* feat(db): a2a task history module over migration-002 tables
* feat(a2a): persist task lifecycle to a2a_tasks with 30d retention purge
* feat(api): a2a task history listing + historical detail fallback
* feat(dashboard): orchestration History tab (Airflow-grid) over persisted runs (2.2)
* fix(dashboard): assert history preset window + loading and time axis in the grid
* chore(dashboard): history i18n + changelog
* fix(dashboard): explicit history purge cascade + keep live drawer off the History tab
* docs: document OMNIROUTE_A2A_HISTORY_RETENTION_DAYS
* refactor(dashboard): split HistoryTab helpers under the complexity ratchet
---------
Co-authored-by: Markus Hartung <diegosouzapw@users.noreply.github.com>
Migrates the Claude, Grok, LMArena, Notion and Perplexity web-cookie transports from the tls-client-node/Koffi sidecar to the exactly pinned wreq-js 3.2.0 runtime, keeping the per-provider browser/OS profiles, making request cookies ephemeral, bounding and generation-protecting the shared native transport pool, removing the legacy downloader and repair path, and carrying the native binding and license evidence through the npm, standalone, Electron, Docker and Bun packaging surfaces.
This is the consolidation of the two competing migrations, and the consolidation was decided by evidence rather than by preference. #11753's six suites were installed over this implementation and run as an independent specification: 31 of 36 passed. All five failures are artefacts of #11753 being the older design, not coverage gaps —
- two hardcode the 3.0.0 pin in their assertions (this branch pins 3.2.0, which is what the release tip already resolves; #11753's 3.0.0 would have conflicted);
- one reads open-sse/services/chatgptTlsClient.ts, deleted when #11754 retired ChatGPT Web, so the test is stale against the current tip;
- two import WREQ_JS_NATIVE_BINARY_NAMES / resolveWreqJsNativeBinaryName, which this branch redesigned into WREQ_JS_NATIVE_BINDINGS / resolveWreqJsNativeBinding plus WREQ_JS_VERSION — a rename from modelling natives as file names to modelling them as package bindings, verified as an API difference rather than a lost capability (the linux-x64-gnu .node is present and serviceable).
This branch is also the strict superset by scope: 7 files exclusive to it, including the wreq-js Rust license inventory and notices, .trivyignore, open-sse/utils/tlsClient.ts and assembleStandalone.mjs. #11753 had one exclusive file, its changelog fragment. Nothing needed porting, so #11753 is superseded rather than merged, and the changelog entry credits both.
Reconciled on merge: clean against the tip. The new migration suite (tests/unit/tls-client-wreq-migration.test.ts, 1374 lines, 31 cases) is frozen at its exact LOC with the rationale — it shares one native-transport harness, so splitting it mid-merge would duplicate that harness for no coverage gain. Verified that no existing cap moves.
Verified: 182/182 across the eight TLS, native-manifest, postinstall, standalone-bundle, pack-artifact and provider-validation suites, typecheck:core clean, check:cycles OK, check-changelog-integrity OK, check-file-size OK, and every changed TypeScript file parses.
Restores ChatGPT Web on a clean-room browser transport, merged on the operator's explicit decision.
Worth stating precisely, because this touches a provenance decision: the PR does not revert #11754. It narrows RETIRED_COMMON_CHATGPT_WEB_PROVIDER_IDS to the single GPL-derived alias cgpt-web and registers chatgpt-web as a separate clean-room id. The old implementation stays retired and blocked; the retirement machinery, its error code and its 410 contract are untouched. All four retirement suites agree with that distinction and pass unchanged.
The 44 protected agent-instruction surfaces this PR touches (AGENTS.md, llm.txt and its 42 mirrors, README) were verified rather than trusted: masking digits and comparing the removed and added line sets gives 264 lines on each side, identical — every change is a provider-count substitution, with no sentence added, removed or reworded.
Reconciled on merge: clean against the tip, with the two chat chokepoints this PR grows (src/sse/handlers/chat.ts +40, open-sse/handlers/chatCore.ts +30) recorded in the file-size baseline under an annotation. Verified that exactly those two caps move and nothing else, so the #12411 ratchet holds. The rebaseline is carried on this branch rather than left in a validation worktree — the propagation mistake that put the 2026-09-02 merge waves base-red in #12434.
Verified: 504/504 across the PR's 44 test files plus all four chatgpt-web retirement suites, check:provider-consistency OK (272 REGISTRY entries, 355 canonical providers), check-file-size OK, and every changed TypeScript file parses.
Thanks @backryun — separating the clean-room id from the retired alias, instead of reopening the old one, is what made this reviewable.
P1 of #12150: transcript text no longer reaches call logs or durable memory in the clear.
Reconciled on merge — the two persisted-requestBody assertions were failing, and the failure signature was misleading enough to be worth recording. They reported "expected: true, actual: false", which reads like the redaction not applying. It was not: pollForCallLog waited at most 120 tries x 20ms = 2.4s for the asynchronous SQLite write, then returned null, so assert.ok(row) failed before any redaction assertion ran. The observed durations were 5253ms and 4467ms against that 2.4s ceiling — a starved runner, not a leak. The control test failing alongside the positive one was the tell: a real redaction defect would break one direction, not both.
Replaced the fixed try count with a 30s wall-clock deadline: far past any healthy write, still bounded, and a fast machine still returns on the first pass. A privacy test should not depend on how busy the box is.
Verified: 4/4 three consecutive times under synthetic load, and 3/3 unloaded beforehand. Note the synthetic load reached ~7, below the ~38 where the original failure appeared, but the budget is now 12.5x larger and deadline-based rather than count-based.
Three regressions inherited by every PR rebased onto release/v3.8.51, caught and documented with the exact failing output.
The one that mattered most: src/shared/providers/webSessionCredentials.ts did not parse. The UC merge (#11513) inserted the uc: entry inside maxai.storageKeys and lost the array's closing ], plus the entry's }, leaving `ERROR: Expected "]" but found ":"` at line 351. That module is imported by the provider API routes, bulk-web-session, autoCombo's virtualFactory, keepaliveThreshold and dashboard components, so the break was live on the tip and flooded unrelated catalog tests with transform failures. That was my conflict resolution, not the contributor's code — thank you for catching it and for tracing it to the root commit rather than patching around the symptom.
Also fixed: the duplicate bin/cli/utils/volatileEnvPath.mjs entry in PACK_ARTIFACT_REQUIRED_PATHS (findMissingArtifactPaths reported it twice), and UC image models made prefix-addressable without letting them claim historical bare model ids belonging to other providers.
Reconciled on merge: #12394 landed the busy_timeout/probe work first, so src/lib/db/core.ts takes the tip's side. probeUtils.ts is the union of both rather than either side — this PR's message regex is wider (SQLite also reports "database table is locked", "database schema is locked" and "database is busy"), while #12394 added the driver code/errcode path that keeps a transient lock from being classified as corruption and renaming the database away. Taking either alone would have dropped the other half; this PR's own ENOENT test is what surfaced it.
Verified: 76/76 across uc-image, probe-9541-repro, web-session-contract, pack-artifact-policy, bulk-web-session-import and exclusive-connection-leases, and every changed .ts file parses.
Thanks @backryun.
GET /v1/providers/gemini-business/models returned nothing because gemini-business had no RegistryEntry: the listing route resolves the provider through getRegistryEntry and filters the unified catalog by owned_by, and open-sse/config/providers/index.ts only registered gemini and gemini-web.
Adds a registry entry mirroring gemini_webProvider — id gemini-business, alias gembiz, cookie auth — with the twelve ids from the executor's MODEL_CATEGORY_MAP. Each model is declared toolCalling: false, supportsReasoning: false, the same live-behaviour contract applied to gemini-web in #9356: the executor returns plain text, hard-wires the thinking mode and parses no tool calls.
Reconciled on merge: the only conflict was the reserved-prefix count assertion, which the tip had moved. Took the tip's text and measured the real value with this PR applied — 406 to 408, the gemini-business id plus its gembiz alias — rather than carrying the branch's number.
Validated in a combined worktree with all 25 PRs of this batch boarded together (typecheck:core clean, 443/443 node-runner plus 14/14 vitest, all static gates green), and re-verified standalone on the current tip after the other 24 landed: 33/33 across provider-node-reserved-prefix, gemini-business-model-registry-12107 and web-cookie-validation-fallback, with check:provider-consistency OK at 272 REGISTRY entries and 355 canonical providers.
Thanks @pacocartones.
The Leaderboard rendered apiKeyId.slice(0, 8)… under a column translated as "name". The route now enriches each entry with the key's display name — route-local, so the shared getTopN helper and the federation leaderboard stay id-only — and the page renders name ?? shortId with the full id in a title attribute. The lookup selects only id and name from api_keys, chunked at 200 ids, with unknown ids and blank names omitted; no key material leaves the DB layer.
Validated in a combined worktree with all 25 PRs of this batch boarded together: typecheck:core clean, 443/443 node-runner tests plus 14/14 vitest across every test file the batch touches, and check-changelog-integrity, check:cycles (418 files), check:provider-consistency (272 REGISTRY entries, 355 canonical providers), check:docs-counts, check:docs-sync (42 locales) and check-file-size all green.
Thanks @pacocartones.
The Profile page already rendered a streak card but fed it a hard-coded useState(0) with a "streak data comes from future API" note — while streaks.ts tracked per-key streaks all along and the MCP gamification_profile tool already returned them. GET /api/gamification/level now returns streak: { current, longest } next to level: the key's own streak with apiKeyId, the operator-wide maximum otherwise, matching the aggregate mode getAggregateXp uses (#3484). No new route, no OpenAPI change, no new i18n keys; a missing or zero streak keeps the card hidden exactly as before.
Validated in a combined worktree with all 25 PRs of this batch boarded together: typecheck:core clean, 443/443 node-runner tests plus 14/14 vitest across every test file the batch touches, and check-changelog-integrity, check:cycles (418 files), check:provider-consistency (272 REGISTRY entries, 355 canonical providers), check:docs-counts, check:docs-sync (42 locales) and check-file-size all green.
Thanks @pacocartones.
getDbInstance() ran PRAGMA journal_mode = WAL as the connection's first statement, before PRAGMA busy_timeout, and openSqliteDatabase() passes no driver-level timeout. A process opening the database while another closed its WAL connection — checkpoint plus WAL delete hold an EXCLUSIVE lock for a few hundred microseconds — therefore died with `database is locked` instead of waiting. That is the flake behind exclusive-connection-leases.test.ts on release/v3.8.51 runs 33525300898 and 33493797519 and on unrelated PR runs.
The second half is worse than the flake: isTransientProbeError matched /SQLITE_BUSY/ against error.message, but both drivers report the plain text `database is locked` and put the code in .code / .errcode. A transient lock during the corruption probe therefore took the corrupt-database path and renamed the file to storage.sqlite.probe-failed-… with "Manual recovery required". The probe now recognises the drivers' real BUSY/PROTOCOL/IOERR signals.
Validated in a combined worktree with all 25 PRs of this batch boarded together: typecheck:core clean, 443/443 node-runner tests plus 14/14 vitest across every test file the batch touches, and check-changelog-integrity, check:cycles (418 files), check:provider-consistency (272 REGISTRY entries, 355 canonical providers), check:docs-counts, check:docs-sync (42 locales) and check-file-size all green.
Thanks @pacocartones.
validateScoreChange() — the documented 1000 XP/min per-API-key limit plus the velocity anomaly check — was exported but never called, so the award path applied every XP delta unconditionally. It now runs before addXp; a rejected award is logged at warn level and skipped, and the fire-and-forget path never throws.
The second finding is the one that made the first invisible: getRecentXp's window query was inert. created_at is stored by the table default as YYYY-MM-DD HH:MM:SS and was compared lexically against a JS ISO string, so same-day rows never matched and the limit could not have tripped even if it had been wired. The window start is now computed in SQLite, matching the style computeZScore already used.
Validated in a combined worktree with all 25 PRs of this batch boarded together: typecheck:core clean, 443/443 node-runner tests plus 14/14 vitest across every test file the batch touches, and check-changelog-integrity, check:cycles (418 files), check:provider-consistency (272 REGISTRY entries, 355 canonical providers), check:docs-counts, check:docs-sync (42 locales) and check-file-size all green.
Thanks @pacocartones.
Four files embedded the U+0000 separator of a memo/group key as a raw NUL byte rather than the \\0 escape the codebase uses for the same idiom elsewhere. The runtime value is identical, but the raw byte trips the binary heuristics of git, GitHub and ripgrep: git diff --numstat reported `- -`, the introducing PRs rendered three of the files as "Binary file not shown", and rg silently skipped them in recursive mode. Rewritten as escapes, with a guard test keeping raw NUL bytes out of src/, open-sse/ and tests/.
Validated in a combined worktree with all 25 PRs of this batch boarded together: typecheck:core clean, 443/443 node-runner tests plus 14/14 vitest across every test file the batch touches, and check-changelog-integrity, check:cycles (418 files), check:provider-consistency (272 REGISTRY entries, 355 canonical providers), check:docs-counts, check:docs-sync (42 locales) and check-file-size all green.
Thanks @pacocartones.
getBestVisionModel() validates a configured fixedModel with hasUsableCredentialsForModel() before short-circuiting (#8430). auto / auto/* ids are virtual combos with no provider row, so that check always reported a confirmed false and the combo was silently discarded in favour of global auto-selection — it never got the chance to rotate its members. This mirrors the exemption the reroute guard in visionBridge.ts already carries; concrete fixedModel ids keep the #8430 fall-through unchanged.
Validated in a combined worktree with all 25 PRs of this batch boarded together: typecheck:core clean, 443/443 node-runner tests plus 14/14 vitest across every test file the batch touches, and check-changelog-integrity, check:cycles (418 files), check:provider-consistency (272 REGISTRY entries, 355 canonical providers), check:docs-counts, check:docs-sync (42 locales) and check-file-size all green.
Thanks @pacocartones.
The opt-in Codex quota auto-ping pinned gpt-5.1-codex-mini. OpenAI shut that model down on 2026-07-23 and the repo's own lifecycle registry already rejects it on the request path, but the scheduler never consulted that gate — every window slide sent a dead id, hit the 15-minute failure cooldown, and retried the same id forever. The ping model now resolves per tick from the provider catalog through isModelSelectable(), the same gate chatCore uses, with the registry import kept lazy because this module sits on the instrumentation boot path (#12074). When nothing is selectable the provider is paused before any throttle slot, usage read or executor call, with one warning per state change.
Validated in a combined worktree with all 25 PRs of this batch boarded together: typecheck:core clean, 443/443 node-runner tests plus 14/14 vitest across every test file the batch touches, and check-changelog-integrity, check:cycles (418 files), check:provider-consistency (272 REGISTRY entries, 355 canonical providers), check:docs-counts, check:docs-sync (42 locales) and check-file-size all green.
Thanks @pacocartones.
Claude Code (claude / cc) is correctly classified as a flat-rate subscription, so the analytics API reports $0 — accurate as billed cost, and useless as a view of what the subscription actually consumed. Neither the Costs nor the Analytics dashboard had a token-price-equivalent view.
The fix keeps both meanings rather than picking one: ordinary analytics callers keep billed-cost semantics ($0 for flat-rate), /dashboard/costs and /dashboard/analytics opt in explicitly via includeFlatRateEstimates=true, the response reports whether estimates were included so a caller cannot mistake them for vendor billing records, and the figures on /dashboard/costs are labelled as flat-rate estimates rather than presented as spend. Omitted, false and unknown values all retain the existing behaviour.
Scope note carried from the description: this is a checkpoint on #11459, not its full closure — the issue stays open.
Verified in a combined worktree with three sibling PRs of this batch: typecheck:core clean, 134/134 focused tests (4 skipped), and i18n UI coverage PASS across all 42 locales for the 43-file locale pass.
One cross-PR interaction worth recording, since it is invisible from either side: this grows CostOverviewTab.tsx from 1282 to 1318 lines, which is fine against the tip's current 2002 cap but exceeds the 1283 that #12411 (file-size ratchet re-tightening) would freeze. Neither PR fails alone. Merged first on purpose so #12411's mechanical --update recomputes against the real post-merge LOC — the cap still only goes down.
Thanks @xiaoyaner0201 — the opt-in contract plus the "were estimates included" flag is the right shape for this.
The contact sheet, the dedup comparator and the drill-down each validated JPEG data-URIs independently. They now share src/lib/guardrails/videoBridgeFrameContract.ts. No behaviour change; the sibling tests that asserted a per-module message were aligned to the shared one. Closes the Standards-4 residue from the 2026-08-18 Video Bridge review.
Verified in a combined worktree with three sibling PRs of this batch: typecheck:core clean, 134/134 focused tests (4 skipped), i18n UI coverage PASS across all 42 locales.
On dashboard/memory?tab=engine the Embedding Model quick-select (and the rerank selector) built their lists from a keyword heuristic over the CHAT catalog (AI_MODELS) plus OpenRouter live discovery. Providers whose embedding models are not in that catalog never appeared — mistral, gemini, nvidia nim, groq, vercel-ai-gateway and others that serve embeddings on a standard OpenAI-compatible /embeddings endpoint — and typing such a model by hand failed at runtime with "Unknown embedding provider".
The fix is one generic mechanism rather than a list of per-provider patches: deriveEmbeddingProviderForChatProvider() turns any chat-registry entry with a /chat/completions base into an OpenAI-compatible /embeddings config, with curated EMBEDDING_PROVIDERS entries always winning; the embeddings service resolves a derived config for unknown-but-configured providers instead of rejecting them; deriveRerankProviderForChatProvider() does the same for Cohere-compatible /rerank; and both memory selectors fall back to a free-text provider/model input when no static catalog exists. No provider is special-cased by name, so adding one to the chat registry now makes it embedding- and rerank-capable here automatically.
Verified on the current release tip: merged clean, typecheck:core clean, check:cycles OK across 417 files, and 35/35 across the PR's five new suites (qdrant-quick-select-catalog, memory-provider-listings, rerank-provider-listings, embedding-generic-provider-fallback, rerank-generic-provider-fallback) plus the updated hard-session-lease-bypass-inventory and embeddings-handler.
Note: the base-red disclaimer in the description referenced #9985 against release/v3.8.50 — that window is closed and the current tip carries no open base-red, so nothing was inherited here.
Thanks @rqzbeh — deriving the capability instead of enumerating providers is the version of this that stays correct as the registry grows.
* fix(memory): measure the embedding width instead of waiting for a probe
resolveEmbeddingSource() reports dimensions: null for any source the
hard-coded registry does not describe, and a self-hosted endpoint is by
definition absent from it. Both write paths then deadlocked on that null:
- scheduleVectorUpsert called ensureReady() with the null resolution, which
declines to create vec_memories, and then ignored the {ready:false} answer
and upserted anyway -- straight into the catch, so every memory was stored,
marked needs_reindex, and never vectorized;
- reindexPending refused to embed until the width was known, and the width
could only ever come from an embedding.
Nothing surfaced it: POST /api/memory returned 200 and the health check
stayed green while rowCount stayed at 0.
The comment on EmbeddingResolution.dimensions already calls this a lazy
probe; nobody performed the probe. The upsert path holds a finished vector
when it calls ensureReady, so measure it there, and let reindex spend one
embedding up front to measure -- reusing that vector rather than paying for
it twice. withMeasuredDimensions rebuilds the signature the same way the
resolution did, identity first, so two endpoints serving the same model id
still reindex independently.
scheduleVectorUpsert now also honours a {ready:false} answer instead of
upserting into a table that is not there.
Fixes#12154
* chore(changelog): point the fragment at the real PR number
* fix(memory): extract reindex helpers so the complexity ratchet stays green
runReindexBatch grew past max-lines-per-function and cognitive-complexity
when the lazy-probe path landed. Split measure/ready/item helpers without
changing the #12154 behavior.
Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
---------
Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
Co-authored-by: Diego Rodrigues de Sa e Souza <diegosouza.pw@gmail.com>
* fix(sse): map normalized xhigh to max for GLM-5.x+, DeepSeek-V4+, and provider aliases
* feat(sse): support native max reasoning effort and per-model clamping
* test(sse): add unit tests for Qwen 3.8, Claude 4.7+, GPT-5.6, and 2026 reasoning models
* fix(sse): align tests and file-size split for native max effort
Keep `max` as a first-class canonical tier. Split the new sanitizer
coverage out of base-executor-sanitize-effort.test.ts so the file stays
under testCap, and update discovery/catalog/vscode assertions to expect
native max instead of the old xhigh alias.
Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
* fix(sse): keep combo effort lists and drop unused collectSSE helper
Combo vscode routes still advertise the 5-tier list. Canonical `max` is
preserved in discovery (#9160) and github model metadata. Remove the
unused collectSSE helper that failed the absolute ESLint gate.
Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
---------
Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
Co-authored-by: Chewji <Chewji9875@users.noreply.github.com>
* feat(compression): make proactive context-compression threshold a live setting
The proactive compression trigger ratio was a hardcoded COMPRESSION_THRESHOLD =
0.7 in chatCore. Operators could not move compression relative to a client's own
compaction point (e.g. Codex Desktop self-compacts at ~0.85 of its window, so
the 0.7 proxy threshold always preempts the client's compaction with the
proxy's lossier one — see #8932 for what that produced before 3.8.50).
New: key_value namespace 'compression', key 'proactiveConfig',
{"thresholdRatio": 0.7}. Clamped [0.1, 0.99], 30s TTL cache, ipFilter
persistence pattern (#6131), synchronous read stays in the hot path. Default
unchanged; missing/invalid rows fall back to 0.7.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
* test(compression): cover the live proactive-compression threshold (read, validity bounds, fallback, TTL)
Locks in getProactiveCompressionRatio() (src/lib/db/compression.ts), the
key_value-backed replacement for chatCore's hardcoded 0.7:
- shipped default 0.7 when no compression/proactiveConfig row exists
- 30s TTL cache: a fresh DB write stays invisible until the TTL lapses
(clock mocked via node:test mock timers, Date API — the module keeps
its cache private with no reset hook)
- valid override read from key_value, boundary values 0.1/0.99 included
- out-of-range ratios fall back to the DEFAULT (a validity window, not
clamping to the nearest bound — matching the shipped comment)
- broken JSON / non-numeric thresholdRatio: 0.7, without throwing
Guard verified by mutation: switching the window to clamping fails the
out-of-range case.
---------
Co-authored-by: root-cli (Hermes ops) <info@livewellwith.us>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
Co-authored-by: Diego Rodrigues de Sa e Souza <diegosouza.pw@gmail.com>
* feat(usage): devin-cli agentic quota + openrouter credits in Provider Limits
Two provider families with live quota APIs were missing from the Provider
Limits dashboard because their list entries were absent:
- devin-cli: new usage leaf querying the Codeium seat-management Connect API
(exa.seat_management_pb.SeatManagementService/GetUserStatus, protobuf over
POST with the raw `Basic <token>-<token>` auth header the CLI itself uses).
Surfaces the plan name plus daily/weekly agentic quota percentages with
reset timestamps from the GetUserStatus plan_status payload, via a minimal
hand-rolled protobuf encoder/reader (no proto dependency warranted for two
fixed messages).
- openrouter: the /key + /credits quota fetcher (#6842) was already wired
into the dispatcher but gated out of the bulk sync — add it to
USAGE_SUPPORTED_PROVIDERS and PROVIDER_LIMITS_APIKEY_PROVIDERS so key
limits and account credits actually surface.
* fix(build): externalize tiktoken so tiktoken_bg.wasm resolves at runtime
The vendored ChatGPT Web connector v4.0.7 (#12181) imports tiktoken
(get_encoding) at module level. tiktoken's node build reads
tiktoken_bg.wasm via a __dirname-relative fs.readFileSync during import;
when Next bundles the package the wasm asset is not traced into the server
chunk, and page-data collection for every route reaching the tokenizer
(e.g. /api/providers/[id]/chatgpt-web-codex-doctor) aborts with
"Missing tiktoken_bg.wasm" — breaking the whole standalone build.
Externalize it like the other runtime-resolved native/wasm packages
(sql.js, sqlite-vec, better-sqlite3): the require stays at runtime, where
node_modules/tiktoken/tiktoken_bg.wasm resolves normally.
* fix(openrouter): /credits balance survives a /key failure
OpenRouter is credit-based, not subscription-based: the authoritative
remaining-credits signal is GET /api/v1/credits (total_credits -
total_usage, the documented "get remaining credits" endpoint), while the
/key limit fields are optional per-key caps that most accounts never set.
fetchOpenrouterQuota previously treated /key as mandatory — any /key
failure (429 rate limit, transient error, unexpected shape) discarded the
whole payload and the Usage dashboard showed "OpenRouter (usage endpoint
unreachable)" even though /credits was reachable. Now:
- /key unavailable + /credits OK → credits-only quota (creditBalance =
total_credits - total_usage) instead of null
- /key 401/403 alone no longer means an invalid token; only a double
auth-rejection (both endpoints) does
- null is returned only when both endpoints fail, and the dashboard label
reflects that ("credits endpoint unreachable")
* fix(openrouter): render AI Credits as a USD credit count in Provider Limits
The Provider Limits card's dollar renderer only activates on
isCredits/creditCount rows (QuotaCardExpanded), but openrouter went through
parseGeneric — which drops `currency` and never sets those flags — so the
credits balance rendered as a meaningless "100% left" (the unlimited-credits
row is always 100%) instead of the actual credit count.
Route openrouter's `credits` quota through buildCreditsQuota() like the
DeepSeek/AgentRouter credits rows: label "AI Credits", dollar-formatted
balance. Free-tier request windows keep the generic percentage treatment.
* fix(usage): document DEVIN_SEAT_API_URL and split quota parsers
Keep fetchOpenrouterQuota and decodeProtoFields under the complexity
ratchets, and add the seat-management URL to the env/docs contract.
Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
* test(usage): drop duplicated GLM quota-ordering test in provider-limits-ui
* test(usage): drop stale openrouter ACCEPTED_DIVERGENCE
OpenRouter is now in both USAGE_FETCHER_PROVIDERS and
USAGE_SUPPORTED_PROVIDERS, so the recorded aggregator divergence
is no longer real. Add the changelog fragment.
Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
---------
Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
Co-authored-by: Diego Rodrigues de Sa e Souza <diegosouza.pw@gmail.com>
The weight table said stability accounts for "low latency stdDev / error rate". Grep errorRate in scoring.ts and you find it declared on ProviderCandidate and read nowhere — while combo.ts pulls 24 hours of usage history behind a ten-sample floor, falls back to real-time metrics, and hands every candidate an errorRate the scorer ignores. Two candidates, one failing 1% of calls and one failing 99%, scored identically at 0.459486.
This declares reliability as a sixteenth factor: 1 - failureRate, using the same formula, field precedence and rate-bounding speedRanking.ts already applies, so a corrupt reading means "nothing observed" rather than "fails every call". It ships at weight 0, leaving the ranking unchanged to the digit — the honest default, since which weight this deserves is a product call backed by traffic the author does not have. Two declared-but-silent factors already ship (cacheAffinity, resetWindowAffinity), so the pattern is not new. The stability row now describes what that factor actually computes: latency variance.
The rest is the mechanical 15 → 16 across nineteen documents and the forty-two llm.txt mirrors — sourced from check:docs-counts rather than a grep, the first real use of the gate #12316 extended.
Protected-surface note: this PR touches AGENTS.md, llm.txt and its 42 mirrors, and skills/omni-combos-routing/SKILL.md. Every changed line in those 45 files is a digit substitution and nothing else — masking all digits makes the removed and added lines identical, with no sentence added, removed or reworded. Reviewed and approved on that basis before merging.
Verified on the author's rebased head: check:docs-counts green (the gate that now enforces the count this PR moves), typecheck:core clean, and 71/71 focused tests across scoring-reliability-factor, combo-scoring-weights-schema-coverage, check-docs-counts-sync, lkgp-enabled-context, intelligent-routing-options and the combo-matrix auto integration suite.
Thanks @maxmad64bis — shipping the factor at weight 0 and saying plainly that the weight is someone else's call is the right way to land this.
* fix(guardrails): pass providerId to getResolvedModelCapabilities in checkComboVision (#12112)
Signed-off-by: Minxi Hou <houminxi@gmail.com>
* chore(quality): register combo-vision providerId test in the stryker tap set
---------
Signed-off-by: Minxi Hou <houminxi@gmail.com>
Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
When fetching version metadata from npm registry or GitHub APIs, if the remote connection stalls during stream reading, the 10-second AbortController timer aborts the request signal but the underlying body reader stream was not listening to the abort signal. This caused readBoundedJson reader.read() loop to hang until external socket close.
Now readBoundedJson listens to AbortSignal abort events, triggers reader.cancel(), and releases locks immediately on abort.
Free Provider Rankings sorted by the top model's Arena score, so a provider stayed first even if it failed every call — the reliability column from #11546 already showed what each one actually served, but ordering ignored it.
Adds an opt-in ?sortBy=reliability (API) and a "Most reliable first" toggle (page) sharing one comparator in freeProviderRankingsUsage.ts: measured providers first by successRate desc with ELO on ties, then unmeasured in their incoming order. The default is unchanged and locked by tests. successRate is null below MIN_USAGE_REQUESTS = 5 (existing, never zero), and ordering runs before slice(0, limit) so limit counts in the requested order. The toggle composes with the existing sortTypeFirst/groupByType grouping — a stable sort keeps reliability order within each group.
Opt-in is the right default here: the page is for discovery, including providers never called.
13 tests across three files (6 new for the comparator in isolation, plus filter and route coverage including unknown sortBy → 400 and the default path staying off call_logs).
Verified in a combined batch worktree with all 11 PRs of this batch: 174/174 focused tests, typecheck:core clean, complexity 2706/3218, cognitive-complexity 1221/1437, check:cycles and check:docs-counts green. The 42-locale i18n pass was checked with the CI gates: check-ui-keys-coverage PASS (all 42 locales at or above 65%) and check-ui-value-drift PASS against the release tip.
Thanks @maxmad64bis.
docs/routing/AUTO-COMBO.md documented four mode packs; six ship. It printed 0.14 where modePacks.ts says 0.1333. And nothing was watching: four documents stated a scoring-factor count and check:docs-counts covered none of them. Wiring them up turned the gate red on seven real drifts — ARCHITECTURE.md and REPOSITORY_MAP.md at "9-factor" (code: 15) and "4 mode packs" (code: 6), RESILIENCE_GUIDE.md and SKILL.md at 13, AUTO-COMBO-GUIDE.md at both 5 and 13. ARCHITECTURE.md did not merely have the wrong number: it named nine factors that are not the engine's, and its four "mode packs" were the auto/* request prefixes.
A product fact fell out of writing the table: no pack sets quality, and applying a pack replaces the weight map wholesale (weights = pack in engine.ts, not a merge), so quality carries 0.03 by default and normalizes to 0 under any pack — pick a mode pack and the observed-quality signal stops voting. Documented, not changed.
The gate reads pack names from the module through the tsx subprocess that already reads every other code-derived count, matching the three spellings the docs actually use; on the reference document a missing claim now fails rather than passing. The dashboard was behind too (four of six packs offered); the count is dropped from the strategy label rather than corrected, since nothing reads selector labels and a right-today number goes stale unnoticed.
Verified in a combined batch worktree: 174/174 focused tests across all 11 PRs of this batch, typecheck:core clean, and check:docs-counts green with the four newly-wired documents.
Thanks @maxmad64bis — finding the quality-under-a-pack behaviour while writing a docs table is the kind of thing a table is for.
Two of the fifteen factors calculateScore applies could not be set by anyone. scoringWeightsSchema is a plain z.object, so zod strips what it does not name: PUT a combo with connectionDensity and you get a 200 back with nothing saved, and normalizeScoringWeights then reads the gap as a deliberate zero — switching off anti-concentration and the quality signal. DEFAULT_INTELLIGENT_WEIGHTS, the dashboard's own copy, missed the same two and every non-zero value differed from the engine's; summing to 1.05, validateWeights rejected them outright.
This adds the two keys to both lists and takes the dashboard defaults from DEFAULT_WEIGHTS. The scorer is not touched.
One behaviour change, and it is the point: a combo whose stored weights omitted the two keys was running with them at zero and the other thirteen renormalized upward. It now uses the engine's distribution (quota 0.1549 → 0.1429, health 0.1740 → 0.1605) and a test pins those numbers.
Left alone and documented rather than widened: rounded percentages now total 101% (six factors at 4.76% each render as 5%), and five stale .default() values in the schema that only bite when a config omits the key.
Verified in a combined batch worktree: 174/174 focused tests across all 11 PRs of this batch, typecheck:core clean, complexity 2706/3218, cognitive-complexity 1221/1437, check:cycles and check:docs-counts green.
Thanks @maxmad64bis — the red-before-green note (7 of 8 failing, and naming the one that passes on purpose) is exactly the evidence that makes a behaviour change reviewable.
GET /api/free-tier/summary could answer from a Radar overlay built 2026-08-02 while the release ships a catalog curated 2026-08-30 (FREE_CATALOG_CURATED_AT) — totals computed from older data, still tagged catalogSource: radar-overlay. The route now refuses any overlay built before the shipped catalog and falls back to that catalog through the operator's local state.
Tightens #11550 using the generatedAt persisted by #11435.
Verified in a combined batch worktree: 174/174 focused tests across all 11 PRs of this batch (this PR's free-tier-summary-radar-overlay suite included), typecheck:core clean, check-file-size, check-changelog-integrity, check:cycles and check:docs-counts green.
Thanks @maxmad64bis.
* fix(usage): console-aware Token Plan guidance and subscription hint on bailian 401
The personal Token Plan is sold through two consoles with different portals,
gateway hosts and login tickets. Two operator-facing messages ignored the split:
- The quota guidance always said 'get the cookie at home.qwencloud.com', even
for connections served by the Alibaba Model Studio console — following it
verbatim produces a cookie the gateway rejects (console mismatch →
BailianGateway.Login.NotLogined). The guidance now derives the console from
the provider via resolveConsoleSite, matching what the fetcher will do with
the pasted cookie.
- Key validation mapped upstream 401 to a bare 'Invalid API key'. An expired
Token Plan subscription produces the exact same upstream 401 (observed live
2026-09-01: subscription ended 08-23, the working key started failing), so
the message now names the subscription as a cause worth checking.
* test(providers): align the remaining bailian 401/403 message pins to prefix match
search-provider-validation.test.ts pinned the exact 'Invalid API key' string for
the bailian validator; the message now also names an expired Token Plan
subscription. Same property asserted (401/403 => invalid), prefix match.
The Token Plan console cookie is a browser credential for the operator's
cloud-console account — same class as the ollama/opencode cookies that
sanitizeProviderSpecificDataForResponse already strips — but the four
qwen/alibaba fields (qwenCloudCookie, qwenCloudSecToken, alibabaConsoleCookie,
alibabaConsoleSecToken) were missing from the strip list, so GET /api/providers
returned the operator's console session in the clear to any dashboard session.
The edit modal depended on that leak: it initialized the cookie fields from the
round-tripped response. It now starts them empty, matching the ollama pattern —
the quota-scraping assign skips empty fields and the PUT handler's partial merge
preserves keys the payload does not carry, so 'leave blank to keep the stored
cookie' (already what the field hints promise) holds for real.
Found in the 2026-09-01 audit of the Token Plan quota feature.
The 1proxy marketplace integration was decommissioned in v3.8.4; the code
survived only through the localDb barrel, deleted in #12055. Everything below
had zero consumers (grep-proven across src/, open-sse/, bin/, electron/,
scripts/ and tests/):
- src/lib/oneproxySync.ts and src/lib/oneproxyRotator.ts deleted.
- src/lib/db/oneproxy.ts: upsertOneproxyProxy, getOneproxyProxyById,
getOneproxyProxyForRotation and markOneproxyProxyFailed removed (their only
consumers were the two deleted modules); listOneproxyProxies and the record
interface stay — open-sse/utils/proxyFallback.ts still uses them.
- src/shared/validation/oneproxySchemas.ts and the unmounted
settings/components/OneproxyTab.tsx deleted (no importer anywhere; the live
UI is the FreePool* tabs over /api/settings/free-proxies).
- ONEPROXY_ENABLED feature flag removed (readerless since oneproxySync died —
the toggle no longer controlled anything); flag-count contract test aligned
54 → 53.
- Docs: PROXY_GUIDE (component rows, env rows, the three omniroute/
oneproxyRotator snippet sections), CODEBASE_DOCUMENTATION, REPOSITORY_MAP,
ENVIRONMENT (ONEPROXY_* rows), FEATURE_FLAGS, .env.example — canonical +
pl/zh-CN/zh-TW mirrors.
- The 308 compat redirects under /api/settings/oneproxy/ stay (deliberate API
compat), as do the live free-proxy provider and proxy_registry rows.
check:dead-code drops 424 → 417 (baseline kept at the velocity-phase 500 —
banking shrinks is paused until v4.0, headroom grows to 16.6%).
check:docs-all, check:env-doc-sync, typecheck:core and the 8 free-proxy/
proxy-fallback test files are green.
Closes#12091
* fix(oauth): keep Claude personal and Team organizations apart
One Anthropic identity reaches its personal workspace and every Team
organization it belongs to with the same email AND the same accountUUID,
each with its own tokens, plan and rate limits. The OAuth dedup matched on
email alone for every provider except Codex, so authenticating the second
organization overwrote the first connection instead of adding one: only the
most recent organization stayed usable. organizationUUID is the field that
separates them (cliUserID cannot be used, it changes on every login).
Disambiguate on organizationUUID, mirroring how Codex uses
workspaceId/chatgptUserId (#7737):
- findExistingOAuthConnectionMatch routes claude through a new
isSameClaudeAccount helper, so a login only merges into an existing row
when the organization agrees;
- isMatchingOauthIdentity gains organizationUUID as a third optional
disambiguator, compared strictly two-sided;
- createProviderConnection passes the incoming organizationUUID, closing the
same hole on the create path.
Rows stored before Claude returned organizationUUID keep the bare-email
match, so re-authenticating an existing connection still updates it in place
instead of forking a duplicate. No behaviour change for other providers.
* docs(oauth): changelog fragment for #12222
* fix(oauth): mark empty Antigravity projectId as degraded (#11284)
The #11284 gate only fired when projectDiscoveryOutcome was set. Paste
credentials, persistOAuthConnection, and agy CLI import could persist
projectId="" as testStatus=active, so the dashboard showed Connected
while fetchAvailableModels returned 403.
Degrade on empty projectId itself. Keep the refresh token stored so
request-time bootstrap can still self-heal.
Signed-off-by: Minxi Hou <houminxi@gmail.com>
* fix(oauth): clear stale degrade fields and bind CLI imports to builtin client
persistOAuthConnection left errorCode/lastError on the row when a later
connect discovered a Cloud Code projectId. agy CLI import also kept a
leftover custom: oauthClient marker from dashboard OAuth, so the next
refresh hit the operator web client instead of the public desktop client.
Signed-off-by: Minxi Hou <houminxi@gmail.com>
* fix(oauth): null error fields on healthy create paths too
Update already cleared errorCode/lastError* when a projectId appeared.
Create payloads still omitted the keys; match the update shape so a
fresh row cannot keep a leftover degrade marker.
Signed-off-by: Minxi Hou <houminxi@gmail.com>
* test(oauth): pin healthy create/upsert nulling of degrade fields
Forge flagged create payloads omitting errorCode/lastError* when a
projectId is present. Production already writes explicit nulls; the
reader strips them via cleanNulls, so pin both the payload shape and
the upsert path that must overwrite a leftover degrade marker.
Signed-off-by: Minxi Hou <houminxi@gmail.com>
* fix(oauth): type create payload from AntigravityDegradedProjectState
The persistence helper duplicated a subset of the degrade type and
dropped warning. Align the parameter so the HTTP-only warning field
cannot drift from the exported type.
Signed-off-by: Minxi Hou <houminxi@gmail.com>
* fix(oauth): persist degrade status through a single override helper
OAuth exchange/poll-callback spread the whole degrade object, which
wrote warning into the SQLite row and left healthy updates as {}.
Centralize testStatus/errorCode/lastError* so they always win over a
spread tokenData payload.
Signed-off-by: Minxi Hou <houminxi@gmail.com>
---------
Signed-off-by: Minxi Hou <houminxi@gmail.com>
Refresh the existing MIT-licensed miuuyy/codex-chatgpt-web vendor snapshot and its OmniRoute integration as one reviewable change.
Co-authored-by: backryun <backryun@daonlab.local>
* fix(memory): honest probe-driven FTS5 keyword status + memory_id rowid sync
The "no such module: fts5" complaint on FTS5-less runtime builds (sql.js/WASM
under a global install) was masked by a hardcoded keyword.available=true in
engineStatus and an unsanitized FTS5 MATCH path. Address root cause:
- engineStatus(): probe runtime via supportsFts5(db) instead of hardcoding
available=true; keywordEngineStatus() reports the true backend (FTS5 vs
none) with a reason. Schema, OpenAPI, dashboard chip updated to match.
- store.ts: sync memory_id to the SQLite rowid on insert (+ self-heal legacy
NULL rows). Migration 023 keys the FTS5 external-content trigger off
memory_id, but plain INSERT left it NULL so the JOIN returned 0 rows —
keyword/hybrid search silently returned nothing on FTS5-capable builds.
- retrieval.ts: apply sanitizeFts5Query() to the preview MATCH path.
Tests updated/added across memory-engine-status, memory-retrieve-preview,
memory-schemas-roundtrip, memory-store, and the integration engine-status
test (dropping the hardcoded "always available" assertion). 66 unit tests
pass; lint and typecheck clean.
* fix(memory): sanitize FTS5 queries for memory retrieval
Prevent SQLite FTS5 syntax errors by sanitizing query terms and replacing FTS control operators with double-quoted tokens.
When a combo is duplicated or imported, its inner data JSON blob may
retain a stale id from the template. withRowId previously kept the inner
string id instead of prioritizing the database primary key (row.id),
causing GET /api/combos to return mismatched ids and breaking subsequent
DELETE / PUT operations with 404.
Also add an error notification branch to handleDelete in the combos page
so failed delete requests surface actionable feedback instead of failing
silently.
Signed-off-by: Minxi Hou <houminxi@gmail.com>
Every Codex quota read goes through throttleQuotaFetch() — the #6009/#6058
gate that spaces genuine upstream calls so many accounts behind one IP do not
fire in the same second, which is the pattern documented to have got a Codex
OAuth token revoked. The auto-ping scheduler called getCodexUsage() directly,
so the one Codex path that runs unattended every 60s per connection was the
one skipping the mitigation written for Codex.
The tick walks connections sequentially but without spacing, so N enabled
connections still produce N upstream usage requests within a few hundred ms.
Gate the read on the same throttle, injected through deps like every other
effect in this module. Placed after the skip checks so a connection filtered
out by the circuit breaker, a cooldown or the failure cache does not consume
a slot and delay the connections that do reach the network.
This does not change the polling cadence. Codex sets pingWhenResetAtSlides
because its resetAt slides forward while the window is idle, so the per-tick
re-fetch is deliberate and is left alone.
Closes#11904
