Repository navigation
Release v3.7.5 - #1753
Release v3.7.5#1753
Conversation
CI Coverage Report
Coverage artifact was not available for this run. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 5fd56207cf
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| return NextResponse.json({ | ||
| success: true, | ||
| action: parsed.action, | ||
| status, |
There was a problem hiding this comment.
POST /api/tunnels/ngrok always responds with success: true once startNgrokTunnel() resolves, but startNgrokTunnel() resolves even on errors (it returns a status with phase: "error"/needs_auth instead of throwing). In that case this handler still sends HTTP 200 success, so the dashboard treats the action as successful (shows a success toast and clears token) even though no tunnel was started; this breaks retry flows for invalid/missing auth tokens and other startup failures.
Useful? React with 👍 / 👎.
| supported: true, | ||
| installed: true, |
There was a problem hiding this comment.
getNgrokTunnelStatus() hardcodes supported: true and installed: true, so the unsupported/not_installed phases in this module are effectively unreachable. On environments where @ngrok/ngrok cannot load (unsupported OS/arch or missing native package), the API still reports ngrok as installed/supported until start fails, which causes incorrect UI state and hides the intended preflight gating.
Useful? React with 👍 / 👎.
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request integrates native ngrok tunnel support into the dashboard, allowing users to manage public tunnels directly. It also includes a bug fix to correctly apply blocked provider filters to non-chat models. I have identified several areas for improvement in the new ngrok implementation, specifically regarding type safety and potential race conditions in the tunnel management logic.
| export async function stopNgrokTunnel(): Promise<NgrokTunnelStatus> { | ||
| const listener = globalForNgrok.__ngrokListener; |
There was a problem hiding this comment.
There is a potential race condition in stopNgrokTunnel. If it is called while startNgrokTunnel is still in progress (e.g., awaiting the dynamic import or the ngrok.forward call), the listener won't have been assigned to the global storage yet. Awaiting the startPromise ensures that any pending start operation completes before attempting to close the listener, preventing the tunnel from being left running in an orphaned state.
export async function stopNgrokTunnel(): Promise<NgrokTunnelStatus> {
if (startPromise) await startPromise;
const listener = globalForNgrok.__ngrokListener;| type NgrokTunnelPhase = | ||
| | "unsupported" | ||
| | "not_installed" | ||
| | "stopped" | ||
| | "needs_auth" | ||
| | "starting" | ||
| | "running" | ||
| | "error"; | ||
|
|
||
| type NgrokTunnelStatus = { | ||
| supported: boolean; | ||
| installed: boolean; | ||
| running: boolean; | ||
| publicUrl: string | null; | ||
| apiUrl: string | null; | ||
| targetUrl: string; | ||
| phase: NgrokTunnelPhase; | ||
| lastError: string | null; | ||
| }; |
There was a problem hiding this comment.
The NgrokTunnelPhase and NgrokTunnelStatus types are identical to those defined in src/lib/ngrokTunnel.ts. Duplicating these types increases the risk of them falling out of sync as the feature evolves. It is recommended to import these types from the library file instead of redefining them here to maintain a single source of truth.
|
|
||
| // Next.js hot-reloading safe global storage for the listener | ||
| const globalForNgrok = globalThis as unknown as { | ||
| __ngrokListener: any; |
There was a problem hiding this comment.
Using any for the global listener storage bypasses TypeScript's type safety. Since the @ngrok/ngrok package provides types, you should use import type { Listener } from "@ngrok/ngrok" at the top of the file and type this as Listener | undefined. This provides better IDE support and safety without triggering a full import of the package at the top level.
| const ngrok = await import("@ngrok/ngrok"); | ||
|
|
||
| const targetUrl = getLocalTargetUrl(); | ||
| const listenerOptions: any = { addr: targetUrl }; |
There was a problem hiding this comment.
The listenerOptions object is typed as any, which is unnecessary. You can use the Config type from @ngrok/ngrok (via import type) to ensure that the options passed to ngrok.forward are valid and benefit from autocomplete.
* fix: resolve build issues for v3.7.5 and apply memory/translation fixes 1. antigravityHeaders.ts: restore ANTIGRAVITY_LOAD_CODE_ASSIST_* exports for oauth.ts compatibility 2. next.config.mjs: add @ngrok/ngrok to serverExternalPackages and webpack externals to handle native .node modules 3. Memory system: UPSERT logic to prevent duplicate entries with same apiKeyId + key 4. Chinese translations: complete CLI tools and memory dashboard localizations 5. Test fixes: unique keys for pagination tests to comply with unique constraint Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: address Gemini Code Assist review feedback 1. store.ts: add expires_at to UPDATE statement in UPSERT logic - Previously, expires_at was not being persisted to database on update - This caused state mismatch between returned Memory object and actual DB row 2. package-lock.json: revert react-markdown registry to official npmjs.org - Mirror-specific registry URL (npmmirror.com) should not be in lockfile Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* fix(antigravity): normalize Gemini bridge payloads Clamp Claude bridge output tokens, use Gemini-valid system roles and tool names, and serialize antigravity requests from a cloned body so Cloud Code payload shaping stays valid. * fix(cli): stop fallback after unsafe known paths Preserve known-path security checks by stopping command discovery when a configured CLI path is suspicious or non-executable, instead of falling through to PATH discovery. * test(memory): make query result assertion deterministic Avoid relying on database result ordering when checking filtered memory keys so the unit suite remains stable across runs. * fix(review): preserve safe cloning and CLI reasons Handle non-cloneable antigravity request bodies without throwing and preserve specific CLI known-path failure reasons instead of masking them as not_found.
… awaits (#1771) When a combo target takes too long, the request-level deadline fires and calls abortController.abort() on the stream controller, but the abort signal never reaches pending awaits in acquireAccountSemaphore() or withRateLimit(). These awaits sit between stream controller creation and executor.execute(), causing requests to hang indefinitely past the 600s deadline. Pass streamController.signal to both functions so they can respond to abort events and terminate early when the request deadline expires. Signed-off-by: wucm667 <stevenwucongmin@gmail.com>
* Fix model sync import handling * Align model import storage semantics * Address model review feedback
* Move DB health to management API * Address DB health review feedback
… refresh (#1754) * fix(kiro): support organization IDC OAuth with regional endpoints and refresh * fix(kiro): refresh IDC tokens with stored region --------- Co-authored-by: ngocdb <ngocdb@ngocdb.local>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: a07d9b62a8
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| apiUrl: currentUrl ? getTunnelApiUrl(currentUrl) : null, | ||
| targetUrl, | ||
| phase: currentUrl === null ? (tokenAvailable ? "stopped" : "needs_auth") : "running", | ||
| lastError: null, |
There was a problem hiding this comment.
getNgrokTunnelStatus() always clears lastError, so any failure returned by startNgrokTunnel() disappears on the next status poll. In the dashboard flow, status is re-fetched immediately after actions, which means invalid token/import/runtime errors are lost and operators cannot see why startup failed. Keep the last startup error in shared tunnel state (until a successful start/explicit stop) so subsequent GET /api/tunnels/ngrok calls remain diagnosable.
Useful? React with 👍 / 👎.
| rememberedResponseToolCalls.set(normalizedResponseId, { | ||
| functionCalls, |
There was a problem hiding this comment.
The tool-call replay cache is keyed only by response_id, and Codex rehydrates missing function_call items from that global map using only previous_response_id. If two active sessions/providers emit the same response ID (common with OpenAI-compatible backends using short or deterministic IDs), follow-up tool outputs can pull the wrong function signature/arguments and corrupt the turn. Include connection or credential scope in the cache key to prevent cross-session collisions.
Useful? React with 👍 / 👎.
- Add ABSOLUTE PROHIBITION section to review-prs.md - Add PR PROHIBITION rule to resolve-issues.md - Add contributor credit rule to AGENTS.md Review Focus - Based on audit finding: 37 PRs had code absorbed without merge credit
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 1a0da9c13f
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| now, | ||
| memory.sessionId, | ||
| memory.type, | ||
| memory.expiresAt ?? null, |
There was a problem hiding this comment.
In the new UPSERT path, createMemory() writes memory.expiresAt directly into SQLite, but API callers pass expiresAt as a Date (z.coerce.date() in /api/memory). Unlike the insert/update code paths that call .toISOString(), this branch can pass a non-serializable value to better-sqlite3, causing writes to fail when a duplicate key is updated with a non-null expiry.
Useful? React with 👍 / 👎.
| return await Promise.race([limiter.schedule(scheduleOpts, fn), abortPromise]); | ||
| } finally { |
There was a problem hiding this comment.
The abort handling now races limiter.schedule(...) against an abort promise, but on abort it only rejects the caller and never cancels the scheduled Bottleneck job. That leaves stale queued work running later (with an already-aborted request), which still consumes limiter capacity and can throttle unrelated live requests during client disconnect churn.
Useful? React with 👍 / 👎.
…e credit This commit formally recognizes 29 contributors whose code was manually integrated across releases v3.4.0 through v3.7.4 without proper GitHub merge credit. Their PRs were resolved locally due to merge conflicts but closed instead of merged, preventing them from appearing in the Contributors graph. We have updated our workflows to ensure this never happens again. Co-authored-by: Randi <55005611+rdself@users.noreply.github.com> Co-authored-by: Benson K B <4044180+benzntech@users.noreply.github.com> Co-authored-by: clousky2020 <33016567+clousky2020@users.noreply.github.com> Co-authored-by: Raxxoor <7317522+dhaern@users.noreply.github.com> Co-authored-by: Jason Landbridge <15127381+JasonLandbridge@users.noreply.github.com> Co-authored-by: slewis3600 <35925982+slewis3600@users.noreply.github.com> Co-authored-by: Markus Hartung <12826053+hartmark@users.noreply.github.com> Co-authored-by: Hernan Javier Ardila Sanchez <204746071+herjarsa@users.noreply.github.com> Co-authored-by: 3_1_3_u <5846351+andruwa13@users.noreply.github.com> Co-authored-by: Paijo <14921983+oyi77@users.noreply.github.com> Co-authored-by: i1hwan <35260883+i1hwan@users.noreply.github.com> Co-authored-by: xandr0s <1709302+xandr0s@users.noreply.github.com> Co-authored-by: backryun <24198422+backryun@users.noreply.github.com> Co-authored-by: Owen <36758131+kang-heewon@users.noreply.github.com> Co-authored-by: Ravi Tharuma <25951435+RaviTharuma@users.noreply.github.com> Co-authored-by: Chris <3751981+christopher-s@users.noreply.github.com> Co-authored-by: Wellington Fonseca <5421548+wlfonseca@users.noreply.github.com> Co-authored-by: Ethan Hunt <136065060+only4copilot@users.noreply.github.com> Co-authored-by: tombii <6607822+tombii@users.noreply.github.com> Co-authored-by: AndrewDragonIV <7906124+AndrewDragonIV@users.noreply.github.com> Co-authored-by: Danh Thanh <50534210+dt418@users.noreply.github.com> Co-authored-by: Will F <30637450+willbnu@users.noreply.github.com> Co-authored-by: defhouse <232128212+defhouse@users.noreply.github.com> Co-authored-by: Skydwest <186351198+mercs2910@users.noreply.github.com> Co-authored-by: zenobit <6384793+zen0bit@users.noreply.github.com> Co-authored-by: Ivan <16905671+razllivan@users.noreply.github.com> Co-authored-by: foxy1402 <45601526+foxy1402@users.noreply.github.com> Co-authored-by: Luan Dias <65574834+luandiasrj@users.noreply.github.com> Co-authored-by: Sergei Korolev <891832+knopki@users.noreply.github.com> Co-authored-by: dail45 <69967573+dail45@users.noreply.github.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 21e5d202a2
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| const listener = await ngrok.forward(listenerOptions); | ||
| globalForNgrok.__ngrokListener = listener; |
There was a problem hiding this comment.
If stopNgrokTunnel() is called while startNgrokTunnel() is still awaiting ngrok.forward(...), the disable path clears only the current listener, but startup later writes a fresh listener back into global state. In that race, a user/API caller can issue disable and still end up with a running tunnel moments later. This is reproducible with concurrent enable/disable requests and leaves the UI/state inconsistent with the requested action.
Useful? React with 👍 / 👎.
…retry on exhausted quotas (#1767) Previously, PROVIDER_FAILURE_ERROR_CODES only included {408, 500, 502, 503, 504}, meaning 429 responses never counted toward the circuit breaker threshold. This caused exhausted accounts to be retried every 3-5 seconds indefinitely instead of being blocked by the provider breaker. Adding 429 ensures persistent rate limiting triggers the circuit breaker after the configured failure threshold, giving the provider time to recover.
…quota drain (#1761) Previously, OmniRoute unconditionally injected thinking: {type: 'adaptive'} and output_config: {effort: 'high'} for Claude Opus 4.7 in Claude Code client requests. This caused Claude Max 5h quota to drain in ~15 minutes. Now checks the original client body: if thinking or output_config are explicitly set (even to null or a different value), the injection is skipped. Users can opt-out by sending thinking: null or output_config: {effort: 'low'}.
Integrated into release/v3.7.5
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 072b4cd414
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| const runtimeLocation = useMemo(() => { | ||
| if (typeof window === "undefined") { | ||
| return { | ||
| isLocalhost: false, | ||
| isTrueLocalhost: false, | ||
| placeholderUrl: "/callback?code=...", |
There was a problem hiding this comment.
This computes isLocalhost/isTrueLocalhost during render via useMemo, which produces different initial markup between SSR (window undefined => false) and client hydration on localhost/LAN (window defined => true). That can trigger hydration mismatch warnings and briefly render the wrong OAuth instructions/flow branch for users connecting from local addresses. The previous mount-time state update pattern avoided this by keeping server and first client render consistent.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 33d4a89a47
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| sessionCache.set(cacheKey, { | ||
| sessionData, | ||
| subscriptionCache, | ||
| teamAccount, | ||
| fetchedAt: Date.now(), | ||
| }); |
There was a problem hiding this comment.
Add eviction for sessionCache entries: the new TTL only controls cache hits, but expired entries are never removed from the map. In a long-running process handling many distinct Blackbox cookies (for example, multiple users or rotated session tokens), this causes unbounded in-memory growth even after entries are stale, which can degrade performance and increase memory pressure over time.
Useful? React with 👍 / 👎.
| /please login|login required|authentication required/i.test(responseText) && | ||
| !isSubscriptionError; |
There was a problem hiding this comment.
The auth error check is based on broad substring matching of the raw completion text, so valid assistant outputs that include phrases like “please login” or “authentication required” are reclassified as HTTP 401 errors. This will incorrectly fail legitimate prompts that discuss authentication wording, because the upstream 200 response body is treated as an auth failure solely by text content.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 0578d8a2db
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if (isImportedSource(model.source) || discoveredIds.has(modelId)) { | ||
| removedCustomModels.push(model); | ||
| continue; |
There was a problem hiding this comment.
This loop now removes every discovered/imported model from customModels and never re-adds them, which means synced models exist only in syncedAvailableModels. Non-chat routes still source discovered models from getAllCustomModels() (for example src/app/api/v1/images/generations/route.ts and src/app/api/v1/embeddings/route.ts), so after a sync those imported image/embedding models stop appearing in model lists and custom-provider fallback matching can fail. In practice, users who rely on /api/providers/[id]/sync-models for non-chat capabilities can get "invalid model" behavior even though sync succeeded.
Useful? React with 👍 / 👎.
Delete temporary patch scripts and local OMC session files that should not ship with the repository. Also remove the Next.js config file and expand editor and TypeScript exclusions to ignore large local workspace directories and reduce unnecessary indexing.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: f18e7690cc
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| source: "/v1/:path*", | ||
| destination: "/api/v1/:path*", |
There was a problem hiding this comment.
Removing next.config.mjs drops the only rewrite rules that map /v1/*, /chat/completions, /responses*, and /models to the real /api/v1/* handlers, but there are still active call sites that use those legacy paths (for example open-sse/mcp-server/server.ts calls omniRouteFetch("/v1/chat/completions") at line 367). With no replacement rewrite layer in this commit, those requests now resolve to 404s, breaking MCP/tooling and external clients that target the documented compatibility endpoints.
Useful? React with 👍 / 👎.
| experimental: { | ||
| serverActions: { | ||
| bodySizeLimit: process.env.OMNIROUTE_SERVER_ACTIONS_BODY_LIMIT || "50mb", | ||
| }, |
There was a problem hiding this comment.
Deleting this config also removes experimental.serverActions.bodySizeLimit, which previously raised the framework-level request cap to 50MB; without it, Next.js falls back to the default small limit and can reject multipart/form-data requests before route handlers execute. That regresses large upload flows such as /api/v1/audio/transcriptions (multipart endpoint) even though the app-level guard is configured for much larger payloads in src/shared/middleware/bodySizeGuard.ts.
Useful? React with 👍 / 👎.
Integrated into release/v3.7.5
Integrated into release/v3.7.5
… Plus tier (#1792) Integrated into release/v3.7.5
Integrated into release/v3.7.5
Integrated into release/v3.7.5
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: b152286e36
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| delete body.previous_response_id; | ||
| if (Array.isArray(body.input) && body.input.length === 0) { | ||
| delete body.input; |
There was a problem hiding this comment.
The new cleanup path unconditionally deletes previous_response_id and then drops input when it is empty, which breaks state-only follow-up requests (common in Codex/Cursor flows that send just a prior response reference). In that scenario the transformed payload is left with neither input nor previous_response_id, so upstream /responses calls fail with a 4xx schema/validation error instead of continuing the turn or triggering the intended replay fallback.
Useful? React with 👍 / 👎.
Add an editable key name field at the top of the permissions modal, allowing users to rename API keys alongside existing permission settings. The backend already supported name updates via PATCH /api/keys/:id — this wires the UI to send the name field and refreshes the key list on success. Changes: - Add keyName state and text input to PermissionsModal - Update handleUpdatePermissions to validate and send name in PATCH body - Add integration test for rename via PATCH (valid, empty, too-long names) - Update E2E mock to handle PATCH requests
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 630e82b459
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| })); | ||
|
|
||
| if (injectedFunctionCalls.length > 0) { | ||
| body.input = [...injectedFunctionCalls, ...inputItems]; |
There was a problem hiding this comment.
When previous_response_id is present, this function first prepends locally remembered conversation items to body.input, but this assignment rebuilds body.input from inputItems (the original pre-replay array) and drops that restored history. The regression triggers on follow-ups that include function_call_output without matching function_call: the fallback injection path runs, strips the replayed context, and sends only injected calls + current turn, which can break Codex continuation quality and cause repeated tool execution because prior conversation state is lost.
Useful? React with 👍 / 👎.
Integrated into release/v3.7.5
Integrated into release/v3.7.5
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: b00a4f7397
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if (rememberedConversationItems.length > 0) { | ||
| body.input = [...rememberedConversationItems, ...inputItems]; | ||
| } |
There was a problem hiding this comment.
When previous_response_id is present, this unconditionally prepends locally remembered conversation items to body.input, even if the caller already included full history. In clients that retry by resending prior turns (common after a 404/compat fallback), this duplicates large chunks of context, inflates token usage, and can push requests over context limits or distort tool-call sequencing. Gate the prepend behind a missing-context check (or dedupe) instead of always concatenating.
Useful? React with 👍 / 👎.
|
* docs(changelog): record PR cheaper-inference#1748 for next release * fix(models): apply blocked providers filter to non-chat catalog models (cheaper-inference#1752) * chore(release): v3.7.5 — integrate ngrok tunnel and fix models filter (cheaper-inference#1753, cheaper-inference#1752) * chore(release): update changelog format for v3.7.5 * Speed up endpoint initial render * Address endpoint review feedback * Add endpoint loading model translations * fix: resolve build issues and implement memory UPSERT logic (cheaper-inference#1763) * fix: resolve build issues for v3.7.5 and apply memory/translation fixes 1. antigravityHeaders.ts: restore ANTIGRAVITY_LOAD_CODE_ASSIST_* exports for oauth.ts compatibility 2. next.config.mjs: add @ngrok/ngrok to serverExternalPackages and webpack externals to handle native .node modules 3. Memory system: UPSERT logic to prevent duplicate entries with same apiKeyId + key 4. Chinese translations: complete CLI tools and memory dashboard localizations 5. Test fixes: unique keys for pagination tests to comply with unique constraint Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: address Gemini Code Assist review feedback 1. store.ts: add expires_at to UPDATE statement in UPSERT logic - Previously, expires_at was not being persisted to database on update - This caused state mismatch between returned Memory object and actual DB row 2. package-lock.json: revert react-markdown registry to official npmjs.org - Mirror-specific registry URL (npmmirror.com) should not be in lockfile Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * fix(antigravity): normalize Gemini bridge payloads (cheaper-inference#1769) * fix(antigravity): normalize Gemini bridge payloads Clamp Claude bridge output tokens, use Gemini-valid system roles and tool names, and serialize antigravity requests from a cloned body so Cloud Code payload shaping stays valid. * fix(cli): stop fallback after unsafe known paths Preserve known-path security checks by stopping command discovery when a configured CLI path is suspicious or non-executable, instead of falling through to PATH discovery. * test(memory): make query result assertion deterministic Avoid relying on database result ordering when checking filtered memory keys so the unit suite remains stable across runs. * fix(review): preserve safe cloning and CLI reasons Handle non-cloneable antigravity request bodies without throwing and preserve specific CLI known-path failure reasons instead of masking them as not_found. * fix(sse): propagate AbortSignal to pre-fetch semaphore and rate-limit awaits (cheaper-inference#1771) When a combo target takes too long, the request-level deadline fires and calls abortController.abort() on the stream controller, but the abort signal never reaches pending awaits in acquireAccountSemaphore() or withRateLimit(). These awaits sit between stream controller creation and executor.execute(), causing requests to hang indefinitely past the 600s deadline. Pass streamController.signal to both functions so they can respond to abort events and terminate early when the request deadline expires. Signed-off-by: wucm667 <stevenwucongmin@gmail.com> * Fix model sync import handling (cheaper-inference#1755) * Fix model sync import handling * Align model import storage semantics * Address model review feedback * fix(codex): stabilize copilot responses reasoning and tool replay (cheaper-inference#1750) * chore(xiaomi): Update Xiaomi provider model list (cheaper-inference#1759) * Move DB health to management API (cheaper-inference#1757) * Move DB health to management API * Address DB health review feedback * fix(kiro): support organization IDC OAuth with regional endpoints and refresh (cheaper-inference#1754) * fix(kiro): support organization IDC OAuth with regional endpoints and refresh * fix(kiro): refresh IDC tokens with stored region --------- Co-authored-by: ngocdb <ngocdb@ngocdb.local> * chore(workflows): add strict PR contributor credit policy - Add ABSOLUTE PROHIBITION section to review-prs.md - Add PR PROHIBITION rule to resolve-issues.md - Add contributor credit rule to AGENTS.md Review Focus - Based on audit finding: 37 PRs had code absorbed without merge credit * chore(release): acknowledge 29 community contributors with retroactive credit This commit formally recognizes 29 contributors whose code was manually integrated across releases v3.4.0 through v3.7.4 without proper GitHub merge credit. Their PRs were resolved locally due to merge conflicts but closed instead of merged, preventing them from appearing in the Contributors graph. We have updated our workflows to ensure this never happens again. Co-authored-by: Randi <55005611+rdself@users.noreply.github.com> Co-authored-by: Benson K B <4044180+benzntech@users.noreply.github.com> Co-authored-by: clousky2020 <33016567+clousky2020@users.noreply.github.com> Co-authored-by: Raxxoor <7317522+dhaern@users.noreply.github.com> Co-authored-by: Jason Landbridge <15127381+JasonLandbridge@users.noreply.github.com> Co-authored-by: slewis3600 <35925982+slewis3600@users.noreply.github.com> Co-authored-by: Markus Hartung <12826053+hartmark@users.noreply.github.com> Co-authored-by: Hernan Javier Ardila Sanchez <204746071+herjarsa@users.noreply.github.com> Co-authored-by: 3_1_3_u <5846351+andruwa13@users.noreply.github.com> Co-authored-by: Paijo <14921983+oyi77@users.noreply.github.com> Co-authored-by: i1hwan <35260883+i1hwan@users.noreply.github.com> Co-authored-by: xandr0s <1709302+xandr0s@users.noreply.github.com> Co-authored-by: backryun <24198422+backryun@users.noreply.github.com> Co-authored-by: Owen <36758131+kang-heewon@users.noreply.github.com> Co-authored-by: Ravi Tharuma <25951435+RaviTharuma@users.noreply.github.com> Co-authored-by: Chris <3751981+christopher-s@users.noreply.github.com> Co-authored-by: Wellington Fonseca <5421548+wlfonseca@users.noreply.github.com> Co-authored-by: Ethan Hunt <136065060+only4copilot@users.noreply.github.com> Co-authored-by: tombii <6607822+tombii@users.noreply.github.com> Co-authored-by: AndrewDragonIV <7906124+AndrewDragonIV@users.noreply.github.com> Co-authored-by: Danh Thanh <50534210+dt418@users.noreply.github.com> Co-authored-by: Will F <30637450+willbnu@users.noreply.github.com> Co-authored-by: defhouse <232128212+defhouse@users.noreply.github.com> Co-authored-by: Skydwest <186351198+mercs2910@users.noreply.github.com> Co-authored-by: zenobit <6384793+zen0bit@users.noreply.github.com> Co-authored-by: Ivan <16905671+razllivan@users.noreply.github.com> Co-authored-by: foxy1402 <45601526+foxy1402@users.noreply.github.com> Co-authored-by: Luan Dias <65574834+luandiasrj@users.noreply.github.com> Co-authored-by: Sergei Korolev <891832+knopki@users.noreply.github.com> Co-authored-by: dail45 <69967573+dail45@users.noreply.github.com> * fix(combo): include 429 in provider circuit breaker to stop infinite retry on exhausted quotas (cheaper-inference#1767) Previously, PROVIDER_FAILURE_ERROR_CODES only included {408, 500, 502, 503, 504}, meaning 429 responses never counted toward the circuit breaker threshold. This caused exhausted accounts to be retried every 3-5 seconds indefinitely instead of being blocked by the provider breaker. Adding 429 ensures persistent rate limiting triggers the circuit breaker after the configured failure threshold, giving the provider time to recover. * fix(claude): respect client thinking/effort params to prevent forced quota drain (cheaper-inference#1761) Previously, OmniRoute unconditionally injected thinking: {type: 'adaptive'} and output_config: {effort: 'high'} for Claude Opus 4.7 in Claude Code client requests. This caused Claude Max 5h quota to drain in ~15 minutes. Now checks the original client body: if thinking or output_config are explicitly set (even to null or a different value), the injection is skipped. Users can opt-out by sending thinking: null or output_config: {effort: 'low'}. * Add MseeP.ai badge to README.md (cheaper-inference#1727) Integrated into release/v3.7.5 * chore(docs): update CHANGELOG for PR cheaper-inference#1727 * fix(tests): update stream-utils assertion for responses api compliance * feat: Fix support for claude-cli using Gemini provider (cheaper-inference#1779) Integrated into release/v3.7.5 * fix(codex): align client identity metadata (cheaper-inference#1778) Integrated into release/v3.7.5 * fix(blackbox-web): correct cookie name and populate session/subscription fields (cheaper-inference#1776) Integrated into release/v3.7.5 * Fix Codex /responses/compact passthrough (cheaper-inference#1777) Integrated into release/v3.7.5 * test(reasoning-cache): isolate DB state using mkdtempSync to prevent 401 middleware errors * chore(release): v3.7.5 — integrate remaining PRs and finalize stability * chore(config): remove local patch artifacts and trim workspace config Delete temporary patch scripts and local OMC session files that should not ship with the repository. Also remove the Next.js config file and expand editor and TypeScript exclusions to ignore large local workspace directories and reduce unnecessary indexing. * fix(antigravity): cap Claude bridge output tokens (cheaper-inference#1785) Integrated into release/v3.7.5 * fix(codex): stabilize Copilot responses replay state (cheaper-inference#1791) Integrated into release/v3.7.5 * fix(chatgpt-web): restore validator + expand model catalog to ChatGPT Plus tier (cheaper-inference#1792) Integrated into release/v3.7.5 * fix(antigravity): scrub internal OmniRoute headers (cheaper-inference#1794) Integrated into release/v3.7.5 * fix(grok-web): fix Grok validator and cookie parsing (cheaper-inference#1793) Integrated into release/v3.7.5 * chore(release): v3.7.5 — finalize changelog for LTS patch * feat(api-keys): add rename support in permissions modal Add an editable key name field at the top of the permissions modal, allowing users to rename API keys alongside existing permission settings. The backend already supported name updates via PATCH /api/keys/:id — this wires the UI to send the name field and refreshes the key list on success. Changes: - Add keyName state and text input to PermissionsModal - Update handleUpdatePermissions to validate and send name in PATCH body - Add integration test for rename via PATCH (valid, empty, too-long names) - Update E2E mock to handle PATCH requests * chore(release): finalize v3.7.5 LTS release with schema and db initialization fixes * test: fix json escaping in stream-utilities test * fix(build): restore next.config.mjs that was accidentally deleted * fix(sse): decrement pending requests on passthrough mode failure (cheaper-inference#1798) Integrated into release/v3.7.5 * fix(grok-web): repair validator probe + accept full cookie blobs (cheaper-inference#1793) Integrated into release/v3.7.5 * docs(i18n): sync documentation updates to 40 languages --------- Signed-off-by: wucm667 <stevenwucongmin@gmail.com> Co-authored-by: diegosouzapw <diegosouzapw@users.noreply.github.com> Co-authored-by: R.D. <rogerproself@gmail.com> Co-authored-by: clousky2020 <33016567+clousky2020@users.noreply.github.com> Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> Co-authored-by: cloudy <37777261+uwuclxdy@users.noreply.github.com> Co-authored-by: wucm667 <109257021+wucm667@users.noreply.github.com> Co-authored-by: Randi <55005611+rdself@users.noreply.github.com> Co-authored-by: ivan-mezentsev <ivan@mezentsev.me> Co-authored-by: backryun <bakryun0718@proton.me> Co-authored-by: Dao Bao Ngoc <42265865+daongoc315@users.noreply.github.com> Co-authored-by: ngocdb <ngocdb@ngocdb.local> Co-authored-by: Benson K B <4044180+benzntech@users.noreply.github.com> Co-authored-by: Raxxoor <7317522+dhaern@users.noreply.github.com> Co-authored-by: Jason Landbridge <15127381+JasonLandbridge@users.noreply.github.com> Co-authored-by: slewis3600 <35925982+slewis3600@users.noreply.github.com> Co-authored-by: Markus Hartung <12826053+hartmark@users.noreply.github.com> Co-authored-by: Hernan Javier Ardila Sanchez <204746071+herjarsa@users.noreply.github.com> Co-authored-by: 3_1_3_u <5846351+andruwa13@users.noreply.github.com> Co-authored-by: Paijo <14921983+oyi77@users.noreply.github.com> Co-authored-by: i1hwan <35260883+i1hwan@users.noreply.github.com> Co-authored-by: xandr0s <1709302+xandr0s@users.noreply.github.com> Co-authored-by: backryun <24198422+backryun@users.noreply.github.com> Co-authored-by: Owen <36758131+kang-heewon@users.noreply.github.com> Co-authored-by: Ravi Tharuma <25951435+RaviTharuma@users.noreply.github.com> Co-authored-by: Chris <3751981+christopher-s@users.noreply.github.com> Co-authored-by: Wellington Fonseca <5421548+wlfonseca@users.noreply.github.com> Co-authored-by: Ethan Hunt <136065060+only4copilot@users.noreply.github.com> Co-authored-by: tombii <6607822+tombii@users.noreply.github.com> Co-authored-by: AndrewDragonIV <7906124+AndrewDragonIV@users.noreply.github.com> Co-authored-by: Danh Thanh <50534210+dt418@users.noreply.github.com> Co-authored-by: Will F <30637450+willbnu@users.noreply.github.com> Co-authored-by: defhouse <232128212+defhouse@users.noreply.github.com> Co-authored-by: Skydwest <186351198+mercs2910@users.noreply.github.com> Co-authored-by: zenobit <6384793+zen0bit@users.noreply.github.com> Co-authored-by: Ivan <16905671+razllivan@users.noreply.github.com> Co-authored-by: foxy1402 <45601526+foxy1402@users.noreply.github.com> Co-authored-by: Luan Dias <65574834+luandiasrj@users.noreply.github.com> Co-authored-by: Sergei Korolev <891832+knopki@users.noreply.github.com> Co-authored-by: dail45 <69967573+dail45@users.noreply.github.com> Co-authored-by: MseeP.ai <mseep@skydeck.ai> Co-authored-by: Markus Hartung <mail@hartmark.se> Co-authored-by: Raxxoor <manker_lol@hotmail.com> Co-authored-by: Jack <5443152+hijak@users.noreply.github.com> Co-authored-by: Sergey Morozov <tr0st@bk.ru> Co-authored-by: payne <baboialex95@gmail.com> Co-authored-by: Antigravity Assistant <bot@antigravity.local> Co-authored-by: Andrew Munsell <andrew@wizardapps.net>
* docs(changelog): record PR cheaper-inference#1748 for next release * fix(models): apply blocked providers filter to non-chat catalog models (cheaper-inference#1752) * chore(release): v3.7.5 — integrate ngrok tunnel and fix models filter (cheaper-inference#1753, cheaper-inference#1752) * chore(release): update changelog format for v3.7.5 * Speed up endpoint initial render * Address endpoint review feedback * Add endpoint loading model translations * fix: resolve build issues and implement memory UPSERT logic (cheaper-inference#1763) * fix: resolve build issues for v3.7.5 and apply memory/translation fixes 1. antigravityHeaders.ts: restore ANTIGRAVITY_LOAD_CODE_ASSIST_* exports for oauth.ts compatibility 2. next.config.mjs: add @ngrok/ngrok to serverExternalPackages and webpack externals to handle native .node modules 3. Memory system: UPSERT logic to prevent duplicate entries with same apiKeyId + key 4. Chinese translations: complete CLI tools and memory dashboard localizations 5. Test fixes: unique keys for pagination tests to comply with unique constraint Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: address Gemini Code Assist review feedback 1. store.ts: add expires_at to UPDATE statement in UPSERT logic - Previously, expires_at was not being persisted to database on update - This caused state mismatch between returned Memory object and actual DB row 2. package-lock.json: revert react-markdown registry to official npmjs.org - Mirror-specific registry URL (npmmirror.com) should not be in lockfile Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * fix(antigravity): normalize Gemini bridge payloads (cheaper-inference#1769) * fix(antigravity): normalize Gemini bridge payloads Clamp Claude bridge output tokens, use Gemini-valid system roles and tool names, and serialize antigravity requests from a cloned body so Cloud Code payload shaping stays valid. * fix(cli): stop fallback after unsafe known paths Preserve known-path security checks by stopping command discovery when a configured CLI path is suspicious or non-executable, instead of falling through to PATH discovery. * test(memory): make query result assertion deterministic Avoid relying on database result ordering when checking filtered memory keys so the unit suite remains stable across runs. * fix(review): preserve safe cloning and CLI reasons Handle non-cloneable antigravity request bodies without throwing and preserve specific CLI known-path failure reasons instead of masking them as not_found. * fix(sse): propagate AbortSignal to pre-fetch semaphore and rate-limit awaits (cheaper-inference#1771) When a combo target takes too long, the request-level deadline fires and calls abortController.abort() on the stream controller, but the abort signal never reaches pending awaits in acquireAccountSemaphore() or withRateLimit(). These awaits sit between stream controller creation and executor.execute(), causing requests to hang indefinitely past the 600s deadline. Pass streamController.signal to both functions so they can respond to abort events and terminate early when the request deadline expires. Signed-off-by: wucm667 <stevenwucongmin@gmail.com> * Fix model sync import handling (cheaper-inference#1755) * Fix model sync import handling * Align model import storage semantics * Address model review feedback * fix(codex): stabilize copilot responses reasoning and tool replay (cheaper-inference#1750) * chore(xiaomi): Update Xiaomi provider model list (cheaper-inference#1759) * Move DB health to management API (cheaper-inference#1757) * Move DB health to management API * Address DB health review feedback * fix(kiro): support organization IDC OAuth with regional endpoints and refresh (cheaper-inference#1754) * fix(kiro): support organization IDC OAuth with regional endpoints and refresh * fix(kiro): refresh IDC tokens with stored region --------- Co-authored-by: ngocdb <ngocdb@ngocdb.local> * chore(workflows): add strict PR contributor credit policy - Add ABSOLUTE PROHIBITION section to review-prs.md - Add PR PROHIBITION rule to resolve-issues.md - Add contributor credit rule to AGENTS.md Review Focus - Based on audit finding: 37 PRs had code absorbed without merge credit * chore(release): acknowledge 29 community contributors with retroactive credit This commit formally recognizes 29 contributors whose code was manually integrated across releases v3.4.0 through v3.7.4 without proper GitHub merge credit. Their PRs were resolved locally due to merge conflicts but closed instead of merged, preventing them from appearing in the Contributors graph. We have updated our workflows to ensure this never happens again. Co-authored-by: Randi <55005611+rdself@users.noreply.github.com> Co-authored-by: Benson K B <4044180+benzntech@users.noreply.github.com> Co-authored-by: clousky2020 <33016567+clousky2020@users.noreply.github.com> Co-authored-by: Raxxoor <7317522+dhaern@users.noreply.github.com> Co-authored-by: Jason Landbridge <15127381+JasonLandbridge@users.noreply.github.com> Co-authored-by: slewis3600 <35925982+slewis3600@users.noreply.github.com> Co-authored-by: Markus Hartung <12826053+hartmark@users.noreply.github.com> Co-authored-by: Hernan Javier Ardila Sanchez <204746071+herjarsa@users.noreply.github.com> Co-authored-by: 3_1_3_u <5846351+andruwa13@users.noreply.github.com> Co-authored-by: Paijo <14921983+oyi77@users.noreply.github.com> Co-authored-by: i1hwan <35260883+i1hwan@users.noreply.github.com> Co-authored-by: xandr0s <1709302+xandr0s@users.noreply.github.com> Co-authored-by: backryun <24198422+backryun@users.noreply.github.com> Co-authored-by: Owen <36758131+kang-heewon@users.noreply.github.com> Co-authored-by: Ravi Tharuma <25951435+RaviTharuma@users.noreply.github.com> Co-authored-by: Chris <3751981+christopher-s@users.noreply.github.com> Co-authored-by: Wellington Fonseca <5421548+wlfonseca@users.noreply.github.com> Co-authored-by: Ethan Hunt <136065060+only4copilot@users.noreply.github.com> Co-authored-by: tombii <6607822+tombii@users.noreply.github.com> Co-authored-by: AndrewDragonIV <7906124+AndrewDragonIV@users.noreply.github.com> Co-authored-by: Danh Thanh <50534210+dt418@users.noreply.github.com> Co-authored-by: Will F <30637450+willbnu@users.noreply.github.com> Co-authored-by: defhouse <232128212+defhouse@users.noreply.github.com> Co-authored-by: Skydwest <186351198+mercs2910@users.noreply.github.com> Co-authored-by: zenobit <6384793+zen0bit@users.noreply.github.com> Co-authored-by: Ivan <16905671+razllivan@users.noreply.github.com> Co-authored-by: foxy1402 <45601526+foxy1402@users.noreply.github.com> Co-authored-by: Luan Dias <65574834+luandiasrj@users.noreply.github.com> Co-authored-by: Sergei Korolev <891832+knopki@users.noreply.github.com> Co-authored-by: dail45 <69967573+dail45@users.noreply.github.com> * fix(combo): include 429 in provider circuit breaker to stop infinite retry on exhausted quotas (cheaper-inference#1767) Previously, PROVIDER_FAILURE_ERROR_CODES only included {408, 500, 502, 503, 504}, meaning 429 responses never counted toward the circuit breaker threshold. This caused exhausted accounts to be retried every 3-5 seconds indefinitely instead of being blocked by the provider breaker. Adding 429 ensures persistent rate limiting triggers the circuit breaker after the configured failure threshold, giving the provider time to recover. * fix(claude): respect client thinking/effort params to prevent forced quota drain (cheaper-inference#1761) Previously, OmniRoute unconditionally injected thinking: {type: 'adaptive'} and output_config: {effort: 'high'} for Claude Opus 4.7 in Claude Code client requests. This caused Claude Max 5h quota to drain in ~15 minutes. Now checks the original client body: if thinking or output_config are explicitly set (even to null or a different value), the injection is skipped. Users can opt-out by sending thinking: null or output_config: {effort: 'low'}. * Add MseeP.ai badge to README.md (cheaper-inference#1727) Integrated into release/v3.7.5 * chore(docs): update CHANGELOG for PR cheaper-inference#1727 * fix(tests): update stream-utils assertion for responses api compliance * feat: Fix support for claude-cli using Gemini provider (cheaper-inference#1779) Integrated into release/v3.7.5 * fix(codex): align client identity metadata (cheaper-inference#1778) Integrated into release/v3.7.5 * fix(blackbox-web): correct cookie name and populate session/subscription fields (cheaper-inference#1776) Integrated into release/v3.7.5 * Fix Codex /responses/compact passthrough (cheaper-inference#1777) Integrated into release/v3.7.5 * test(reasoning-cache): isolate DB state using mkdtempSync to prevent 401 middleware errors * chore(release): v3.7.5 — integrate remaining PRs and finalize stability * chore(config): remove local patch artifacts and trim workspace config Delete temporary patch scripts and local OMC session files that should not ship with the repository. Also remove the Next.js config file and expand editor and TypeScript exclusions to ignore large local workspace directories and reduce unnecessary indexing. * fix(antigravity): cap Claude bridge output tokens (cheaper-inference#1785) Integrated into release/v3.7.5 * fix(codex): stabilize Copilot responses replay state (cheaper-inference#1791) Integrated into release/v3.7.5 * fix(chatgpt-web): restore validator + expand model catalog to ChatGPT Plus tier (cheaper-inference#1792) Integrated into release/v3.7.5 * fix(antigravity): scrub internal OmniRoute headers (cheaper-inference#1794) Integrated into release/v3.7.5 * fix(grok-web): fix Grok validator and cookie parsing (cheaper-inference#1793) Integrated into release/v3.7.5 * chore(release): v3.7.5 — finalize changelog for LTS patch * feat(api-keys): add rename support in permissions modal Add an editable key name field at the top of the permissions modal, allowing users to rename API keys alongside existing permission settings. The backend already supported name updates via PATCH /api/keys/:id — this wires the UI to send the name field and refreshes the key list on success. Changes: - Add keyName state and text input to PermissionsModal - Update handleUpdatePermissions to validate and send name in PATCH body - Add integration test for rename via PATCH (valid, empty, too-long names) - Update E2E mock to handle PATCH requests * chore(release): finalize v3.7.5 LTS release with schema and db initialization fixes * test: fix json escaping in stream-utilities test * fix(build): restore next.config.mjs that was accidentally deleted * fix(sse): decrement pending requests on passthrough mode failure (cheaper-inference#1798) Integrated into release/v3.7.5 * fix(grok-web): repair validator probe + accept full cookie blobs (cheaper-inference#1793) Integrated into release/v3.7.5 * docs(i18n): sync documentation updates to 40 languages --------- Signed-off-by: wucm667 <stevenwucongmin@gmail.com> Co-authored-by: diegosouzapw <diegosouzapw@users.noreply.github.com> Co-authored-by: R.D. <rogerproself@gmail.com> Co-authored-by: clousky2020 <33016567+clousky2020@users.noreply.github.com> Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> Co-authored-by: cloudy <37777261+uwuclxdy@users.noreply.github.com> Co-authored-by: wucm667 <109257021+wucm667@users.noreply.github.com> Co-authored-by: Randi <55005611+rdself@users.noreply.github.com> Co-authored-by: ivan-mezentsev <ivan@mezentsev.me> Co-authored-by: backryun <bakryun0718@proton.me> Co-authored-by: Dao Bao Ngoc <42265865+daongoc315@users.noreply.github.com> Co-authored-by: ngocdb <ngocdb@ngocdb.local> Co-authored-by: Benson K B <4044180+benzntech@users.noreply.github.com> Co-authored-by: Raxxoor <7317522+dhaern@users.noreply.github.com> Co-authored-by: Jason Landbridge <15127381+JasonLandbridge@users.noreply.github.com> Co-authored-by: slewis3600 <35925982+slewis3600@users.noreply.github.com> Co-authored-by: Markus Hartung <12826053+hartmark@users.noreply.github.com> Co-authored-by: Hernan Javier Ardila Sanchez <204746071+herjarsa@users.noreply.github.com> Co-authored-by: 3_1_3_u <5846351+andruwa13@users.noreply.github.com> Co-authored-by: Paijo <14921983+oyi77@users.noreply.github.com> Co-authored-by: i1hwan <35260883+i1hwan@users.noreply.github.com> Co-authored-by: xandr0s <1709302+xandr0s@users.noreply.github.com> Co-authored-by: backryun <24198422+backryun@users.noreply.github.com> Co-authored-by: Owen <36758131+kang-heewon@users.noreply.github.com> Co-authored-by: Ravi Tharuma <25951435+RaviTharuma@users.noreply.github.com> Co-authored-by: Chris <3751981+christopher-s@users.noreply.github.com> Co-authored-by: Wellington Fonseca <5421548+wlfonseca@users.noreply.github.com> Co-authored-by: Ethan Hunt <136065060+only4copilot@users.noreply.github.com> Co-authored-by: tombii <6607822+tombii@users.noreply.github.com> Co-authored-by: AndrewDragonIV <7906124+AndrewDragonIV@users.noreply.github.com> Co-authored-by: Danh Thanh <50534210+dt418@users.noreply.github.com> Co-authored-by: Will F <30637450+willbnu@users.noreply.github.com> Co-authored-by: defhouse <232128212+defhouse@users.noreply.github.com> Co-authored-by: Skydwest <186351198+mercs2910@users.noreply.github.com> Co-authored-by: zenobit <6384793+zen0bit@users.noreply.github.com> Co-authored-by: Ivan <16905671+razllivan@users.noreply.github.com> Co-authored-by: foxy1402 <45601526+foxy1402@users.noreply.github.com> Co-authored-by: Luan Dias <65574834+luandiasrj@users.noreply.github.com> Co-authored-by: Sergei Korolev <891832+knopki@users.noreply.github.com> Co-authored-by: dail45 <69967573+dail45@users.noreply.github.com> Co-authored-by: MseeP.ai <mseep@skydeck.ai> Co-authored-by: Markus Hartung <mail@hartmark.se> Co-authored-by: Raxxoor <manker_lol@hotmail.com> Co-authored-by: Jack <5443152+hijak@users.noreply.github.com> Co-authored-by: Sergey Morozov <tr0st@bk.ru> Co-authored-by: payne <baboialex95@gmail.com> Co-authored-by: Antigravity Assistant <bot@antigravity.local> Co-authored-by: Andrew Munsell <andrew@wizardapps.net>
* docs(changelog): record PR cheaper-inference#1748 for next release * fix(models): apply blocked providers filter to non-chat catalog models (cheaper-inference#1752) * chore(release): v3.7.5 — integrate ngrok tunnel and fix models filter (cheaper-inference#1753, cheaper-inference#1752) * chore(release): update changelog format for v3.7.5 * Speed up endpoint initial render * Address endpoint review feedback * Add endpoint loading model translations * fix: resolve build issues and implement memory UPSERT logic (cheaper-inference#1763) * fix: resolve build issues for v3.7.5 and apply memory/translation fixes 1. antigravityHeaders.ts: restore ANTIGRAVITY_LOAD_CODE_ASSIST_* exports for oauth.ts compatibility 2. next.config.mjs: add @ngrok/ngrok to serverExternalPackages and webpack externals to handle native .node modules 3. Memory system: UPSERT logic to prevent duplicate entries with same apiKeyId + key 4. Chinese translations: complete CLI tools and memory dashboard localizations 5. Test fixes: unique keys for pagination tests to comply with unique constraint Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: address Gemini Code Assist review feedback 1. store.ts: add expires_at to UPDATE statement in UPSERT logic - Previously, expires_at was not being persisted to database on update - This caused state mismatch between returned Memory object and actual DB row 2. package-lock.json: revert react-markdown registry to official npmjs.org - Mirror-specific registry URL (npmmirror.com) should not be in lockfile Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * fix(antigravity): normalize Gemini bridge payloads (cheaper-inference#1769) * fix(antigravity): normalize Gemini bridge payloads Clamp Claude bridge output tokens, use Gemini-valid system roles and tool names, and serialize antigravity requests from a cloned body so Cloud Code payload shaping stays valid. * fix(cli): stop fallback after unsafe known paths Preserve known-path security checks by stopping command discovery when a configured CLI path is suspicious or non-executable, instead of falling through to PATH discovery. * test(memory): make query result assertion deterministic Avoid relying on database result ordering when checking filtered memory keys so the unit suite remains stable across runs. * fix(review): preserve safe cloning and CLI reasons Handle non-cloneable antigravity request bodies without throwing and preserve specific CLI known-path failure reasons instead of masking them as not_found. * fix(sse): propagate AbortSignal to pre-fetch semaphore and rate-limit awaits (cheaper-inference#1771) When a combo target takes too long, the request-level deadline fires and calls abortController.abort() on the stream controller, but the abort signal never reaches pending awaits in acquireAccountSemaphore() or withRateLimit(). These awaits sit between stream controller creation and executor.execute(), causing requests to hang indefinitely past the 600s deadline. Pass streamController.signal to both functions so they can respond to abort events and terminate early when the request deadline expires. Signed-off-by: wucm667 <stevenwucongmin@gmail.com> * Fix model sync import handling (cheaper-inference#1755) * Fix model sync import handling * Align model import storage semantics * Address model review feedback * fix(codex): stabilize copilot responses reasoning and tool replay (cheaper-inference#1750) * chore(xiaomi): Update Xiaomi provider model list (cheaper-inference#1759) * Move DB health to management API (cheaper-inference#1757) * Move DB health to management API * Address DB health review feedback * fix(kiro): support organization IDC OAuth with regional endpoints and refresh (cheaper-inference#1754) * fix(kiro): support organization IDC OAuth with regional endpoints and refresh * fix(kiro): refresh IDC tokens with stored region --------- Co-authored-by: ngocdb <ngocdb@ngocdb.local> * chore(workflows): add strict PR contributor credit policy - Add ABSOLUTE PROHIBITION section to review-prs.md - Add PR PROHIBITION rule to resolve-issues.md - Add contributor credit rule to AGENTS.md Review Focus - Based on audit finding: 37 PRs had code absorbed without merge credit * chore(release): acknowledge 29 community contributors with retroactive credit This commit formally recognizes 29 contributors whose code was manually integrated across releases v3.4.0 through v3.7.4 without proper GitHub merge credit. Their PRs were resolved locally due to merge conflicts but closed instead of merged, preventing them from appearing in the Contributors graph. We have updated our workflows to ensure this never happens again. Co-authored-by: Randi <55005611+rdself@users.noreply.github.com> Co-authored-by: Benson K B <4044180+benzntech@users.noreply.github.com> Co-authored-by: clousky2020 <33016567+clousky2020@users.noreply.github.com> Co-authored-by: Raxxoor <7317522+dhaern@users.noreply.github.com> Co-authored-by: Jason Landbridge <15127381+JasonLandbridge@users.noreply.github.com> Co-authored-by: slewis3600 <35925982+slewis3600@users.noreply.github.com> Co-authored-by: Markus Hartung <12826053+hartmark@users.noreply.github.com> Co-authored-by: Hernan Javier Ardila Sanchez <204746071+herjarsa@users.noreply.github.com> Co-authored-by: 3_1_3_u <5846351+andruwa13@users.noreply.github.com> Co-authored-by: Paijo <14921983+oyi77@users.noreply.github.com> Co-authored-by: i1hwan <35260883+i1hwan@users.noreply.github.com> Co-authored-by: xandr0s <1709302+xandr0s@users.noreply.github.com> Co-authored-by: backryun <24198422+backryun@users.noreply.github.com> Co-authored-by: Owen <36758131+kang-heewon@users.noreply.github.com> Co-authored-by: Ravi Tharuma <25951435+RaviTharuma@users.noreply.github.com> Co-authored-by: Chris <3751981+christopher-s@users.noreply.github.com> Co-authored-by: Wellington Fonseca <5421548+wlfonseca@users.noreply.github.com> Co-authored-by: Ethan Hunt <136065060+only4copilot@users.noreply.github.com> Co-authored-by: tombii <6607822+tombii@users.noreply.github.com> Co-authored-by: AndrewDragonIV <7906124+AndrewDragonIV@users.noreply.github.com> Co-authored-by: Danh Thanh <50534210+dt418@users.noreply.github.com> Co-authored-by: Will F <30637450+willbnu@users.noreply.github.com> Co-authored-by: defhouse <232128212+defhouse@users.noreply.github.com> Co-authored-by: Skydwest <186351198+mercs2910@users.noreply.github.com> Co-authored-by: zenobit <6384793+zen0bit@users.noreply.github.com> Co-authored-by: Ivan <16905671+razllivan@users.noreply.github.com> Co-authored-by: foxy1402 <45601526+foxy1402@users.noreply.github.com> Co-authored-by: Luan Dias <65574834+luandiasrj@users.noreply.github.com> Co-authored-by: Sergei Korolev <891832+knopki@users.noreply.github.com> Co-authored-by: dail45 <69967573+dail45@users.noreply.github.com> * fix(combo): include 429 in provider circuit breaker to stop infinite retry on exhausted quotas (cheaper-inference#1767) Previously, PROVIDER_FAILURE_ERROR_CODES only included {408, 500, 502, 503, 504}, meaning 429 responses never counted toward the circuit breaker threshold. This caused exhausted accounts to be retried every 3-5 seconds indefinitely instead of being blocked by the provider breaker. Adding 429 ensures persistent rate limiting triggers the circuit breaker after the configured failure threshold, giving the provider time to recover. * fix(claude): respect client thinking/effort params to prevent forced quota drain (cheaper-inference#1761) Previously, OmniRoute unconditionally injected thinking: {type: 'adaptive'} and output_config: {effort: 'high'} for Claude Opus 4.7 in Claude Code client requests. This caused Claude Max 5h quota to drain in ~15 minutes. Now checks the original client body: if thinking or output_config are explicitly set (even to null or a different value), the injection is skipped. Users can opt-out by sending thinking: null or output_config: {effort: 'low'}. * Add MseeP.ai badge to README.md (cheaper-inference#1727) Integrated into release/v3.7.5 * chore(docs): update CHANGELOG for PR cheaper-inference#1727 * fix(tests): update stream-utils assertion for responses api compliance * feat: Fix support for claude-cli using Gemini provider (cheaper-inference#1779) Integrated into release/v3.7.5 * fix(codex): align client identity metadata (cheaper-inference#1778) Integrated into release/v3.7.5 * fix(blackbox-web): correct cookie name and populate session/subscription fields (cheaper-inference#1776) Integrated into release/v3.7.5 * Fix Codex /responses/compact passthrough (cheaper-inference#1777) Integrated into release/v3.7.5 * test(reasoning-cache): isolate DB state using mkdtempSync to prevent 401 middleware errors * chore(release): v3.7.5 — integrate remaining PRs and finalize stability * chore(config): remove local patch artifacts and trim workspace config Delete temporary patch scripts and local OMC session files that should not ship with the repository. Also remove the Next.js config file and expand editor and TypeScript exclusions to ignore large local workspace directories and reduce unnecessary indexing. * fix(antigravity): cap Claude bridge output tokens (cheaper-inference#1785) Integrated into release/v3.7.5 * fix(codex): stabilize Copilot responses replay state (cheaper-inference#1791) Integrated into release/v3.7.5 * fix(chatgpt-web): restore validator + expand model catalog to ChatGPT Plus tier (cheaper-inference#1792) Integrated into release/v3.7.5 * fix(antigravity): scrub internal OmniRoute headers (cheaper-inference#1794) Integrated into release/v3.7.5 * fix(grok-web): fix Grok validator and cookie parsing (cheaper-inference#1793) Integrated into release/v3.7.5 * chore(release): v3.7.5 — finalize changelog for LTS patch * feat(api-keys): add rename support in permissions modal Add an editable key name field at the top of the permissions modal, allowing users to rename API keys alongside existing permission settings. The backend already supported name updates via PATCH /api/keys/:id — this wires the UI to send the name field and refreshes the key list on success. Changes: - Add keyName state and text input to PermissionsModal - Update handleUpdatePermissions to validate and send name in PATCH body - Add integration test for rename via PATCH (valid, empty, too-long names) - Update E2E mock to handle PATCH requests * chore(release): finalize v3.7.5 LTS release with schema and db initialization fixes * test: fix json escaping in stream-utilities test * fix(build): restore next.config.mjs that was accidentally deleted * fix(sse): decrement pending requests on passthrough mode failure (cheaper-inference#1798) Integrated into release/v3.7.5 * fix(grok-web): repair validator probe + accept full cookie blobs (cheaper-inference#1793) Integrated into release/v3.7.5 * docs(i18n): sync documentation updates to 40 languages --------- Signed-off-by: wucm667 <stevenwucongmin@gmail.com> Co-authored-by: diegosouzapw <diegosouzapw@users.noreply.github.com> Co-authored-by: R.D. <rogerproself@gmail.com> Co-authored-by: clousky2020 <33016567+clousky2020@users.noreply.github.com> Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> Co-authored-by: cloudy <37777261+uwuclxdy@users.noreply.github.com> Co-authored-by: wucm667 <109257021+wucm667@users.noreply.github.com> Co-authored-by: Randi <55005611+rdself@users.noreply.github.com> Co-authored-by: ivan-mezentsev <ivan@mezentsev.me> Co-authored-by: backryun <bakryun0718@proton.me> Co-authored-by: Dao Bao Ngoc <42265865+daongoc315@users.noreply.github.com> Co-authored-by: ngocdb <ngocdb@ngocdb.local> Co-authored-by: Benson K B <4044180+benzntech@users.noreply.github.com> Co-authored-by: Raxxoor <7317522+dhaern@users.noreply.github.com> Co-authored-by: Jason Landbridge <15127381+JasonLandbridge@users.noreply.github.com> Co-authored-by: slewis3600 <35925982+slewis3600@users.noreply.github.com> Co-authored-by: Markus Hartung <12826053+hartmark@users.noreply.github.com> Co-authored-by: Hernan Javier Ardila Sanchez <204746071+herjarsa@users.noreply.github.com> Co-authored-by: 3_1_3_u <5846351+andruwa13@users.noreply.github.com> Co-authored-by: Paijo <14921983+oyi77@users.noreply.github.com> Co-authored-by: i1hwan <35260883+i1hwan@users.noreply.github.com> Co-authored-by: xandr0s <1709302+xandr0s@users.noreply.github.com> Co-authored-by: backryun <24198422+backryun@users.noreply.github.com> Co-authored-by: Owen <36758131+kang-heewon@users.noreply.github.com> Co-authored-by: Ravi Tharuma <25951435+RaviTharuma@users.noreply.github.com> Co-authored-by: Chris <3751981+christopher-s@users.noreply.github.com> Co-authored-by: Wellington Fonseca <5421548+wlfonseca@users.noreply.github.com> Co-authored-by: Ethan Hunt <136065060+only4copilot@users.noreply.github.com> Co-authored-by: tombii <6607822+tombii@users.noreply.github.com> Co-authored-by: AndrewDragonIV <7906124+AndrewDragonIV@users.noreply.github.com> Co-authored-by: Danh Thanh <50534210+dt418@users.noreply.github.com> Co-authored-by: Will F <30637450+willbnu@users.noreply.github.com> Co-authored-by: defhouse <232128212+defhouse@users.noreply.github.com> Co-authored-by: Skydwest <186351198+mercs2910@users.noreply.github.com> Co-authored-by: zenobit <6384793+zen0bit@users.noreply.github.com> Co-authored-by: Ivan <16905671+razllivan@users.noreply.github.com> Co-authored-by: foxy1402 <45601526+foxy1402@users.noreply.github.com> Co-authored-by: Luan Dias <65574834+luandiasrj@users.noreply.github.com> Co-authored-by: Sergei Korolev <891832+knopki@users.noreply.github.com> Co-authored-by: dail45 <69967573+dail45@users.noreply.github.com> Co-authored-by: MseeP.ai <mseep@skydeck.ai> Co-authored-by: Markus Hartung <mail@hartmark.se> Co-authored-by: Raxxoor <manker_lol@hotmail.com> Co-authored-by: Jack <5443152+hijak@users.noreply.github.com> Co-authored-by: Sergey Morozov <tr0st@bk.ru> Co-authored-by: payne <baboialex95@gmail.com> Co-authored-by: Antigravity Assistant <bot@antigravity.local> Co-authored-by: Andrew Munsell <andrew@wizardapps.net>
[3.7.5] — 2026-04-29
✨ New Features
🐛 Bug Fixes
defaultproperties from tool-call JSON schemas during egress to prevent injection errors ([BUG] Anthropic→OpenAI tool translation: optional fields not omitted in tool_use, WebSearch returns empty result blocks #1782)quota_snapshotstable to core DB schema initialization to prevent startup failures on fresh installs🛠️ Maintenance
🏆 Community Contributors Acknowledgment
We identified that 37 community PRs across past releases (v3.4.0 → v3.7.4) were manually integrated into release branches but closed instead of properly merged through GitHub, preventing contributors from receiving merge credit on their profiles. We sincerely apologize for this oversight and have since updated our workflows to ensure this never happens again.
The following contributors had their code and ideas integrated across multiple releases without proper merge credit. Thank you for your invaluable contributions to OmniRoute:
Tests