure = streamFailureFinalizers.handleStreamFailure;
   onPipelineStreamError = streamFailureFinalizers.onPipelineStreamError;
-  onClientDisconnectFinalize = (event) =>
-    handleStreamFailure({
-      status: 499,
-      message: `Client disconnected: ${event.reason}`,
-      code: "client_disconnected",
-      type: "client_disconnected",
-    });
+  // #9653: gives a genuine, race-delayed completion a chance to land (see
+  // createClientDisconnectGraceHandler's doc comment) before persisting a false
+  // 499/0-tokens for a request that actually delivered its full response.
+  onClientDisconnectFinalize = streamFailure.createClientDisconnectGraceHandler({
+    isStreamCompletionRecorded: () => streamCompletionRecorded,
+    gracePeriodMs: STREAM_DISCONNECT_GRACE_PERIOD_MS,
+    finalize: (event) =>
+      handleStreamFailure({
+        status: 499,
+        message: `Client disconnected: ${event.reason}`,
+        code: "client_disconnected",
+        type: "client_disconnected",
+      }),
+  });
 
   // For providers using Responses API format, translate stream back to openai (Chat Completions) format
   // UNLESS client is Droid CLI which expects openai-responses format back

open-sse/handlers/chatCore/logTruncation.ts

@@ -3,11 +3,11 @@ import {
   getChatLogMaxDepth,
   getChatLogArrayTailItems,
   getChatLogMaxObjectKeys,
+  getChatLogMaxBodyBytes,
 } from "@/lib/logEnv";
 import { estimateSizeFast } from "../../utils/estimateSize.ts";
 
 export const MEMORY_EXTRACTION_TEXT_LIMIT = 64 * 1024;
-const MAX_LOG_BODY_CHARS = 8 * 1024; // 8KB cap for logged request/response bodies
 
 export function capMemoryExtractionText(value: string): string {
   if (value.length <= MEMORY_EXTRACTION_TEXT_LIMIT) return value;
@@ -60,9 +60,10 @@ export function cloneBoundedChatLogPayload(value: unknown, depth = 0): unknown {
 
 /**
  * Truncate a large object for logging. If its JSON representation exceeds
- * 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.
+ * the configured max body size (getChatLogMaxBodyBytes()), 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
@@ -75,8 +76,9 @@ export function cloneBoundedChatLogPayload(value: unknown, depth = 0): unknown {
 export function truncateForLog(value: unknown): Record<string, unknown> | null | undefined {
   if (value === null || value === undefined) return value as null | undefined;
   if (typeof value !== "object") return value as unknown as Record<string, unknown>;
-  const estimatedSize = estimateSizeFast(value);
-  if (estimatedSize <= MAX_LOG_BODY_CHARS) return value as Record<string, unknown>;
+  const maxBodyBytes = getChatLogMaxBodyBytes();
+  const estimatedSize = estimateSizeFast(value, maxBodyBytes);
+  if (estimatedSize <= maxBodyBytes) return value as Record<string, unknown>;
   // Object is too large — return a summary instead of a deep clone
   const obj = value as Record<string, unknown>;
   const summary: Record<string, unknown> = {

open-sse/handlers/imageGeneration.ts

@@ -1524,7 +1524,7 @@ async function handleFalAIImageGeneration({
     }
 
     const payload = await response.json();
-    const images = await normalizeProviderImagePayload(payload, body, log);
+    const images = await normalizeProviderImagePayload(payload, body, log, "b64_json");
     return saveImageSuccessResult({
       provider,
       model,
@@ -1714,7 +1714,7 @@ async function handleStabilityAIImageGeneration({
       payload = { image: buffer.toString("base64") };
     }
 
-    const images = await normalizeProviderImagePayload(payload, body, log);
+    const images = await normalizeProviderImagePayload(payload, body, log, "b64_json");
     return saveImageSuccessResult({
       provider,
       model,
@@ -1833,7 +1833,7 @@ async function handleBlackForestLabsImageGeneration({
         })
       : initialPayload;
 
-    const images = await normalizeProviderImagePayload(finalPayload, body, log);
+    const images = await normalizeProviderImagePayload(finalPayload, body, log, "url");
     return saveImageSuccessResult({
       provider,
       model,
@@ -1908,7 +1908,7 @@ async function handleRecraftImageGeneration({
     }
 
     const payload = await response.json();
-    const images = await normalizeProviderImagePayload(payload, body, log);
+    const images = await normalizeProviderImagePayload(payload, body, log, "url");
     return saveImageSuccessResult({
       provider,
       model,
@@ -2200,7 +2200,7 @@ function shouldIncludeStabilityMask(model) {
   ]).has(model);
 }
 
-async function normalizeProviderImagePayload(payload, body, log) {
+async function normalizeProviderImagePayload(payload, body, log, defaultFormat) {
   const candidates = [];
 
   const pushCandidate = (value) => {
@@ -2226,7 +2226,7 @@ async function normalizeProviderImagePayload(payload, body, log) {
 
   const normalized = [];
   for (const candidate of candidates) {
-    const item = await normalizeProviderImageCandidate(candidate, body);
+    const item = await normalizeProviderImageCandidate(candidate, body, defaultFormat);
     if (item) normalized.push(item);
   }
 
@@ -2240,8 +2240,8 @@ async function normalizeProviderImagePayload(payload, body, log) {
   return normalized;
 }
 
-async function normalizeProviderImageCandidate(candidate, body) {
-  const wantsBase64 = body?.response_format === "b64_json";
+async function normalizeProviderImageCandidate(candidate, body, defaultFormat) {
+  const wantsBase64 = body?.response_format === "b64_json" || defaultFormat === "b64_json";
   let url = null;
   let b64 = null;
 

open-sse/handlers/imageGeneration/providers/adobeFirefly.ts

@@ -15,14 +15,15 @@ import { saveImageErrorResult, saveImageSuccessResult } from "../../imageGenerat
 import {
   AdobeFireflyError,
   adobeFireflyGenerateImage,
-  adobeFireflyImageTimeoutMs,
-  resolveAdobeAccessToken,
-  resolveAdobeSourceImageReferences,
+  resolveAdobeSourceImageIds,
   resolveAdobeImageModel,
 } from "../../../services/adobeFireflyClient.ts";
-import { getAdobeReferenceUploadLimit } from "../../../services/adobeFireflyModels.ts";
-import { isAdobeFireflyUpscaleModel } from "../../../services/adobeFireflyUpscale.ts";
-import { handleAdobeFireflyImageUpscale } from "../../imageUpscale/adobeFirefly.ts";
+import { ensureAdobeFireflySession } from "../../../services/adobeFireflySession.ts";
+
+function normalizePositiveNumber(value: unknown, fallback: number): number {
+  const n = Number(value);
+  return Number.isFinite(n) && n > 0 ? n : fallback;
+}
 
 export async function handleAdobeFireflyImageGeneration({
   model,
@@ -50,25 +51,22 @@ export async function handleAdobeFireflyImageGeneration({
     images?: unknown;
     [key: string]: unknown;
   };
-  credentials: { apiKey?: string; accessToken?: string };
+  credentials: {
+    apiKey?: string;
+    accessToken?: string;
+    connectionId?: string;
+    providerSpecificData?: {
+      cookie?: unknown;
+      access_token?: unknown;
+      accessToken?: unknown;
+      browserSessionKey?: unknown;
+    } | null;
+  };
   log?: { info?: (...args: unknown[]) => void; error?: (...args: unknown[]) => void };
   fetchImpl?: typeof fetch;
 }) {
   const startTime = Date.now();
   const prompt = typeof body.prompt === "string" ? body.prompt.trim() : "";
-
-  // Topaz upscalers share adobe-firefly but use /v2/3p-images/upsample (no prompt).
-  if (isAdobeFireflyUpscaleModel(model)) {
-    return handleAdobeFireflyImageUpscale({
-      model,
-      provider,
-      body: body as Record<string, unknown>,
-      credentials,
-      log,
-      fetchImpl,
-    });
-  }
-
   if (!prompt) {
     return saveImageErrorResult({
       provider,
@@ -80,7 +78,17 @@ export async function handleAdobeFireflyImageGeneration({
   }
 
   try {
-    const accessToken = await resolveAdobeAccessToken(credentials, fetchImpl);
+    // Durable session: JWT + Cookie once → auto-rebuild ARP from forter/arkose,
+    // cache, optional Playwright warm-up. Submit path rotates ARP on 408.
+    const session = await ensureAdobeFireflySession({
+      credentials,
+      fetchImpl,
+      log,
+    });
+    const accessToken = session.accessToken;
+    const sessionCookie = session.cookie || undefined;
+    const arpSessionId = session.arpSessionId;
+    const timeoutMs = normalizePositiveNumber(body.timeout_ms, 180_000);
     const seed =
       typeof body.seed === "number"
         ? body.seed
@@ -88,44 +96,26 @@ export async function handleAdobeFireflyImageGeneration({
           ? Number(body.seed)
           : undefined;
 
-    // Keep the raw credential blob for Cookie + sherlockToken (x-arp-session-id).
-    // JWT may be embedded in the same paste as cookies (HAR / multi-line).
-    const psd = (credentials as { providerSpecificData?: { cookie?: string } })
-      ?.providerSpecificData;
-    const sessionCookie =
-      (typeof psd?.cookie === "string" && psd.cookie.trim()) ||
-      (typeof credentials?.apiKey === "string" && credentials.apiKey.trim()) ||
-      (typeof credentials?.accessToken === "string" && credentials.accessToken.includes(";")
-        ? credentials.accessToken
-        : undefined);
+    // Cap uploads by model family (matches MediaViewModel GetSourceImageLimit).
+    const { id: resolvedId } = resolveAdobeImageModel(model);
+    const maxRefs = resolvedId.includes("nano-banana") || resolvedId.includes("gpt-image") ? 4 : 2;
 
-    const { spec } = resolveAdobeImageModel(model);
-    const references = await resolveAdobeSourceImageReferences({
+    const sourceImageIds = await resolveAdobeSourceImageIds({
       accessToken,
       body,
-      max: getAdobeReferenceUploadLimit(spec, "image"),
+      max: maxRefs,
       sessionCookie,
+      arpSessionId,
       prompt,
       fetchImpl,
       log,
     });
 
-    const explicitTimeout =
-      typeof body.timeout_ms === "number"
-        ? body.timeout_ms
-        : typeof body.timeout_ms === "string" && body.timeout_ms.trim()
-          ? Number(body.timeout_ms)
-          : undefined;
-    const timeoutMs = adobeFireflyImageTimeoutMs({
-      timeoutMs: explicitTimeout,
-      refCount: references.length,
-    });
-
     log?.info?.(
       "IMAGE",
       `${provider}/${model} (adobe-firefly) | prompt: "${prompt.slice(0, 60)}${prompt.length > 60 ? "..." : ""}"` +
-        (references.length ? ` | refs: ${references.length}` : "") +
-        ` | pollTimeoutMs=${timeoutMs}`
+        (sourceImageIds.length ? ` | refs: ${sourceImageIds.length}` : "") +
+        ` | session=${session.source}`
     );
 
     const result = await adobeFireflyGenerateImage({
@@ -137,8 +127,11 @@ export async function handleAdobeFireflyImageGeneration({
       quality: body.quality,
       seed: Number.isFinite(seed as number) ? (seed as number) : undefined,
       negativePrompt: typeof body.negative_prompt === "string" ? body.negative_prompt : undefined,
-      references: references.length ? references : undefined,
+      sourceImageIds: sourceImageIds.length ? sourceImageIds : undefined,
       sessionCookie,
+      arpSessionId,
+      sessionFingerprint: session.fingerprint,
+      sessionBrowserKey: session.browserSessionKey,
       timeoutMs,
       fetchImpl,
       log,

open-sse/handlers/responsesHandler.ts

@@ -40,7 +40,12 @@ export async function handleResponsesCore({
   const customToolNames = collectResponsesCustomToolNames(body?.tools, inputItems);
 
   // Convert Responses API format to Chat Completions format
-  const convertedBody = convertResponsesApiFormat(body, credentials, modelInfo?.provider);
+  const convertedBody = convertResponsesApiFormat(
+    body,
+    credentials,
+    modelInfo?.provider,
+    modelInfo?.model
+  );
 
   // Ensure stream is enabled
   convertedBody.stream = true;

open-sse/handlers/videoGeneration.ts

@@ -4,7 +4,7 @@
  * Handles POST /v1/videos/generations requests. Proxies to upstream video
  * generation providers (ComfyUI AnimateDiff/SVD, SD WebUI AnimateDiff, and
  * more — see the per-format handlers below). Response format (OpenAI-like):
- * { "created": 1234567890, "data": [{ "b64_json": "...", "format": "mp4" }] }
+ * { "created": 1234567890, "data": [{ "url": "https://…", "format": "mp4" }] }
  */
 
 import { getVideoProvider, parseVideoModel } from "../config/videoRegistry.ts";
@@ -18,6 +18,16 @@ import { handleNovitaVideoGeneration } from "./videoGeneration/novitaHandler.ts"
 import { handleXaiVideoGeneration } from "./videoGeneration/xaiGrokImagineHandler.ts";
 import { handleSegmindVideoGeneration } from "./videoGeneration/providers/segmind.ts";
 import { handleAdobeFireflyVideoGeneration } from "./videoGeneration/adobeFireflyHandler.ts";
+import { handleOpenAIVideoGeneration } from "./videoGeneration/openai.ts";
+import { getVideoJobPreset, handleVideoJobGeneration } from "./videoGeneration/job.ts";
+import {
+  extractRunwayFailureMessage,
+  normalizeRunwayVideoResult,
+  resolvePositiveInteger,
+  resolveRunwayDuration,
+  resolveRunwayPromptImage,
+  resolveRunwayRatio,
+} from "./videoGeneration/runwayHelpers.ts";
 import { getExecutor } from "../executors/index.ts";
 import { getKieTaskId, isJsonObject, parseKieResultJson } from "../utils/kieTask.ts";
 import {
@@ -33,13 +43,94 @@ import {
   resolveComfyUiBaseUrl,
 } from "../utils/comfyuiClient.ts";
 import { saveCallLog } from "@/lib/usageDb";
+import { getAllCustomModels } from "@/lib/db/models";
 import { sanitizeErrorMessage } from "../utils/error.ts";
+import {
+  FetchTimeoutError,
+  fetchWithTimeout,
+  getConfiguredTimeout,
+} from "@/shared/utils/fetchTimeout";
+
+/**
+ * Resolve the base URL for OpenAI-compatible video generation endpoints.
+ * Prefers providerSpecificData.baseUrl (from custom node config), falls back to
+ * top-level credentials.baseUrl, then to the provided fallback.
+ */
+export function resolveVideoBaseUrl(
+  credentials:
+    { baseUrl?: unknown; providerSpecificData?: { baseUrl?: unknown } | null } | null | undefined,
+  fallback: string
+): string {
+  const psd = credentials?.providerSpecificData;
+  const psdBaseUrl =
+    psd && typeof psd === "object" && typeof psd.baseUrl === "string" && psd.baseUrl.trim()
+      ? psd.baseUrl.trim()
+      : null;
+  const topLevelBaseUrl =
+    typeof credentials?.baseUrl === "string" && credentials.baseUrl.trim()
+      ? credentials.baseUrl.trim()
+      : null;
+  const nodeBaseUrl = psdBaseUrl || topLevelBaseUrl;
+
+  if (!nodeBaseUrl) return fallback;
+
+  // Trim trailing slashes
+  let normalized = nodeBaseUrl;
+  while (normalized.endsWith("/")) normalized = normalized.slice(0, -1);
+  if (normalized.endsWith("/videos/generations")) return normalized;
+  const stripped = normalized.replace(/\/videos\/generations$/, "");
+  return `${stripped}/videos/generations`;
+}
+
+/**
+ * Read generationConfig.preset from the custom model row for the given
+ * provider/model id. Returns null when the model has no preset configured (or
+ * the registry is unreadable), so callers can fall back to the sync path.
+ */
+async function getCustomModelVideoPreset(
+  providerId: string,
+  modelId: string
+): Promise<string | null> {
+  try {
+    const customModelsMap = (await getAllCustomModels()) as Record<
+      string,
+      Array<Record<string, unknown>>
+    >;
+    const models = customModelsMap[providerId];
+    if (!Array.isArray(models)) return null;
+    for (const model of models) {
+      if (!model || typeof model !== "object" || model.id !== modelId) continue;
+      const generationConfig = model.generationConfig;
+      if (
+        generationConfig &&
+        typeof generationConfig === "object" &&
+        typeof (generationConfig as Record<string, unknown>).preset === "string"
+      ) {
+        return (generationConfig as Record<string, unknown>).preset as string;
+      }
+      return null;
+    }
+    return null;
+  } catch {
+    return null;
+  }
+}
 
 /**
  * Handle video generation request
  */
-export async function handleVideoGeneration({ body, credentials, log }) {
-  const { provider, model } = parseVideoModel(body.model);
+
+/**
+ * Handle video generation request
+ */
+export async function handleVideoGeneration({ body, credentials, log, resolvedProvider = null }) {
+  let { provider, model } = parseVideoModel(body.model);
+  if (resolvedProvider) {
+    provider = resolvedProvider;
+    model = body.model.startsWith(provider + "/")
+      ? body.model.slice(provider.length + 1)
+      : body.model;
+  }
 
   if (!provider) {
     return {
@@ -51,11 +142,59 @@ export async function handleVideoGeneration({ body, credentials, log }) {
 
   const providerConfig = getVideoProvider(provider);
   if (!providerConfig) {
-    return {
-      success: false,
-      status: 400,
-      error: `Unknown video provider: ${provider}`,
+    if (!resolvedProvider) {
+      return {
+        success: false,
+        status: 400,
+        error: `Unknown video provider: ${provider}`,
+      };
+    }
+    // Custom provider node. When the custom model row carries a
+    // generationConfig.preset (e.g. "agnes-video-job"), dispatch through the
+    // submit → poll job pipeline; otherwise mirror the images route and use the
+    // generic OpenAI-compatible handler with a synthetic config.
+    const presetName = await getCustomModelVideoPreset(provider, model);
+    if (presetName !== null) {
+      if (!getVideoJobPreset(presetName)) {
+        return {
+          success: false,
+          status: 502,
+          error: `Unknown video job preset: ${presetName}`,
+        };
+      }
+      if (log)
+        log.info("VIDEO", `Custom model ${provider}/${model} — using job preset ${presetName}`);
+      return handleVideoJobGeneration({
+        model,
+        presetName,
+        body,
+        credentials,
+        log,
+      });
+    }
+    if (log)
+      log.info("VIDEO", `Custom model ${provider}/${model} — using OpenAI-compatible handler`);
+    const syntheticConfig = {
+      id: provider,
+      baseUrl: resolveVideoBaseUrl(
+        credentials,
+        "http://generative.language.googleapis.com/v1beta/openai/videos/generations"
+      ),
+      authType: "apikey",
+      authHeader: "bearer",
+      format: "openai-video",
     };
+    return handleOpenAIVideoGeneration({
+      model,
+      body,
+      credentials,
+      provider,
+      providerConfig: syntheticConfig,
+      log,
+    });
+  }
+  if (providerConfig.format === "openai-video") {
+    return handleOpenAIVideoGeneration({ model, provider, providerConfig, body, credentials, log });
   }
 
   if (providerConfig.format === "vertex-veo") {
@@ -158,7 +297,10 @@ export async function handleVideoGeneration({ body, credentials, log }) {
       log,
     });
   }
-
+  if (resolvedProvider) {
+    // Custom provider with no matching built-in format — use OpenAI-compatible fallback
+    return handleOpenAIVideoGeneration({ model, provider, providerConfig, body, credentials, log });
+  }
   return {
     success: false,
     status: 400,
@@ -832,148 +974,6 @@ const RUNWAY_TERMINAL_FAILURE_STATUSES = new Set([
   "DELETED",
 ]);
 
-function resolveRunwayPromptImage(body) {
-  const directCandidates = [
-    body.promptImage,
-    body.prompt_image,
-    body.image,
-    body.image_url,
-    body.imageUrl,
-    body.provider_options?.promptImage,
-    body.provider_options?.prompt_image,
-  ];
-
-  for (const candidate of directCandidates) {
-    if (typeof candidate === "string" && candidate.trim()) return candidate.trim();
-    if (candidate && typeof candidate === "object") return candidate;
-    if (Array.isArray(candidate) && candidate.length > 0) return candidate;
-  }
-
-  const arrayCandidates = [
-    body.imageUrls,
-    body.image_urls,
-    body.provider_options?.imageUrls,
-    body.provider_options?.image_urls,
-  ];
-  for (const candidate of arrayCandidates) {
-    if (Array.isArray(candidate) && candidate.length > 0) return candidate;
-  }
-
-  return null;
-}
-
-function resolveRunwayRatio(body) {
-  const aspectRatio = typeof body.aspect_ratio === "string" ? body.aspect_ratio : body.aspectRatio;
-  if (aspectRatio === "1280:720" || aspectRatio === "720:1280") return aspectRatio;
-  if (aspectRatio === "16:9") return "1280:720";
-  if (aspectRatio === "9:16") return "720:1280";
-
-  const size = typeof body.size === "string" ? body.size : "";
-  const [widthRaw, heightRaw] = size.split("x");
-  const width = Number(widthRaw);
-  const height = Number(heightRaw);
-  if (Number.isFinite(width) && Number.isFinite(height) && width > 0 && height > 0) {
-    return width >= height ? "1280:720" : "720:1280";
-  }
-
-  return "1280:720";
-}
-
-function resolveRunwayDuration(body) {
-  if (Number.isFinite(body.duration)) {
-    return clampRunwayDuration(body.duration);
-  }
-
-  if (Number.isFinite(body.frames) && Number.isFinite(body.fps) && Number(body.fps) > 0) {
-    return clampRunwayDuration(Number(body.frames) / Number(body.fps));
-  }
-
-  return 5;
-}
-
-function clampRunwayDuration(value) {
-  const duration = Math.round(Number(value));
-  if (!Number.isFinite(duration)) return 5;
-  return Math.min(10, Math.max(2, duration));
-}
-
-function resolvePositiveInteger(value, fallback) {
-  const numeric = Number(value);
-  if (!Number.isFinite(numeric) || numeric <= 0) return fallback;
-  return Math.floor(numeric);
-}
-
-function extractRunwayOutputUrls(task) {
-  const rawOutput = Array.isArray(task?.output)
-    ? task.output
-    : Array.isArray(task?.result)
-      ? task.result
-      : [];
-
-  return rawOutput
-    .map((entry) => {
-      if (typeof entry === "string") return entry;
-      if (!entry || typeof entry !== "object") return null;
-      return entry.url || entry.uri || entry.videoUrl || entry.video_url || null;
-    })
-    .filter((value) => typeof value === "string" && value.length > 0);
-}
-
-function extractRunwayFailureMessage(task) {
-  const directCandidates = [
-    task?.failure,
-    task?.failureReason,
-    task?.error,
-    task?.errorMessage,
-    task?.message,
-  ];
-  for (const candidate of directCandidates) {
-    if (typeof candidate === "string" && candidate.trim()) return candidate.trim();
-  }
-
-  if (task?.failure && typeof task.failure === "object") {
-    const nestedCandidates = [
-      task.failure.message,
-      task.failure.reason,
-      task.failure.error,
-      task.failure.code,
-    ];
-    for (const candidate of nestedCandidates) {
-      if (typeof candidate === "string" && candidate.trim()) return candidate.trim();
-    }
-  }
-
-  return null;
-}
-
-async function normalizeRunwayVideoResult(task, body) {
-  const urls = extractRunwayOutputUrls(task);
-  if (urls.length === 0) {
-    throw new Error(
-      `Runway task completed without output URLs: ${JSON.stringify(task).slice(0, 400)}`
-    );
-  }
-
-  if (body.response_format === "url") {
-    return urls.map((url) => ({ url, format: "mp4" }));
-  }
-
-  const videos = [];
-  for (const url of urls) {
-    const response = await fetch(url);
-    if (!response.ok) {
-      throw new Error(`Runway output fetch failed (${response.status})`);
-    }
-    const arrayBuffer = await response.arrayBuffer();
-    videos.push({
-      b64_json: Buffer.from(arrayBuffer).toString("base64"),
-      format: "mp4",
-    });
-  }
-
-  return videos;
-}
-
 async function handleHaiperVideoGeneration({
   model,
   provider,

open-sse/handlers/videoGeneration/adobeFireflyHandler.ts

@@ -9,11 +9,10 @@ import { sanitizeErrorMessage } from "../../utils/error.ts";
 import {
   AdobeFireflyError,
   adobeFireflyGenerateVideo,
-  resolveAdobeAccessToken,
-  resolveAdobeSourceImageReferences,
+  resolveAdobeSourceImageIds,
   resolveAdobeVideoModel,
 } from "../../services/adobeFireflyClient.ts";
-import { getAdobeReferenceUploadLimit } from "../../services/adobeFireflyModels.ts";
+import { ensureAdobeFireflySession } from "../../services/adobeFireflySession.ts";
 
 function normalizePositiveNumber(value: unknown, fallback: number): number {
   const n = Number(value);
@@ -32,7 +31,17 @@ export async function handleAdobeFireflyVideoGeneration({
   provider: string;
   providerConfig?: { baseUrl?: string };
   body: Record<string, unknown>;
-  credentials?: { apiKey?: string; accessToken?: string } | null;
+  credentials?: {
+    apiKey?: string;
+    accessToken?: string;
+    connectionId?: string;
+    providerSpecificData?: {
+      cookie?: unknown;
+      access_token?: unknown;
+      accessToken?: unknown;
+      browserSessionKey?: unknown;
+    } | null;
+  } | null;
   log?: { info?: (...args: unknown[]) => void; error?: (...args: unknown[]) => void };
   fetchImpl?: typeof fetch;
 }) {
@@ -47,7 +56,14 @@ export async function handleAdobeFireflyVideoGeneration({
   }
 
   try {
-    const accessToken = await resolveAdobeAccessToken(credentials, fetchImpl);
+    const session = await ensureAdobeFireflySession({
+      credentials,
+      fetchImpl,
+      log,
+    });
+    const accessToken = session.accessToken;
+    const sessionCookie = session.cookie || undefined;
+    const arpSessionId = session.arpSessionId;
     const timeoutMs = normalizePositiveNumber(body.timeout_ms, 300_000);
     const seed =
       typeof body.seed === "number"
@@ -55,22 +71,16 @@ export async function handleAdobeFireflyVideoGeneration({
         : typeof body.seed === "string" && String(body.seed).trim()
           ? Number(body.seed)
           : undefined;
-    // Keep raw paste for Cookie + sherlockToken (x-arp-session-id).
-    const psd = (credentials as { providerSpecificData?: { cookie?: string } })
-      ?.providerSpecificData;
-    const sessionCookie =
-      (typeof psd?.cookie === "string" && psd.cookie.trim()) ||
-      (typeof credentials?.apiKey === "string" && credentials.apiKey.trim()) ||
-      (typeof credentials?.accessToken === "string" && credentials.accessToken.includes(";")
-        ? credentials.accessToken
-        : undefined);
 
-    const { spec } = resolveAdobeVideoModel(String(model));
-    const references = await resolveAdobeSourceImageReferences({
+    // Kling i2v / Veo ref / Sora frame: upload reference images first.
+    const { id: videoModelId } = resolveAdobeVideoModel(String(model));
+    const maxFrames = videoModelId.includes("kling") || videoModelId.includes("sora") ? 2 : 3;
+    const sourceImageIds = await resolveAdobeSourceImageIds({
       accessToken,
       body,
-      max: getAdobeReferenceUploadLimit(spec, "image"),
+      max: maxFrames,
       sessionCookie,
+      arpSessionId,
       prompt,
       fetchImpl,
       log,
@@ -79,7 +89,8 @@ export async function handleAdobeFireflyVideoGeneration({
     log?.info?.(
       "VIDEO",
       `${provider}/${model} (adobe-firefly) | prompt: "${prompt.slice(0, 60)}${prompt.length > 60 ? "..." : ""}"` +
-        (references.length ? ` | refs: ${references.length}` : "")
+        (sourceImageIds.length ? ` | frames: ${sourceImageIds.length}` : "") +
+        ` | session=${session.source}`
     );
 
     const result = await adobeFireflyGenerateVideo({
@@ -99,8 +110,11 @@ export async function handleAdobeFireflyVideoGeneration({
             ? body.negativePrompt
             : undefined,
       generateAudio: body.generate_audio !== false && body.generateAudio !== false,
-      references: references.length ? references : undefined,
+      sourceImageIds: sourceImageIds.length ? sourceImageIds : undefined,
       sessionCookie,
+      arpSessionId,
+      sessionFingerprint: session.fingerprint,
+      sessionBrowserKey: session.browserSessionKey,
       timeoutMs,
       fetchImpl,
       log,

open-sse/handlers/videoGeneration/job.ts

@@ -0,0 +1,418 @@
+/**
+ * Async job/poll video generation for custom OpenAI-compatible provider nodes
+ * whose /videos surface is a submit → poll → fetch-result API (e.g. Agnes
+ * Video V2.0, muapi.ai, OpenAI Sora). Presets are declarative data — the
+ * handler here is one family; everything else is per-preset config.
+ *
+ * Response shape stays OpenAI-like: { created, data: [{ url, format: "mp4" }] } so the
+ * /v1/videos/generations route returns the same contract as the synchronous
+ * path.
+ */
+
+import {
+  fetchWithTimeout,
+  FetchTimeoutError,
+  getConfiguredTimeout,
+} from "@/shared/utils/fetchTimeout";
+import { sanitizeErrorMessage } from "../../utils/error.ts";
+import { sleep } from "../../utils/sleep.ts";
+
+interface LogLike {
+  info?: (tag: string, msg: string, meta?: unknown) => void;
+  warn?: (tag: string, msg: string, meta?: unknown) => void;
+  error?: (tag: string, msg: string, meta?: unknown) => void;
+}
+
+interface CredentialsLike {
+  providerSpecificData?: { baseUrl?: unknown } | null;
+  baseUrl?: unknown;
+  apiKey?: unknown;
+  accessToken?: unknown;
+}
+
+/** Dot-path reader restricted to plain objects/arrays (no prototypes). */
+function readPath(value: unknown, path: string): unknown {
+  if (!path) return value;
+  let current: unknown = value;
+  for (const segment of path.split(".")) {
+    if (current === null || current === undefined) return undefined;
+    if (typeof current !== "object") return undefined;
+    if (Array.isArray(current)) {
+      const index = Number(segment);
+      if (!Number.isInteger(index) || index < 0 || index >= current.length) return undefined;
+      current = current[index];
+      continue;
+    }
+    if (!Object.prototype.hasOwnProperty.call(current, segment)) return undefined;
+    current = (current as Record<string, unknown>)[segment];
+  }
+  return current;
+}
+
+/** Non-empty string from a dot path, or null. */
+function readStringPath(value: unknown, path: string): string | null {
+  const found = readPath(value, path);
+  return typeof found === "string" && found.trim() ? found : null;
+}
+
+function isDoneStatus(
+  status: unknown,
+  done: string[],
+  failed: string[]
+): "done" | "failed" | "pending" {
+  if (typeof status !== "string") return "pending";
+  if (failed.includes(status)) return "failed";
+  if (done.includes(status)) return "done";
+  return "pending";
+}
+
+export type VideoJobPreset = {
+  id: string;
+  displayName: string;
+  /** auth header name plus value scheme */
+  authHeaderName: "x-api-key" | "Authorization";
+  authScheme: "bearer" | "raw";
+  baseUrlFallback: string;
+  submit: {
+    method: "POST";
+    /** may contain {model} — substituted before POST */
+    path: string;
+    buildBody: (params: {
+      model?: string;
+      prompt?: string;
+      duration?: number;
+      extras: Record<string, unknown>;
+    }) => Record<string, unknown>;
+  };
+  /** dot path into the submit response identifying the job */
+  taskIdPath: string;
+  poll: {
+    /** contains {taskId} */
+    pathTemplate: string;
+  };
+  statusPath: string;
+  statusDone: string[];
+  statusFailed: string[];
+  /** dot path into the poll response holding the finished video URL/array */
+  resultPath: string;
+  maxPolls: number;
+  pollIntervalMs: number;
+};
+
+// #9820: declarative presets for the shipping async job/poll video providers.
+const VIDEO_JOB_PRESETS: Record<string, VideoJobPreset> = {
+  "agnes-video-job": {
+    id: "agnes-video-job",
+    displayName: "Agnes Video V2.0",
+    authHeaderName: "x-api-key",
+    authScheme: "raw",
+    // Real default, matching the Agnes Video V2.0 reference: POST /v1/videos with
+    // x-api-key auth; GET /v1/videos/{task_id} returns status/progress/metadata.
+    baseUrlFallback: "https://apihub.agnes-ai.com",
+    submit: {
+      method: "POST",
+      path: "/v1/videos",
+      buildBody: ({ model, prompt, extras }) => ({
+        model,
+        prompt,
+        // passthrough of image/mode/num_frames/frame_rate/…  — the generic
+        // route body uses .catchall, so provider-specific knobs survive.
+        ...extras,
+      }),
+    },
+    taskIdPath: "task_id",
+    poll: { pathTemplate: "/v1/videos/{taskId}" },
+    statusPath: "status",
+    statusDone: ["completed"],
+    statusFailed: ["failed"],
+    resultPath: "metadata.url",
+    maxPolls: 60,
+    pollIntervalMs: 2000,
+  },
+  "muapi-video-job": {
+    id: "muapi-video-job",
+    displayName: "muapi.ai",
+    authHeaderName: "x-api-key",
+    authScheme: "raw",
+    // muapi.ai video/audio surface is Replicate-style: POST /api/v1/{model}
+    // returns { request_id }; poll GET /api/v1/predictions/{id}/result.
+    baseUrlFallback: "https://api.muapi.ai",
+    submit: {
+      method: "POST",
+      path: "/api/v1/{model}",
+      buildBody: (params) => {
+        const { prompt, duration, extras } = params;
+        return {
+          prompt,
+          ...(typeof duration === "number" ? { duration } : {}),
+          ...extras,
+        };
+      },
+    },
+    taskIdPath: "request_id",
+    poll: { pathTemplate: "/api/v1/predictions/{taskId}/result" },
+    statusPath: "status",
+    statusDone: ["completed"],
+    statusFailed: ["failed"],
+    resultPath: "outputs",
+    maxPolls: 60,
+    pollIntervalMs: 2000,
+  },
+  "sora-job": {
+    id: "sora-job",
+    displayName: "OpenAI Sora",
+    authHeaderName: "Authorization",
+    authScheme: "bearer",
+    baseUrlFallback: "https://api.openai.com",
+    submit: {
+      method: "POST",
+      path: "/v1/videos",
+      buildBody: (params) => {
+        const { model, prompt, duration, extras } = params;
+        // seconds is a STRING enum ("4"|"8"|"12") in the Sora API; absolute
+        // size mapping is intentionally not forced here.
+        return {
+          model,
+          prompt,
+          ...(typeof duration === "number" ? { seconds: String(duration) } : {}),
+          ...extras,
+        };
+      },
+    },
+    taskIdPath: "id",
+    poll: { pathTemplate: "/v1/videos/{taskId}" },
+    statusPath: "status",
+    statusDone: ["completed"],
+    statusFailed: ["failed"],
+    resultPath: "data",
+    maxPolls: 60,
+    pollIntervalMs: 2000,
+  },
+};
+
+/** Resolve a configured job preset; null when the preset is unknown/none. */
+export function getVideoJobPreset(presetName: unknown): VideoJobPreset | null {
+  if (typeof presetName !== "string") return null;
+  const preset = VIDEO_JOB_PRESETS[presetName];
+  return preset ?? null;
+}
+
+/**
+ * Handle a video-generation job via the submit→poll preset pipeline.
+ * Returns the same shape as the sync handlers: { success, data?: …, status?, error? }.
+ */
+export async function handleVideoJobGeneration({
+  model,
+  presetName,
+  body,
+  credentials,
+  log,
+  maxPolls: maxPollsOverride,
+  pollIntervalMs: pollIntervalOverride,
+}: {
+  model: string;
+  presetName: string;
+  body: Record<string, unknown>;
+  credentials?: unknown;
+  log?: {
+    info?: (tag: string, msg: string, meta?: unknown) => void;
+    error?: (tag: string, msg: string) => void;
+  };
+  maxPolls?: number;
+  pollIntervalMs?: number;
+}) {
+  const preset = getVideoJobPreset(presetName);
+  if (!preset) {
+    return {
+      success: false,
+      status: 400,
+      error: `Unknown video job preset: ${presetName}`,
+    };
+  }
+
+  const baseUrl = resolveJobBaseUrl(credentials, preset.baseUrlFallback);
+  log?.info?.("VIDEO", `Job preset ${presetName} submitting ${model}`);
+  log?.info?.("VIDEO", JSON.stringify({ baseUrl }));
+
+  const bodyForPreset = preset.submit.buildBody({
+    model: model,
+    prompt: typeof body.prompt === "string" ? body.prompt : undefined,
+    duration: typeof body.duration === "number" ? body.duration : undefined,
+    // passthrough of the remainder — the API keeps catchall extras
+    extras: Object.fromEntries(
+      Object.entries(body ?? {}).filter(
+        ([key]) => key !== "model" && key !== "prompt" && key !== "duration"
+      )
+    ),
+  });
+
+  const submitPath = preset.submit.path.replace("{model}", encodeURIComponent(model));
+  const submitUrl = `${baseUrl}${submitPath}`; // baseUrl never ends with "/"
+  const submitResult = await fetchJson(submitUrl, {
+    method: preset.submit.method,
+    headers: buildJobHeaders(preset, credentials),
+    body: JSON.stringify(bodyForPreset),
+    log,
+  });
+  if (submitResult.ok === false) {
+    return { success: false, status: submitResult.status, error: submitResult.error };
+  }
+
+  const taskId = readStringPath(submitResult.data, preset.taskIdPath);
+  if (!taskId) {
+    return {
+      success: false,
+      status: 502,
+      error: `Video provider did not return a job id (${presetName})`,
+    };
+  }
+
+  // Poll loop.
+  const maxPolls = maxPollsOverride ?? preset.maxPolls;
+  const pollInterval = pollIntervalOverride ?? preset.pollIntervalMs;
+
+  for (let attempt = 1; attempt <= maxPolls; attempt += 1) {
+    await sleep(pollInterval);
+    const pollUrl = `${baseUrl}${preset.poll.pathTemplate.replace("{taskId}", encodeURIComponent(taskId))}`;
+    const pollResult = await fetchJson(pollUrl, {
+      method: "GET",
+      headers: buildJobHeaders(preset, credentials),
+      log,
+    });
+    if (pollResult.ok === false) {
+      return { success: false, status: pollResult.status, error: pollResult.error };
+    }
+
+    const status = readPath(pollResult.data, preset.statusPath);
+    const jobState = isDoneStatus(status, preset.statusDone, preset.statusFailed);
+    if (jobState === "done") {
+      const url = readResultUrl(pollResult.data, preset.resultPath);
+      if (!url) {
+        return {
+          success: false,
+          status: 502,
+          error: `Video job completed but no result URL found (${presetName})`,
+        };
+      }
+      log?.info?.("VIDEO", `Job completed after ${attempt} poll(s)`);
+      return {
+        success: true,
+        data: {
+          created: Math.floor(Date.now() / 1000),
+          data: [{ url, format: "mp4" }],
+        },
+      };
+    }
+    if (jobState === "failed") {
+      return {
+        success: false,
+        status: 502,
+        error: `Video job failed (${presetName})`,
+      };
+    }
+  }
+
+  return {
+    success: false,
+    status: 504,
+    error: `Video job timed out after ${maxPolls} polls (${presetName})`,
+  };
+}
+
+function buildJobHeaders(preset: VideoJobPreset, credentials?: unknown): Record<string, string> {
+  const creds = credentials as CredentialsLike | null | undefined;
+  const apiKey =
+    typeof creds?.apiKey === "string" && creds.apiKey
+      ? creds.apiKey
+      : typeof creds?.accessToken === "string" && creds.accessToken
+        ? creds.accessToken
+        : "";
+  const headers: Record<string, string> = { "Content-Type": "application/json" };
+  if (!apiKey) return headers;
+  if (preset.authScheme === "raw") {
+    headers[preset.authHeaderName] = apiKey;
+  } else {
+    headers[preset.authHeaderName] = `Bearer ${apiKey}`;
+  }
+  return headers;
+}
+
+function resolveJobBaseUrl(credentials: unknown, fallback: string): string {
+  const creds = credentials as CredentialsLike | null | undefined;
+  const psdBaseUrl =
+    creds?.providerSpecificData?.baseUrl != null &&
+    typeof creds.providerSpecificData.baseUrl === "string" &&
+    creds.providerSpecificData.baseUrl.trim()
+      ? (creds.providerSpecificData.baseUrl as string).trim()
+      : null;
+  const topLevelBaseUrl =
+    creds?.baseUrl != null && typeof creds.baseUrl === "string" && creds.baseUrl.trim()
+      ? (creds.baseUrl as string).trim()
+      : null;
+  const nodeBaseUrl = psdBaseUrl || topLevelBaseUrl;
+  if (!nodeBaseUrl) return fallback.replace(/\/+$/, "");
+  let normalized = nodeBaseUrl;
+  while (normalized.endsWith("/")) normalized = normalized.slice(0, -1);
+  return normalized;
+}
+
+async function fetchJson(
+  url: string,
+  {
+    method,
+    headers,
+    body,
+    log,
+  }: {
+    method: string;
+    headers: Record<string, string>;
+    body?: string;
+    log?: LogLike;
+  }
+): Promise<{ ok: true; data: unknown } | { ok: false; status: number; error: string }> {
+  try {
+    const response = await fetchWithTimeout(url, {
+      method,
+      headers,
+      ...(body !== undefined ? { body } : {}),
+      timeoutMs: getConfiguredTimeout(),
+    });
+    if (!response.ok) {
+      const errorText = await response.text();
+      log?.error?.("VIDEO", `Upstream ${response.status} for ${url}: ${errorText.slice(0, 200)}`);
+      return { ok: false, status: response.status, error: errorText };
+    }
+    const data = await response.json();
+    return { ok: true, data };
+  } catch (err: unknown) {
+    const message = err instanceof Error ? err.message : String(err);
+    const isTimeout =
+      err instanceof FetchTimeoutError || (err instanceof Error && err.name === "AbortError");
+    log?.error?.(
+      "VIDEO",
+      `${isTimeout ? "Timeout" : "Request error"} for ${url}: ${sanitizeErrorMessage(message)}`
+    );
+    return {
+      ok: false,
+      status: isTimeout ? 504 : 502,
+      error: `Video provider error: ${sanitizeErrorMessage(message)}`,
+    };
+  }
+}
+
+function readResultUrl(data: unknown, resultPath: string): string | null {
+  const found = readPath(data, resultPath);
+  if (typeof found === "string" && found.trim()) return found.trim();
+  if (Array.isArray(found)) {
+    const first = found[0];
+    // muapi-style: resultPath "outputs" resolves to ["https://…"].
+    if (typeof first === "string" && first.trim()) return first.trim();
+    // sora-style: resultPath "data" resolves to [{ url: "https://…" }].
+    if (first && typeof first === "object" && !Array.isArray(first)) {
+      const urlEntry = (first as Record<string, unknown>).url;
+      if (typeof urlEntry === "string" && urlEntry.trim()) return urlEntry.trim();
+    }
+    return null;
+  }
+  return null;
+}

open-sse/handlers/videoGeneration/openai.ts

@@ -0,0 +1,156 @@
+import {
+  fetchWithTimeout,
+  FetchTimeoutError,
+  getConfiguredTimeout,
+} from "@/shared/utils/fetchTimeout";
+import { saveCallLog } from "@/lib/usageDb";
+import { sanitizeErrorMessage } from "../../utils/error.ts";
+
+interface LogLike {
+  info?: (tag: string, msg: string, meta?: unknown) => void;
+  error?: (tag: string, msg: string) => void;
+}
+
+interface CredentialsLike {
+  providerSpecificData?: { baseUrl?: unknown } | null;
+  baseUrl?: unknown;
+  apiKey?: unknown;
+  accessToken?: unknown;
+}
+
+/**
+ * Resolve the video generation endpoint URL from credentials and fallback.
+ * Handles baseUrl from providerSpecificData or top-level credentials.
+ */
+function resolveVideoEndpoint(credentials: unknown, fallback: string): string {
+  const creds = credentials as CredentialsLike | null | undefined;
+  const psdBaseUrl =
+    creds?.providerSpecificData?.baseUrl != null &&
+    typeof creds.providerSpecificData.baseUrl === "string" &&
+    creds.providerSpecificData.baseUrl.trim()
+      ? creds.providerSpecificData.baseUrl.trim()
+      : null;
+  const topLevelBaseUrl =
+    creds?.baseUrl != null && typeof creds.baseUrl === "string" && creds.baseUrl.trim()
+      ? creds.baseUrl.trim()
+      : null;
+  const nodeBaseUrl = psdBaseUrl || topLevelBaseUrl;
+  let n = nodeBaseUrl;
+  while (n.endsWith("/")) n = n.slice(0, -1);
+  if (n.endsWith("/videos/generations")) return n;
+  return `${n}/videos/generations`;
+}
+
+/**
+ * Fetch the video generation endpoint with timeout and error handling.
+ */
+async function fetchVideoEndpoint(
+  url: string,
+  { headers, body, log }: { headers: Record<string, string>; body: string; log?: LogLike }
+) {
+  try {
+    const response = await fetchWithTimeout(url, {
+      method: "POST",
+      headers,
+      body,
+      timeoutMs: getConfiguredTimeout(),
+    });
+    if (!response.ok) {
+      const errorText = await response.text();
+      log?.error?.("VIDEO", `Upstream ${response.status} for ${url}: ${errorText}`);
+      return { success: false, status: response.status, error: errorText };
+    }
+    const data = await response.json();
+    return {
+      success: true,
+      data: { created: data.created || Math.floor(Date.now() / 1000), data: data.data || [] },
+    };
+  } catch (err) {
+    const message = err?.message;
+    const isTimeout = err instanceof FetchTimeoutError || err?.name === "AbortError";
+    log?.error?.(
+      "VIDEO",
+      `${isTimeout ? "Timeout" : "Request error"} for ${url}: ${sanitizeErrorMessage(message || err)}`
+    );
+    return {
+      success: false,
+      status: isTimeout ? 504 : 502,
+      error: `Video provider error: ${sanitizeErrorMessage(message || err)}`,
+    };
+  }
+}
+
+/**
+ * Handle OpenAI-compatible video generation.
+ * This handler is dispatched for custom providers with format "openai-video".
+ */
+export async function handleOpenAIVideoGeneration({
+  model,
+  provider,
+  providerConfig,
+  body,
+  credentials,
+  log,
+}: {
+  model: string;
+  provider: string;
+  providerConfig: { baseUrl: string; authHeader: string };
+  body: unknown;
+  credentials: unknown;
+  log?: LogLike;
+}) {
+  const startTime = Date.now();
+  const creds = credentials as CredentialsLike | null | undefined;
+  const apiToken = creds?.apiKey || creds?.accessToken;
+  const endpoint = resolveVideoEndpoint(credentials, providerConfig.baseUrl);
+  const headers = {
+    "Content-Type": "application/json",
+    ...(providerConfig.authHeader === "x-api-key"
+      ? { "x-api-key": String(apiToken) }
+      : { Authorization: `Bearer ${apiToken}` }),
+  };
+  const bodyObj = body as Record<string, unknown>;
+  const upstreamBody = {
+    model,
+    prompt: (bodyObj.prompt ?? "") as string,
+    ...(typeof bodyObj.duration === "number" && { duration: bodyObj.duration }),
+  };
+  const logRequestBody = {
+    model: bodyObj.model,
+    prompt:
+      typeof bodyObj.prompt === "string"
+        ? bodyObj.prompt.slice(0, 200)
+        : String(bodyObj.prompt ?? ""),
+    duration: bodyObj.duration,
+  };
+  log?.info?.("VIDEO", `OpenAI-compatible video generation: ${provider}/${model} -> ${endpoint}`, {
+    body: logRequestBody,
+  });
+
+  const fetchResult = await fetchVideoEndpoint(endpoint, {
+    headers,
+    body: JSON.stringify(upstreamBody),
+    log,
+  });
+
+  if (!fetchResult.success) {
+    return { success: false, status: fetchResult.status, error: fetchResult.error };
+  }
+
+  // Save call log for billing/tracking
+  await saveCallLog({
+    provider,
+    model: String(bodyObj.model),
+    endpoint: "video",
+    status: fetchResult.status,
+    durationMs: Date.now() - startTime,
+    tokensIn: 0,
+    tokensOut: 0,
+    requestId: null,
+  });
+
+  return {
+    success: true,
+    data: fetchResult.data,
+  };
+}

open-sse/handlers/videoGeneration/runwayHelpers.ts

@@ -0,0 +1,125 @@
+export function resolveRunwayPromptImage(body) {
+  const directCandidates = [
+    body.promptImage,
+    body.prompt_image,
+    body.image,
+    body.image_url,
+    body.imageUrl,
+    body.provider_options?.promptImage,
+    body.provider_options?.prompt_image,
+  ];
+
+  for (const candidate of directCandidates) {
+    if (typeof candidate === "string" && candidate.trim()) return candidate.trim();
+    if (candidate && typeof candidate === "object") return candidate;
+    if (Array.isArray(candidate) && candidate.length > 0) return candidate;
+  }
+
+  const arrayCandidates = [
+    body.imageUrls,
+    body.image_urls,
+    body.provider_options?.imageUrls,
+    body.provider_options?.image_urls,
+  ];
+  for (const candidate of arrayCandidates) {
+    if (Array.isArray(candidate) && candidate.length > 0) return candidate;
+  }
+
+  return null;
+}
+
+export function resolveRunwayRatio(body) {
+  const aspectRatio = typeof body.aspect_ratio === "string" ? body.aspect_ratio : body.aspectRatio;
+  if (aspectRatio === "1280:720" || aspectRatio === "720:1280") return aspectRatio;
+  if (aspectRatio === "16:9") return "1280:720";
+  if (aspectRatio === "9:16") return "720:1280";
+
+  const size = typeof body.size === "string" ? body.size : "";
+  const [widthRaw, heightRaw] = size.split("x");
+  const width = Number(widthRaw);
+  const height = Number(heightRaw);
+  if (Number.isFinite(width) && Number.isFinite(height) && width > 0 && height > 0) {
+    return width >= height ? "1280:720" : "720:1280";
+  }
+
+  return "1280:720";
+}
+
+export function resolveRunwayDuration(body) {
+  if (Number.isFinite(body.duration)) return clampRunwayDuration(body.duration);
+  if (Number.isFinite(body.frames) && Number.isFinite(body.fps) && Number(body.fps) > 0) {
+    return clampRunwayDuration(Number(body.frames) / Number(body.fps));
+  }
+  return 5;
+}
+
+function clampRunwayDuration(value) {
+  const duration = Math.round(Number(value));
+  if (!Number.isFinite(duration)) return 5;
+  return Math.min(10, Math.max(2, duration));
+}
+
+export function resolvePositiveInteger(value, fallback) {
+  const numeric = Number(value);
+  if (!Number.isFinite(numeric) || numeric <= 0) return fallback;
+  return Math.floor(numeric);
+}
+
+function extractRunwayOutputUrls(task) {
+  const rawOutput = Array.isArray(task?.output)
+    ? task.output
+    : Array.isArray(task?.result)
+      ? task.result
+      : [];
+  return rawOutput
+    .map((entry) => {
+      if (typeof entry === "string") return entry;
+      if (!entry || typeof entry !== "object") return null;
+      return entry.url || entry.uri || entry.videoUrl || entry.video_url || null;
+    })
+    .filter((value) => typeof value === "string" && value.length > 0);
+}
+
+export function extractRunwayFailureMessage(task) {
+  const directCandidates = [
+    task?.failure,
+    task?.failureReason,
+    task?.error,
+    task?.errorMessage,
+    task?.message,
+  ];
+  for (const candidate of directCandidates) {
+    if (typeof candidate === "string" && candidate.trim()) return candidate.trim();
+  }
+  if (task?.failure && typeof task.failure === "object") {
+    const nestedCandidates = [
+      task.failure.message,
+      task.failure.reason,
+      task.failure.error,
+      task.failure.code,
+    ];
+    for (const candidate of nestedCandidates) {
+      if (typeof candidate === "string" && candidate.trim()) return candidate.trim();
+    }
+  }
+  return null;
+}
+
+export async function normalizeRunwayVideoResult(task, body) {
+  const urls = extractRunwayOutputUrls(task);
+  if (urls.length === 0) {
+    throw new Error(
+      `Runway task completed without output URLs: ${JSON.stringify(task).slice(0, 400)}`
+    );
+  }
+  if (body.response_format === "url") return urls.map((url) => ({ url, format: "mp4" }));
+
+  const videos = [];
+  for (const url of urls) {
+    const response = await fetch(url);
+    if (!response.ok) throw new Error(`Runway output fetch failed (${response.status})`);
+    const arrayBuffer = await response.arrayBuffer();
+    videos.push({ b64_json: Buffer.from(arrayBuffer).toString("base64"), format: "mp4" });
+  }
+  return videos;
+}

open-sse/services/accountSemaphore.ts

@@ -54,21 +54,8 @@ export function buildAccountSemaphoreKey({
   return `${String(provider)}:${String(accountKey)}`;
 }
 
-/**
- * Effective positive cap, or null when the semaphore is bypassed (unset/<=0).
- *
- * Narrowing companion of {@link isBypassed}: that one returns a plain boolean, so
- * TypeScript cannot narrow `number | null` to `number` in its else-branch (a
- * `x is null | undefined` predicate would be unsound — 0 bypasses too). Callers
- * that need the VALUE after the guard go through here instead of casting.
- */
-function resolveActiveCap(maxConcurrency?: number | null): number | null {
-  if (maxConcurrency == null || maxConcurrency <= 0) return null;
-  return maxConcurrency;
-}
-
 function isBypassed(maxConcurrency?: number | null): boolean {
-  return resolveActiveCap(maxConcurrency) === null;
+  return maxConcurrency == null || maxConcurrency <= 0;
 }
 
 function createNoopReleaseFn(): () => void {
@@ -205,8 +192,7 @@ export function acquire(
     maxQueueSize = DEFAULT_MAX_QUEUE_SIZE,
   }: AcquireAccountSemaphoreOptions = {}
 ): Promise<() => void> {
-  const activeCap = resolveActiveCap(maxConcurrency);
-  if (activeCap === null) {
+  if (isBypassed(maxConcurrency)) {
     return Promise.resolve(createNoopReleaseFn());
   }
 
@@ -214,7 +200,9 @@ export function acquire(
     return Promise.reject(makeAbortError(signal));
   }
 
-  const gate = ensureGate(semaphoreKey, activeCap);
+  // isBypassed() above already excluded null/<=0 — ensureGate requires a plain
+  // number, but a boolean-returning helper isn't a type predicate TS can narrow on.
+  const gate = ensureGate(semaphoreKey, maxConcurrency as number);
   clearCleanupTimer(gate);
 
   if (gate.running < gate.maxConcurrency && !isBlocked(gate)) {

open-sse/services/adobeFireflyReferences.ts

@@ -0,0 +1,97 @@
+import { AdobeFireflyError } from "./adobeFireflyClient.ts";
+import type { AdobeFireflyVideoModelSpec } from "./adobeFireflyClient.ts";
+
+export interface AdobeSourceImageReference {
+  source: string;
+  usage?: string;
+  order?: number;
+}
+
+export function normalizeAdobeReferenceBlobs(
+  modelSpec: AdobeFireflyVideoModelSpec,
+  references: unknown
+): Array<{ id: string; usage: string; order?: number }> {
+  if (!Array.isArray(references)) return [];
+
+  const maxReferences = modelSpec.referenceMode === "image" ? 3 : 2;
+  if (references.length > maxReferences) {
+    throw new AdobeFireflyError(
+      `Adobe Firefly model accepts at most ${maxReferences} ${
+        modelSpec.referenceMode === "image" ? "asset" : "frame"
+      } image references`,
+      400,
+      "bad_image"
+    );
+  }
+
+  return references.map((reference, index) => {
+    if (!reference || typeof reference !== "object") {
+      throw new AdobeFireflyError("Invalid Adobe Firefly reference image", 400, "bad_image");
+    }
+    const value = reference as Record<string, unknown>;
+    const id = typeof value.id === "string" ? value.id.trim() : "";
+    if (!id) {
+      throw new AdobeFireflyError("Adobe Firefly reference image id is required", 400, "bad_image");
+    }
+
+    const expectedUsage = modelSpec.referenceMode === "image" ? "asset" : "frame";
+    const usage = typeof value.usage === "string" ? value.usage.trim() : expectedUsage;
+    if (usage !== expectedUsage) {
+      throw new AdobeFireflyError(
+        `Adobe Firefly model does not support image references with usage '${usage}'`,
+        400,
+        "bad_image"
+      );
+    }
+
+    return expectedUsage === "frame" ? { id, usage, order: index + 1 } : { id, usage };
+  });
+}
+
+export function extractAdobeSourceImageReferences(
+  body: unknown,
+  max = 4
+): AdobeSourceImageReference[] {
+  if (!body || typeof body !== "object") return [];
+  const inputs = (body as Record<string, unknown>).adobe_reference_inputs;
+  if (!Array.isArray(inputs)) return [];
+
+  const references: AdobeSourceImageReference[] = [];
+  for (const input of inputs) {
+    if (!input || typeof input !== "object") continue;
+    const value = input as Record<string, unknown>;
+    if (
+      value.type !== undefined &&
+      value.type !== "input_image" &&
+      value.type !== "image" &&
+      value.type !== "image_url"
+    ) {
+      continue;
+    }
+
+    const imageUrl = value.image_url;
+    const source =
+      typeof value.source === "string"
+        ? value.source.trim()
+        : typeof imageUrl === "string"
+          ? imageUrl.trim()
+          : imageUrl &&
+              typeof imageUrl === "object" &&
+              typeof (imageUrl as Record<string, unknown>).url === "string"
+            ? String((imageUrl as Record<string, unknown>).url).trim()
+            : typeof value.url === "string"
+              ? value.url.trim()
+              : "";
+    if (!source || (!source.startsWith("data:image/") && !/^https?:\/\//i.test(source))) continue;
+
+    const usage =
+      typeof value.usage === "string" && value.usage.trim() ? value.usage.trim() : undefined;
+    const order =
+      typeof value.order === "number" && Number.isInteger(value.order) && value.order > 0
+        ? value.order
+        : undefined;
+    references.push({ source, ...(usage ? { usage } : {}), ...(order ? { order } : {}) });
+    if (references.length >= max) break;
+  }
+  return references;
+}

open-sse/services/adobeFireflySecurity.ts

@@ -0,0 +1,54 @@
+const ADOBE_JWT_IN_TEXT_REGEX =
+  /eyJ[A-Za-z0-9_-]{1,4096}\.[A-Za-z0-9_-]{1,4096}\.[A-Za-z0-9_-]{1,4096}/;
+const ADOBE_JWT_IN_TEXT_GLOBAL_REGEX =
+  /eyJ[A-Za-z0-9_-]{1,4096}\.[A-Za-z0-9_-]{1,4096}\.[A-Za-z0-9_-]{1,4096}/g;
+const ADOBE_JWT_EXACT_REGEX =
+  /^eyJ[A-Za-z0-9_-]{1,4096}\.[A-Za-z0-9_-]{1,4096}\.[A-Za-z0-9_-]{1,4096}$/;
+const FIREFLY_3P_HOST_SUFFIX = "firefly-3p.ff.adobe.io";
+
+export function decodeAdobeJwtPayload(token: string): Record<string, unknown> | null {
+  try {
+    let raw = String(token || "")
+      .trim()
+      .replace(/^bearer\s+/i, "")
+      .trim();
+    const match = raw.match(ADOBE_JWT_IN_TEXT_REGEX);
+    if (match) raw = match[0];
+    const part = raw.split(".")[1];
+    if (!part) return null;
+    const json = Buffer.from(part.replace(/-/g, "+").replace(/_/g, "/"), "base64").toString("utf8");
+    const value: unknown = JSON.parse(json);
+    return value && typeof value === "object" ? (value as Record<string, unknown>) : null;
+  } catch {
+    return null;
+  }
+}
+
+export function findAllAdobeJwts(value: string): string[] {
+  return value.match(ADOBE_JWT_IN_TEXT_GLOBAL_REGEX) ?? [];
+}
+
+export function isExactAdobeJwt(value: string): boolean {
+  return ADOBE_JWT_EXACT_REGEX.test(value);
+}
+
+export function stripAdobeJwts(value: string, replacement = ""): string {
+  return value.replace(ADOBE_JWT_IN_TEXT_GLOBAL_REGEX, replacement);
+}
+
+function hostnameMatches(hostname: string, expected: string): boolean {
+  const normalized = hostname.toLowerCase().replace(/\.$/, "");
+  return normalized === expected || normalized.endsWith(`.${expected}`);
+}
+
+export function isAdobeFireflyApiUrl(rawUrl: string): boolean {
+  try {
+    return hostnameMatches(new URL(rawUrl).hostname, FIREFLY_3P_HOST_SUFFIX);
+  } catch {
+    return false;
+  }
+}
+
+export function isAdobeLoginCookieDomain(domain: string): boolean {
+  return hostnameMatches(domain.replace(/^\./, ""), "adobelogin.com");
+}

open-sse/services/adobeFireflyUpscale.ts

@@ -156,9 +156,10 @@ export function resolveAdobeCreativityLevel(opts: {
     return clampLevel(normalizeExplicitCreativity(Number(explicit)));
   }
 
-  const percent = typeof opts.creativityPercent === "number" && Number.isFinite(opts.creativityPercent)
-    ? Math.max(0, Math.min(100, opts.creativityPercent))
-    : 0;
+  const percent =
+    typeof opts.creativityPercent === "number" && Number.isFinite(opts.creativityPercent)
+      ? Math.max(0, Math.min(100, opts.creativityPercent))
+      : 0;
   return clampLevel(percent / 100);
 }
 
@@ -265,11 +266,7 @@ export async function adobeFireflyUpscaleImage(opts: {
 
   const blobId = String(opts.blobId || "").trim();
   if (!blobId) {
-    throw new AdobeFireflyError(
-      "Adobe Firefly upscale requires a source image",
-      400,
-      "bad_image"
-    );
+    throw new AdobeFireflyError("Adobe Firefly upscale requires a source image", 400, "bad_image");
   }
 
   const factor = normalizeFactor(opts.upsamplerFactor, spec.factors);

open-sse/services/antigravityProjectPersist.ts

@@ -39,7 +39,7 @@ export function preferAntigravityConnectionsWithStoredProject<T extends Record<s
 ): T[] {
   if (!Array.isArray(connections) || connections.length === 0) return connections;
   const hasStoredProject = (connection: T): boolean => {
-    if (typeof connection.projectId === "string" && connection.projectId) return true;
+    if (typeof connection.projectId === "string" && connection.projectId.trim()) return true;
     let psd = connection.providerSpecificData;
     if (typeof psd === "string") {
       try {
@@ -48,12 +48,9 @@ export function preferAntigravityConnectionsWithStoredProject<T extends Record<s
         return false;
       }
     }
-    return Boolean(
-      psd &&
-      typeof psd === "object" &&
-      typeof (psd as Record<string, unknown>).projectId === "string" &&
-      (psd as Record<string, unknown>).projectId
-    );
+    if (!psd || typeof psd !== "object") return false;
+    const projectId = (psd as Record<string, unknown>).projectId;
+    return typeof projectId === "string" && projectId.trim().length > 0;
   };
   const withStoredProject = connections.filter(hasStoredProject);
   return withStoredProject.length > 0 ? withStoredProject : connections;

open-sse/services/antigravityProjectPersistence.ts

@@ -1,13 +1,39 @@
 /**
  * Re-export from `antigravityProjectPersist.ts` plus a connection-preference helper.
+ *
+ * The persistence layer for a runtime-discovered Antigravity projectId lives in
+ * the sibling file `antigravityProjectPersist.ts` (named by its core function).
+ * This module adds `preferAntigravityConnectionsWithStoredProject()`, used by the
+ * quota-strategy engine to give priority to connections whose projectId has
+ * already been discovered and persisted.
  */
+
 import { persistDiscoveredAntigravityProjectId } from "./antigravityProjectPersist.ts";
+
 export { persistDiscoveredAntigravityProjectId };
 
-export function preferAntigravityConnectionsWithStoredProject(
-  connections: Array<Record<string, unknown>>
-): Array<Record<string, unknown>> {
-  return connections.filter(
-    (conn) => conn != null && typeof conn.projectId === "string" && conn.projectId.trim().length > 0
-  );
+/**
+ * Prefer Antigravity connections with a discovered/stored `projectId` for
+ * reset-aware quota routing.
+ *
+ * This is a preference, not a hard requirement: when no candidate has a stored
+ * projectId, retain the full pool rather than making freshly-added accounts unusable.
+ */
+function hasStoredProjectId(connection: Record<string, unknown>): boolean {
+  if (typeof connection.projectId === "string" && connection.projectId.trim().length > 0) {
+    return true;
+  }
+  const providerSpecificData = connection.providerSpecificData;
+  if (providerSpecificData && typeof providerSpecificData === "object") {
+    const nested = (providerSpecificData as Record<string, unknown>).projectId;
+    if (typeof nested === "string" && nested.trim().length > 0) return true;
+  }
+  return false;
+}
+
+export function preferAntigravityConnectionsWithStoredProject<T extends Record<string, unknown>>(
+  connections: T[]
+): T[] {
+  const withStoredProject = connections.filter(hasStoredProjectId);
+  return withStoredProject.length > 0 ? withStoredProject : connections;
 }

open-sse/services/claudeAdaptiveThinking.ts

@@ -54,7 +54,7 @@ export function normalizeClaudeAdaptiveThinking<T extends Record<string, unknown
   delete nextThinking.budget_tokens;
   delete nextThinking.max_tokens;
 
-  return { ...record, thinking: nextThinking } as T;
+  return { ...body, thinking: nextThinking };
 }
 
 /**
@@ -84,7 +84,7 @@ export function normalizeClaudeDisabledThinkingEffort<T extends Record<string, u
   }
 
   return {
-    ...record,
+    ...body,
     output_config: { ...outputConfig, effort: disabledEffortCap },
-  } as T;
+  };
 }

open-sse/services/combo.ts

@@ -3020,7 +3020,8 @@ async function handleRoundRobinCombo({
       return new Response(
         JSON.stringify({
           error: {
-            message: "Service temporarily unavailable: all targets were skipped by pre-dispatch filters",
+            message:
+              "Service temporarily unavailable: all targets were skipped by pre-dispatch filters",
             type: "service_unavailable",
             code: "ALL_TARGETS_SKIPPED",
           },

open-sse/services/combo/comboStructure.ts

@@ -138,9 +138,7 @@ function normalizeRuntimeStep(
       : {}),
     weight,
     label,
-    // `prompt` is a per-step pipeline input and only exists on a model step —
-    // #8894 widened the union with ComboProviderWildcardStep, which has no prompt.
-    prompt: (step.kind === "model" ? step.prompt : null) || null,
+    prompt: step.kind === "model" ? step.prompt || null : null,
   } satisfies ResolvedComboTarget;
 }
 
@@ -485,13 +483,6 @@ function estimateRequestInputTokens(body: Record<string, unknown>): number {
 }
 
 function valueContainsImagePart(value: unknown): boolean {
-  // Delegates to the unified media detector (open-sse/utils/mediaParts.ts) —
-  // single source of truth shared with the vision-bridge guardrail. The
-  // detector keeps this filter's legacy permissive matches (image-ish `type`
-  // in any casing, bare `image_url`/`input_image` keys, source.media_type
-  // image/*, bare data:image strings, recursion capped at depth 8) via
-  // "image_indicator" parts. containsMediaKind short-circuits on the first
-  // hit — this runs on every request, so no full-part collection here.
   return containsMediaKind([{ content: [value] }], "image");
 }
 
@@ -615,6 +606,10 @@ export type CompatFilterOptions = {
   failOpen?: boolean;
 };
 
+export function hasHardCapabilityFailure(reasons: string[]): boolean {
+  return reasons.some((reason) => HARD_COMPAT_REASONS.has(reason));
+}
+
 /**
  * Summarize a capability-filter exhaustion for a 400-class combo error (#8488).
  * Returns null when the empty pool is not attributable to hard requirements.
@@ -718,9 +713,7 @@ export function filterTargetsByRequestCompatibility(
 
   if (compatible.length === targets.length) return targets;
   if (compatible.length === 0) {
-    const hardRejected = rejected.some((entry) =>
-      entry.reasons.some((r) => HARD_COMPAT_REASONS.has(r))
-    );
+    const hardRejected = rejected.some((entry) => hasHardCapabilityFailure(entry.reasons));
     const failOpen = options?.failOpen === true;
 
     log.debug?.(

open-sse/services/combo/fusionPanel.ts

@@ -51,9 +51,9 @@ export function extractFusionPanelSpec(
       panel.push(step.comboName);
       return;
     }
-    // #8894 widened ComboStep with ComboProviderWildcardStep, which carries a
-    // modelPattern instead of a model. getComboModelString() already resolves any
-    // step shape (and returns null for the ones with no concrete model id).
+    // Provider-wildcard steps have no concrete model to dispatch — fusion is a
+    // fixed-size panel of literal models/combo-refs, not a wildcard-expanding
+    // strategy (see file header). Skip rather than push an undefined model.
     const modelStr = getComboModelString(step);
     if (modelStr) panel.push(modelStr);
   });

open-sse/services/combo/quotaStrategies.ts

@@ -89,9 +89,7 @@ async function getQuotaAwareConnectionsForTarget(
               ? (connections as Array<Record<string, unknown>>)
               : [];
             if (provider === "antigravity" || provider === "agy") {
-              activeConnections = preferAntigravityConnectionsWithStoredProject(
-                activeConnections
-              ) as Array<Record<string, unknown>>;
+              activeConnections = preferAntigravityConnectionsWithStoredProject(activeConnections);
             }
             if (
               !resetAwareConnectionCache.has(provider) &&

open-sse/services/compression/engines/cavemanAdapter.ts

@@ -221,6 +221,14 @@ const LITE_SCHEMA: EngineConfigField[] = [
     label: "Preserve system prompt",
     defaultValue: true,
   },
+  {
+    key: "compressToolResults",
+    type: "boolean",
+    label: "Proactively truncate long tool results",
+    description:
+      "Truncates tool results over 2,000 characters during Lite compression. Emergency overflow protection may still trim content when the context exceeds the model budget.",
+    defaultValue: true,
+  },
 ];
 
 function validateLiteConfig(config: Record<string, unknown>): EngineValidationResult {
@@ -231,6 +239,7 @@ function validateLiteConfig(config: Record<string, unknown>): EngineValidationRe
   ) {
     errors.push("preserveSystemPrompt must be a boolean");
   }
+  validateBoolean(config, "compressToolResults", errors);
   return { valid: errors.length === 0, errors };
 }
 
@@ -253,9 +262,17 @@ export const liteEngine: CompressionEngine = {
   },
   apply(body, options) {
     const adapter = adaptBodyForCompression(body);
+    const stepCompressToolResults = options?.stepConfig?.compressToolResults;
     const result = applyLiteCompression(adapter.body, {
       ...options,
       preserveSystemPrompt: options?.config?.preserveSystemPrompt !== false,
+      // buildStepOptions() already merges global config.lite with explicit step.config
+      // (step wins) into stepConfig, so consume that single effective value instead of
+      // AND-ing root and step values — an explicit step `true` must override a global `false`.
+      compressToolResults:
+        typeof stepCompressToolResults === "boolean"
+          ? stepCompressToolResults
+          : (options?.config?.lite?.compressToolResults ?? true),
     });
     return adapter.adapted ? { ...result, body: adapter.restore(result.body) } : result;
   },

open-sse/services/compression/engines/ccr/index.ts

@@ -35,7 +35,6 @@
  *   - Only replace blocks ≥ minChars (default 600).
  *   - `stackable: true`, `stackPriority: 4` (runs just after session-dedup(3)).
  */
-
 import crypto from "node:crypto";
 import {
   deleteAllCcrBlocks,
@@ -292,8 +291,10 @@ function rehydrateEntry(hash: string, principalId: string, now: number): CcrEntr
 
   // Re-admit through the same budgets a fresh store would face. If the block no longer
   // fits, it stays on disk and is served straight from the row instead of being cached.
-  const { principalId: owner, bytes } = entry;
-  if (enforcePrincipalBudget(owner, bytes) && enforceGlobalBudget(owner, bytes)) {
+  if (
+    enforcePrincipalBudget(entry.principalId, entry.bytes) &&
+    enforceGlobalBudget(entry.principalId, entry.bytes)
+  ) {
     const key = buildStoreKey(hash, principalId === ANON ? undefined : principalId);
     ccrStore.set(key, entry);
     ccrTotalBytes += entry.bytes;

open-sse/services/compression/lite.ts

@@ -17,6 +17,7 @@ interface LiteCompressionOptions {
   model?: string;
   supportsVision?: boolean | null;
   preserveSystemPrompt?: boolean;
+  compressToolResults?: boolean;
 }
 
 function trimTrailingHorizontalWhitespace(line: string): string {
@@ -253,9 +254,11 @@ export function applyLiteCompression(
   current = r2.body;
   if (r2.applied) techniquesApplied.push("system-dedup");
 
-  const r3 = compressToolResults(current);
-  current = r3.body;
-  if (r3.applied) techniquesApplied.push("tool-compress");
+  if (options?.compressToolResults !== false) {
+    const r3 = compressToolResults(current);
+    current = r3.body;
+    if (r3.applied) techniquesApplied.push("tool-compress");
+  }
 
   const r4 = removeRedundantContent(current, options);
   current = r4.body;