| Files | Last commit | Last update |
|---|---|---|
feat(google): CLIProxyAPI parity for Gemini inbound, Interactions API and Google outbound fidelity (#8771) * docs(devlog): Google protocol parity roadmap against CLIProxyAPI Adds devlog/_plan/261115_google_protocol_cpa_parity: a 73-row parity matrix comparing CLIProxyAPI (e96d4c4) with opencodex dev (d842c1352e) and PR #8742 (d21ed22ea8) for generateContent, the Interactions API and the Live API, inbound and outbound (AI Studio, Vertex, Antigravity); Google primary-doc facts; the three lane research appendices; and diff-level phase docs 010-070 (Interactions inbound and state, Gemini inbound fidelity, Google error classification, response and request fidelity, delivery). Folded one independent audit round. * feat(gemini): Interactions API inbound on the Chat path Adds POST /v1beta/interactions and the stable POST /v1/interactions. A new codec (src/protocols/codecs/gemini-interactions.ts) maps the create body (string, content or step input; system_instruction; function tools; generation_config incl. thinking_level, thinking_summaries and tool_choice; text response_format) onto one Chat Completions body, and maps the answer back to an Interaction (object "interaction", Google's total_* usage names, requires_action for tool turns, incomplete for length and content filters) or to the Interactions SSE vocabulary (interaction.created, step.start, step.delta, step.stop, interaction.completed, error). Requests the proxy cannot honor are 400s: agent, background, webhook_config, environment, agent_config, hosted tools, non-text content or output, and previous_interaction_id until stored interactions land. The path classifier also owns /v1beta|v1/interactions/{id} (GET, DELETE -> 404) and .../{id}/cancel (POST -> 501), with geminiRouteAllows giving the by-id route two methods for the 405 Allow header and the loopback allowlist. Parity row INT-01..12, 18, 19 in devlog/_plan/261115_google_protocol_cpa_parity/001_parity_matrix.md. * feat(gemini): Interactions state for previous_interaction_id, get and delete Stored interactions live in a process-memory store (src/server/gemini-interactions-store.ts) owned by the salted principal of the data-plane key the caller presented, or one shared "loopback" owner for keyless loopback callers; any other owner sees a missing id. Bounds: 24 h from the last write, 1,000 entries / 64 MiB in total, 200 entries / 16 MiB per owner, and per entry the smaller of 8 MiB and the inbound body limit. The store is on the app-owned memory roster as gemini_interactions (README retained-store count 14 -> 15 in all eight pages). previous_interaction_id continues the stored history; tools, system_instruction and generation_config stay interaction-scoped. A chained function_result must answer a pending call of the stored turn's last answer and lead the input; a resend of those calls is dropped. Turns are stored only after they settle (inside the bridge's frames() for streams). GET returns the stored resource with its input steps, DELETE answers {}, cancel 400 for an owned interaction and 404 otherwise. Buffered answers report x-opencodex-interaction-stored. Parity rows INT-13, INT-14, INT-15. * feat(gemini): carry thought signatures on the Chat wire; countTokens and limits Gemini 3 rejects a replayed functionCall that lost its thoughtSignature. The Chat wire now carries it on tool_calls[].extra_content.google.thought_signature, the carrier the Responses wire already uses: - inbound: src/chat/inbound.ts (bridge) and the Antigravity chat-ir lane read a normalized value, so a malformed or oversized echo is dropped instead of failing the request; - outbound: src/chat/outbound.ts copies a function_call item's extra_content onto streamed, buffered and SSE-folded tool calls, and the direct-encoder lane passes the adapter's provider metadata through adapter-events to the Chat encoder; - native Chat declines a body carrying extra_content (opaque-state-stripped) unless the destination is a Google host, so Google's opaque state never reaches another vendor; - the Gemini inbound codec (#8742) maps thoughtSignature to and from that carrier. Also on the Gemini inbound surface: :countTokens answers a local estimate under the key's model scope (embed methods stay 501), model resources carry inputTokenLimit/outputTokenLimit when configured or known from model metadata and advertise countTokens, `$alt=sse` is accepted, and STATUS_NAMES follows the google.rpc HTTP mapping (405 UNIMPLEMENTED). Parity rows IN-09, CW-01, IN-05, IN-12, IN-03, IN-13. * feat(google): classify errors by ErrorInfo and RetryInfo; quarantine TOS suspensions Google states why it refused and when to retry in typed google.rpc details, which opencodex ignored in favor of prose. A new src/adapters/google-error-details.ts reads ErrorInfo.reason, RetryInfo.retryDelay, ErrorInfo.metadata.quotaResetDelay and an "after <duration>" phrase in protobuf or Go duration syntax ("30s", "1h2m3s"). - QUOTA_EXHAUSTED and INSUFFICIENT_G1_CREDITS_BALANCE are quota exhaustion on every Google label; RATE_LIMIT_EXCEEDED is a transient rate limit on Antigravity even with quota-like prose (CPA's rule); AI Studio and Vertex keep the prose rule. - A retry delay stated only in the 429 body now floors the next send like Retry-After, and a delay past the retry ceiling returns the 429 with that Retry-After. - A 403 with a typed ErrorInfo TOS_VIOLATION (object root, code 403, PERMISSION_DENIED) gets a fixed "Antigravity account suspended (TOS_VIOLATION)" message that never echoes Google's appeal text; dispatch quarantines that account through the existing fenced path and rotates. The hold is stored as verify_account plus needsReauthCause "tos_violation", so builds that do not know the cause still keep the account out of the pool; it shows as needs-reauth(tos) and health "Account suspended". Parity rows AG-X1, AG-X2, AG-X3. * feat(google): report thinking and tool-use tokens; fail closed on every Google mode Google response fidelity, parity rows OUT-V38, OUT-V42, AG-R7, OUT-V40, OUT-V43 (devlog 050). - Usage: output tokens are candidatesTokenCount + thoughtsTokenCount and input includes toolUsePromptTokenCount, so output includes reasoning as on every other adapter and the Gemini inbound codec's subtraction becomes exact. - Fail closed on AI Studio as on Vertex and Antigravity: MALFORMED_FUNCTION_CALL, MAX_TOKENS after a started call, and UNEXPECTED_TOOL_CALL / MALFORMED_RESPONSE when no visible output (non-thought text, inline image or function call) reached the client in the whole turn. The error carries status 502 / upstream_error, so its HTTP outcome no longer depends on message wording (Vertex MALFORMED_FUNCTION_CALL moves from 400 to 502). - IMAGE_SAFETY, IMAGE_PROHIBITED_CONTENT and MALFORMED_RESPONSE with output end as content_filter; the request log keeps MALFORMED_RESPONSE and MISSING_THOUGHT_SIGNATURE. - promptFeedback.blockReason with no candidates is a 400 invalid_prompt naming only the reason enum: a terminal refusal for Codex and a hard stop for combos, never echoing Google's message. - The new test registrations move to the adapters layout shard; layout.json was over 2000 lines. Release note: AI Studio turns that ended with MALFORMED_FUNCTION_CALL, UNEXPECTED_TOOL_CALL or MALFORMED_RESPONSE and no visible output used to complete empty and now fail with 502; reported Google output tokens now include thinking tokens. * feat(google): Vertex us/eu rep hosts, inline audio, and error tool results Google request fidelity, parity rows OUT-V04, OUT-V29, OUT-V27 (devlog 060). - Vertex project endpoints for the us and eu multi-regions use aiplatform.{us,eu}.rep.googleapis.com (CPA's vertexBaseURL table); global and regions are unchanged. vertexHostForLocation runs after the existing location validation, so the authority still cannot be changed by configuration. - WAV and MP3 audio parts are sent as inline_data on every Google mode instead of a text marker; Claude models on Antigravity keep the marker (Antigravity hands them an Anthropic message). - A tool result flagged isError is sent as functionResponse.response.error, the key Google documents for a failed call and CPA writes on its Gemini and Antigravity translators. - AG-M1 (nest Antigravity tool-result images) is deferred: ADR-7279 extends nesting only with live upstream evidence for a route, and the pinned CCA layouts were live fixes. Release note: Vertex providers configured with location us or eu now call the rep hosts. * docs(devlog): record shipped parity rows in the Google CPA matrix Rows closed by 010-060 now name their commit; AG-M1 stays deferred (ADR-7279 evidence bar). * docs: Gemini API and Interactions endpoints, Google 429 and suspension handling proxy-formats gains the /v1beta/models and /v1beta/interactions section and overview rows; adapters documents typed 429 handling; providers-accounts documents needs-reauth(tos). English and Spanish; docs-site build passes (775 pages). * fix(gemini): body admission for countTokens; keep AI Studio's body retry wait Independent review round 1. - :countTokens reads, decodes and parses a body but ran outside the shared inbound-body allowance, so concurrent encoded or raised-limit bodies were not bounded in aggregate. isGeminiBodyPath now covers it and the dispatcher runs it inside runAdmittedBodyWork (no turn slot); a saturated allowance queues it and refuses with the Google-shaped 503. - AI Studio returns raw upstream errors, so a 429 whose wait is stated only in the body (RetryInfo, quotaResetDelay) and exceeds the retry ceiling, or is quota exhaustion, reached callers without Retry-After. The raw answer is now copied with that header; Google's body and status are unchanged and an explicit upstream header still wins. - Interactions is_error to Google's response.error is deferred as matrix row INT-20. * fix(google): keep a 429's body retry wait on the last attempt; resync README hashes Review round 2 and PR CI. - Running out of retry attempts (or the send budget) returned the final 429 before its body was read, so a wait stated only in RetryInfo or quotaResetDelay never became Retry-After. Every final answer now goes through finalGoogleAnswer, raw (AI Studio) or normalized (Vertex, Antigravity); an explicit upstream header still wins. - readme/i18n-manifest.json: the retained-store count edit in README.md was mirrored in every locale in the same commit, so each locale's source hash now names the current README. * fix(gui): check the gateway-client action status before reading its body #8742's action client parsed every answer as an action body, which react-doctor's no-fetch-response-used-without-status-check rule blocks (it fails #8742's own run too). It now branches on the status first: a 2xx or one of the route's refusals (400, 404, 409) is read for its code, and any other status is reached but has no body to parse. No visible change. Co-authored-by: Andrew <wsy1731095638@gmail.com> * fix(gemini): CodeRabbit round on the Google parity branch - Retry delays: only a positive delay counts, and a typed RetryInfo outranks quotaResetDelay and untyped details regardless of order. - Interactions store: a write evicts only its own owner's entries, also at the global limits; when the store is full and the writer owns nothing, the interaction is not kept, so one key never removes another key's history. Only the app-owned memory budget evicts across owners. - Interactions codec: a conversation-level system/developer turn keeps its position, and thinking_summaries "none" reaches Chat as include_reasoning false. - Gemini codec: enum/const/default/example(s) values are copied, not lower-cased; an explicit includeThoughts false is carried. - Model discovery reports a configured output limit as outputTokenLimit. - ocx account gives "Account suspended" an appeal action. - Unknown /v1beta/* paths are a JSON 404 instead of the dashboard page. - Docs: best-effort Interactions retention with its limits, rejected input types, safety_settings ignored, streamed 502 after output; one-line inventory table row. Each code fix has a regression test that failed before it. * fix(gemini): normalize schemas under property names that look like keywords CodeRabbit follow-up. normalizeSchema copied the value of a property named enum, const, default, example or examples as a literal, so its schema kept an upper-case type. Schema maps (properties, patternProperties, $defs, definitions) now normalize every value as a schema while their names stay data; literal keywords inside a schema are still copied verbatim. Also: discovery coverage for a configured output limit overriding generated metadata, and the proxy-formats docs (en/es) name the per-interaction storage limit and the x-opencodex-interaction-stored header. * docs(devlog): plan the agy acp provider (#8480) and add its parity row Owner decision on #8480: a separate local provider over the official agy acp transport. 080 is the diff-level plan; AG-ACP is the matrix row. Implementation ships as its own PR from dev. * docs(devlog): revise the agy acp plan after audit round 1 Resolves B1-B6: fallback bit on the run-turn and sidecar paths, apiKeyPool and resolved-target config checks, state-tracked option selection, --sandbox with request-review and a private workspace, fixed error text, a complete terminal table. * docs(devlog): revise the agy acp plan after audit round 2 R2-1 pure structural selector resolution for config checks; R2-2 the permission engine is the promised boundary and --sandbox is unpromised hardening; R2-3 fail closed when the Gemini model proof is lost. * docs(devlog): phase-aware model proof in the agy acp plan (audit round 3) * fix(google): preserve retry and account trust across the rebase * fix(gemini): bound interaction retention and protect resource responses * fix(google): fall back to a body delay for invalid Retry-After * fix(gui): identify suspended OAuth accounts and show appeal guidance * docs(google): record parity gaps and shipped ACP locality * fix(google): validate Retry-After once before replay * fix(gemini): publish custom model input limits in discovery * test(antigravity): cover redaction in terminal suspension responses * docs(google): keep retry guidance within combined-tree limits * test(retry): cover fractional delays and Google header precedence * fix(google): keep strict Retry-After parsing at the adapter boundary * fix(gemini): honor ACP turn policy on Interactions ingress --------- Co-authored-by: Andrew <wsy1731095638@gmail.com> | 2 days ago | |
feat(oauth): require an explicit acknowledgement before any Google Antigravity sign-in (#8820) * feat(oauth): require an explicit Antigravity risk acknowledgement on the API and CLI Google can disable a whole Google account's Cloud Code Assist access when the Antigravity login is used through a third-party proxy. Until now the CLI and the management API started that login with no warning at all. POST /api/oauth/login and POST /api/oauth/accounts/import now refuse google-antigravity with 403 antigravity_risk_not_acknowledged unless the body carries acknowledgeAntigravityRisk: true. ocx login, ocx account login (including reauth) and ocx account import print the warning and continue only on a y answer at a terminal, or with --accept-antigravity-risk for non-interactive use; otherwise they exit before touching the proxy or reading an import document. * feat(gui): show the Antigravity login as a danger dialog behind an unticked checkbox The dashboard used the generic high-risk copy for Google Antigravity. It now gets its own danger dialog: red frame and title with a warning icon, an error notice quoting Google's ban message verbatim, a checklist of consequences (account-wide ban, appeal not guaranteed, use only an account you can afford to lose), the Gemini API key as the safer path, and a red Continue that stays disabled until the acknowledgement box is ticked. The box starts unticked on every open. Login, add-account and reauth send acknowledgeAntigravityRisk only from the dialog's Continue, and the Cockpit Tools import now goes through the same dialog, including a file picked straight into the hidden input. All eleven locales carry the copy; the unused generic saferPath key is removed. * docs: document the Antigravity account-ban risk and acknowledgement Adds the ban warning to the providers guide, the CLI reference for ocx login / account login / account import, and the 403 refusal to the management API table, plus the devlog plan unit for this change. * fix(oauth): refuse agent-driven Antigravity acknowledgement and scope the warning copy Review follow-ups: - The CLI gate refuses a run driven by a coding agent or CI runner even with --accept-antigravity-risk or a y answer (isAgentDriven), following the user-consent rule; AGENTS_INSTALL.md documents the boundary, including the admin-token API route the server cannot tell apart. - A prompt counts as interactive only when its own stream is a terminal. - The copy says Google may permanently disable Gemini and Antigravity on the account, which is what the reported ban message describes, rather than the whole Google account. - The Cockpit picker's acknowledgement also disarms when the window regains focus, for engines without the file input cancel event. * test: keep cli-account.test.ts at its file-size cap The Antigravity import gate test moves to cli-account-login-options.test.ts, which already drives cmdAccount against a recording fetch; the remaining edits to cli-account.test.ts are line-neutral. * docs(es): mirror the Antigravity acknowledgement docs in Spanish Adds the danger admonition to es/guides/providers.md, the CLI paragraph to es/reference/cli/providers-accounts.md and the 403 to the es management API row, so the Spanish parity test sees the same code spans, admonitions and status codes. The English CLI paragraph is rewrapped so no code span breaks across lines. * fix(gui): hold every Cockpit pick behind its own Antigravity warning Review follow-up. The import no longer arms an acknowledgement before a file is chosen: the button opens the picker directly and every selected file, however it was chosen, waits at the warning dialog and is imported only from that file's Continue. This removes the window in which a dismissed picker (in engines without the cancel event) could let a later direct pick reuse an earlier acceptance. Also documents the acknowledgement contract in the owned structure/providers/antigravity-account-import.md. * test: keep layout.json under the size threshold and let the login-error test pass the Antigravity gate layout.json reached 2000 lines with the new entry, which the file-size ratchet treats as a new oversized file; the entry now shares the line of its neighbour. oauth-login-cli-errors runs ocx login google-antigravity in a child to test discovery diagnostics, so it passes the explicit acknowledgement with the agent seam pinned, and strips the always-printed warning before asserting stderr. | 5 days ago | |
fix: Aside/Prime send system instead of developer; name DashScope's developer-role rejection (#8864) (#8867) * docs(devlog): plan #8864 Aside/Prime developer role and DashScope rejection * fix(clients): Aside and Prime send system instead of developer; Alibaba presets fold the role Aside and Prime reuse Pi's models.json builder but never received the #5674 developer-role option, so their Pi runtime sent the system prompt as "developer" and DashScope answered 400. Both export paths now write compat.supportsDeveloperRole: false (session affinity stays Pi-only), and a shape-discovered regression test covers every Pi-builder client. alibaba-token-plan and alibaba-token-plan-intl now record foldDeveloperRoleToSystem, so the native Chat passthrough folds developer into system for any client, including saved configs and an overridden DashScope base URL. Refs #8864 * docs(es): Prime and Aside integrations state the developer-role compat flag * fix(errors): name a developer-role rejection without quoting the provider A provider that rejects the developer role (DashScope "developer is not one of ...", Moonshot "role 'developer' is not allowed", serde "unknown variant `developer`") used to reach the client as "The provider rejected the request." with nothing in the request log, so the cause took a manual bisect to find. A fixed-phrase verdict on upstream 4xx evidence now selects a proxy-owned actionable sentence on Chat and Messages, and provider bodies record the existing closed pair messages.role/enum, which the dashboard already shows. No upstream text reaches a client body, row, usage.jsonl or the console; a body this proxy formatted keeps validation-only capture. Refs #8864 * test(errors): cover the Messages boundary and keep each row's role pair separate Final review follow-ups for #8864: a Messages request through startServer now asserts the fixed developer-role sentence end to end, and each row gets a fresh messages.role/enum array instead of a shared module constant that a valid list would store by reference. Refs #8864 * test(errors): keep nested objects visible in the boundary console capture The boundary test asserts that no provider text reaches console output, but String() turned an object argument into "[object Object]", so a regression that logged { error: { message } } would have passed. Non-string arguments are now printed with Bun.inspect at full depth. Addresses the CodeRabbit review on #8867. Refs #8864 | 4 days ago | |
fix(codex): restore never reverts config.toml changes made after injection (#8878) * docs(devlog): plan key-level Codex journal restore that never reverts later config edits * test(codex): reproduce restore losing concurrent writes and adopting stale injected values Adds a behavior-neutral seam (runBeforeCodexConfigPublishForTests) at the three restore publications of config.toml and child-process tests under temporary CODEX_HOME/OPENCODEX_HOME/HOME. On this commit 11 of 18 fail for the recorded reasons: start/client journal recovery leaves routing behind, a stale journal is retained, an in-place write landing during restore is overwritten, a rename during restore crashes compensation, an app rewrite to an earlier injection's URL is adopted as user-owned, and a dropped model_catalog_json is not drift. The 7 passing cases guard stop/disconnect/untouched behavior. * fix(codex): restore never reverts config.toml changes made after injection Every OpenCodex publication of config.toml now goes through publishCodexConfigIfUnchanged, which refuses retryably and writes nothing when the file no longer holds the bytes the decision was computed from. The lock's inode witness already refused an atomic rename (how Codex writes) but missed an in-place writer, which restore renamed over while reporting success. Failure compensation reverts config.toml only while it still holds this lock section's own publication, and skips surfaces already holding their preimage, so a foreign write is never rolled back and a refused restore names only what it could not put back. Start and mismatched-client journal recovery use the three-way owned-field restore that disconnect already used. They used to refuse as "incomplete" after any later edit and leave a dead loopback URL in config.toml. The journal records every root routing URL injections under it wrote (injectedRootHistory, optional, version stays 1). A Codex app write-back of an earlier injection's URL is stripped and re-pointed instead of being snapshotted as user-owned. Drift detection also covers a dropped model_catalog_json and such write-backs, and the first heal in a process warns once. structure/codex-home.md and structure/overview.md (INV-RESTORE-02) state the invariant; tests/codex-integration/codex-restore-later-edits.test.ts enforces it. * fix(codex): keep the drift heal notice on the info channel; repair the write-back drift test literal * docs(codex): state compensation ownership for held and unheld callers * fix(codex): report an unreadable preimage surface as unrestored instead of throwing * docs(devlog): record verification and PR for the journal restore unit * test(layout): register the journal restore tests in their own shard to keep layout.json under the ratchet * test(codex): reproduce the CodeRabbit round on #8878 before fixing it Adds setAfterCompensationDecisionForTests, a behavior-neutral seam between compensation's ownership decision and its publish/remove, and seven cases that fail on this commit: - start and mismatched-client recovery crash on ConfigWriteContentChanged when config.toml changes while they restore (journal kept, start aborted); - compensation overwrites (replace) or deletes (remove) a write that lands after it decided the file was OpenCodex's publication; - reinjection over a journal without injectedRootHistory loses the URL it replaces; - a multiline model_catalog_json assignment, and an unparseable config, are reported as a missing catalog key. * fix(codex): fold the CodeRabbit round on #8878 - Compensation publishes over, or removes, config.toml only while the file still holds the exact bytes its ownership check hashed; it no longer re-reads before publishing, and removing a file this section created checks content first. - Start and mismatched-client journal recovery treat ConfigWriteContentChanged as "defer to the next start" (journal kept, start continues); the failed-connect rollback records it as a partial rollback. - markJournalInjectedState records the outgoing latest URLs in injectedRootHistory before overwriting them, so journals written before the history keep them. - Drift decides key presence from the parsed document (multiline assignments are present; an unparseable file reports nothing missing). - The publication docstring, INV-RESTORE-02 and structure/codex-home.md state the remaining compare-then-rename window, and the native-child digest limit is documented where it is recorded. | 4 days ago | |
fix(combos): carry combo hop, attempt timing, Devin admission and Anthropic pool failover fixes (#8871) Lane R1 of the 2026-11-17 triage round: hop on unoffered tool-call refusals (#8563), time a rotated attempt from its own send (#8749), resolve Devin context windows for combo input admission (#8575), and scale Anthropic pool failover within the funded send budget (#8857), with review follow-ups. Co-authored-by: Omar <13569609+omar391@users.noreply.github.com> Co-authored-by: Andrew (agent) <wsy1731095638@gmail.com> Co-authored-by: Vadym O <bolein95@gmail.com> | 4 days ago | |
fix(providers): R3 triage — Devin Claude Code rewrites, OpenCode Go include, Command Code root schemas, JEV continuation effort (#8873) Lane R3 of the 2026-11-17 triage round: rewrite the Claude Code sentences Cognition refuses for Devin (#8703), normalize the OpenCode Go output include (#8827), normalize root tool schemas for Command Code's Claude upstreams (#8591), and keep JEV's decided effort through tool-loop continuations (#8861, continuation fix only), with review follow-ups. Co-authored-by: 袁威豪 <137183806+jhl880617@users.noreply.github.com> Co-authored-by: Ingwannu <186453546+Ingwannu@users.noreply.github.com> Co-authored-by: Mursyid <ciddhere@gmail.com> Co-authored-by: tachyon6 <songjunjun62754@gmail.com> | 4 days ago | |
fix(auth): refuse Anthropic redirects, keep chat caller restrictions, gate explicit SSH routes (#8868) Lane R4 of the 2026-11-17 triage round: refuse redirects on Anthropic bearer requests (#8758), keep caller restrictions visible to the Chat tool-dispatch guard (#8814), and require owner authority for explicit SSH routes (#8806). Co-authored-by: Vadym O <bolein95@gmail.com> Co-authored-by: Epinephrine <luvs01@hanmail.net> | 4 days ago | |
fix: R5 platform and test-hygiene lane (zh-CN/ko-KR checked stop, Windows git config, test network guard) (#8870) Lane R5 of the 2026-11-17 triage round: read zh-CN and ko-KR schtasks status words in the Windows checked stop (#8863), use /dev/null as the empty global git config on Windows (#8746), and refuse real network egress from the test process (#8848). Co-authored-by: andrew05060414 <wsy1731095638@gmail.com> | 4 days ago | |
feat(models): add GPT-6.2 Sol and Astra and upgrade subagent defaults (#8893) * feat(models): add GPT-6.2 Sol and Astra and upgrade subagent defaults * test: sync GPT-6.2 catalog and CI expectations * test(claude): verify the current generated agent roster * feat(providers): seed GPT-6.2 Sol and Astra preemptively on GPT-6.1 Sol providers Mirror each provider's GPT-6.1 Sol row on Kiro, CodeBuddy, Devin (dashed uids and derived pricing), TokenLab, Modelflare and GitHub Copilot, and opt both OpenRouter routes into Fast, following the Sonnet 5.5 and GPT-6.1 Terra preemptive-seed precedent. No provider default changes. Generated with [Devin](https://devin.ai) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com> * docs(providers): separate preemptive Copilot Responses defaults from observed rejections GPT-6.2 Sol/Astra and GPT-6 Sol/Luna default to Responses on Copilot without an observed /chat/completions rejection, so list them as preemptive built-in defaults instead of in the observed-rejection list, in English and every translation that repeats it. Generated with [Devin](https://devin.ai) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com> * fix(models): give GPT-6.2 Sol and Astra their 128K output ceiling The configured-native template moved to GPT-6.2 Sol, but neither GPT-6.2 model had a generated metadata row, so native output admission and configured natives resolved an unknown output ceiling. Add the verified openai, openai-codex and openrouter rows (128K output, API pricing, native/API windows) to the metadata snapshot and regenerate. Generated with [Devin](https://devin.ai) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com> --------- Co-authored-by: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com> | 3 days ago | |
feat(gui): drop Start at Login from the desktop Getting started and let Home hide it (#9007) * feat(gui): drop Start at Login from the desktop checklist and let Home hide Getting started Inside the OpenCodex desktop app, the app's install and first run already turn Start at Login on, so Home and the Set up hub no longer list that step there (Home reads "X of 2"). The browser dashboard keeps its step unchanged. Home's Getting started card gains a Hide control, remembered per browser profile in localStorage; Recent errors and the other Home cards stay. Hiding offers Undo for that visit, and the Set up hub brings the card back with Show on Home. Copy for all 11 locales lives in getting-started-copy.ts. * fix(gui): keep the Getting started choice when storage reads fail Writable but unreadable storage made Hide revert on the next read; the store now answers with the last choice made in this window. Only localStorage events reconcile the choice, so a sessionStorage clear elsewhere cannot drop an unsaved one. The Set up hub test asserts where focus lands after Show on Home. | 1 day ago | |
fix(link): Remote Link Child auth, Child status page and Windows probe (#8972, #8999, #8985) (#9018) * docs(devlog): plan the fb2100 Remote Link Child auth and Windows units * fix(link): read Windows_NT through the quote cmd leaves on a Windows Child (#8972) Win32-OpenSSH runs an exec request as cmd.exe /c "<command>". With the bare "cmd /c echo %OS%" probe cmd keeps those quotes and prints Windows_NT" (reproduced locally with cmd.exe), so Home-initiated enrollment of every Windows Child failed with an opaque 502. - WINDOWS_OS_PROBE is now "cmd.exe /c echo %OS%", which prints a clean Windows_NT. - isWindowsNtAnswer accepts one trailing quote in all three parsers, so a shell that still adds it is understood. - Platform discovery failures throw SshDiagnosticError with a bounded stderr reason; sshRunnerErrorHint reports it, so probe_failed and ocx link cleanup say why instead of looking like an SSH outage. * fix(link): let a Child's admin token read the Home's quota pool (#8999) GOAT reads a Home's ocx.quota-pool/1 through each Remote Link Child at /api/machine/hub-relay/api/quota-pool with the Child's local admin token. The link-Child relay branch admitted only a dashboard session and answered 401, so every Child showed "OpenCodex pool unavailable". The Child's listener now also admits its own admin-token principal for exactly GET and HEAD of /api/quota-pool (linkChildAdminReadAllowed). The parsed path must match exactly, the relay's refresh-key refusal still applies, every other relay path keeps requiring the dashboard session, and the Home only ever sees the link key. structure/quota-pool.md and structure/remote-link.md name the credential. * fix(gui): the Child's Remote Link page reads its own status and shows how to disconnect (#8985) On a connected Remote Link Child the page asked the Home relay for /api/link/status, which the relay refuses as a Home management read (403 link_child_read_only). The page then showed "Remote link request could not be completed", the off switch, and no way out of the link. - RemoteLink reads its status from statusApiBase, which App sets to the Child's machine listener, and gates that read and the page on the Child's own session; Home operations keep the shared plane. - gui/src/api.ts classifies /api/link/status and /api/link/diagnostics on a link Child's machine target as machine-plane requests, so they carry the Child's session instead of none. - A connected Child shows "Disconnect this computer" with the commands that end the link: ocx disconnect --expect-link <id> and ocx restart. The dashboard still cannot change the link itself; the Child listener refuses dashboard-session writes by design. - Copy lives in gui/src/i18n/remote-link-child-copy.ts for all 11 locales. | 1 day ago | |
feat(images,videos,codex): Grok Imagine 2.0/Video 1.5 bridges and quota/reset parity (#9013) * docs(devlog): plan Grok media and quota parity against live xAI/OpenAI behavior * feat(images): default the Grok image bridge to grok-imagine-image-2.0 and send quality grok-imagine-image-quality was retired on 2026-11-02 and is served as 2.0 at low quality, so the old default ignored the requested quality. The bridge and the /v1/images relay now default to grok-imagine-image-2.0, forward quality (low/standard -> low, medium/high/hd -> medium, auto) to 2.0 only, accept the full xAI aspect-ratio set, map 1024x1536/1536x1024 to 2:3/3:2, forward up to five edit sources, and return xAI validation errors only as canonical messages rebuilt from a closed vocabulary. * feat(videos): support Grok Imagine Video 1.5, 1080p and pinned frames The video bridge defaulted to the classic grok-imagine-video (720p, no frame pins). It now defaults to grok-imagine-video-1.5-lite, advertises only what the configured model accepts (1.5: 1080p, first/last frame, four keyframes, 14 references capped at 720p; Lite: first frame and 1080p; classic: first frame and seven references; unknown models keep the old surface), forwards image, last_frame, keyframes and reference_images in xAI's shape, and reports xAI validation errors and failure codes only as canonical messages. Image inputs are https or data:image URLs, or references to images the image bridge downloaded in this process, which resolve to the recorded xAI URL; no local file is read, and those references are only sent to the pinned xAI host. * fix(images): bound quality error details and align edit-source limits Review follow-up for the image bridge. A quality-list validation message is returned only with at most the four known values, each once, rebuilt in canonical order. The /v1/images/edits relay drops duplicate sources before the five-source limit, matching callXaiImages, and the guides now say the limit applies to that relay while the image_gen bridge tool sends one image. * fix(videos): refuse unsupported video settings and keep safe poll detail Review follow-up for the video bridge. An explicit resolution or aspect ratio the configured model cannot honour is now a tool error instead of being dropped, which would have bought a video at the upstream default. A permanent poll failure keeps the allowlisted xAI detail (attached as a separate field by the client) and still never the raw upstream message. * fix(codex): keep non-expiring reset credits and calibrate Pro 100/500 capacity The official Codex client types a reset credit's expires_at as optional, but the management DTO dropped any credit without a string expiry. Such rows are now kept with expires_at: null, together with the display fields the backend sends (reset_type, status, title, description, bounded) and total_earned_count; ids stay private. The auto-redeemer, which schedules by expiry, still sees only expiring credits. Pool capacity weighted Pro 100 (prolite) at the pre-2026-10-29 5x and had no Pro 500 (promax), which counted as one Plus seat. They are now 10x and 50x Plus, per OpenAI's 2026-10-29 doubling and staff-published multipliers. * docs(images): record remaining Grok/OpenAI non-LLM decisions and fix the video default comment * fix(gui): show non-expiring Codex reset credits as "No expiry" The reset-credit endpoint now returns expires_at: null for credits without an expiry, which the dashboard passed to the date formatter as the Unix epoch, so such a credit read "Expires Jan 1, 1970 (0d left)" and was marked urgent. The credit item now shows a localized "No expiry" line without urgency; dated credits are unchanged. * fix: guard reset-credit auto-redeem order, refuse invalid video durations Final-review follow-up. Consume names no credit and the dashboard says the oldest is used first. Now that non-expiring credits are visible, the auto-redeemer also receives them and stands down ("order-not-guaranteed") when a credit granted before the one it protects could be spent in its place. An explicit video duration outside 1-15 whole seconds (or a non-numeric one) is a tool error instead of being clamped or dropped before a paid call; the numeric string form xAI accepts ("8") still works. The pool-quota regression expectations follow the Pro 100 weight of 10 (weighted weekly 71 % instead of 66 %). * fix(codex): auto-redeem fails closed on tied or unorderable reset-credit grants An equal grant time has no known consume tie-breaker and a missing or unparseable one cannot be ordered, so either now stops automatic redemption like an older credit does. The redeemer's inspect reads every upstream row, passing a row without a grant time on as unorderable; the display DTO is unchanged. * fix(codex): never plan auto-redeem for a reset credit without a usable grant time A credit whose grant time is missing or unparseable has no identity the redeemer can protect, so it is never selected; it still blocks other plans as an unorderable competitor. The guard also rejects an unparseable planned grant time before checking competitors. * fix: treat identical reset credits as a tie, scope quality to 2.0, advertise all image ratios Final-approval review follow-up. - Auto-redeem excludes exactly one row as the planned credit; a second row with the same grant and expiry is a distinct credit with an unknown consume order, so the redeemer stands down. - quality is forwarded to grok-imagine-image-2.0 only; another 2.x model has to opt in explicitly rather than receive a field it may reject on a paid call. - The image_gen tool advertises every aspect ratio the xAI client forwards (adds 9:19.5, 19.5:9, 9:20, 20:9, 1:2, 2:1). * fix(images): refuse an edit relay request with any unusable image source Final-approval review follow-up: a source the relay cannot represent as a URL (an empty string, a file_id entry, an object without a URL, a non-string) used to be dropped while the remaining sources went to a paid edit. The whole request is now a 400 before any upstream call. * fix(images): validate every edit relay source as an https or image data URL Final-approval review follow-up: an explicit null source was still ignored and any non-empty string was forwarded. Each supplied source (null included) must now be an https URL without credentials or a base64 PNG/JPEG/WebP data URL, otherwise the request is a 400 before any upstream call. * fix(images): check every URL property of an edit entry and require canonical base64 Final-approval review follow-up: an images[] entry contributed only its first URL-bearing property, so {image_url: null, url} slipped through and a second valid property was dropped; and any run of base64 characters (====) passed as an image data URL. Each image_url / image_url.url / url property is now a source in its own right, and a data URL needs whole base64 quads with padding only at the end. * fix(images): require canonical base64 padding bits in edit data URLs Final-approval review follow-up: a payload such as AB== passed the alphabet, length and padding-position checks although its unused bits are non-zero. The payload must now round-trip through decode and re-encode unchanged. * fix(videos): refuse mode-specific settings xAI cannot honour (CodeRabbit) CodeRabbit follow-up. Before any paid call the video tool now also refuses: 1080p with last_frame or keyframes (they make it reference-to-video, which stops at 720p, like references already did); a keyframe outside the effective clip when duration is omitted (xAI's 8 s default); image plus references on the classic model, which rejects the mix; and an aspect ratio on plain image-to-video, where xAI keeps the image's own ratio. An xAI error body that stalls or is aborted while its detail is read no longer hides the HTTP status: detail extraction is best effort, so a permanent poll failure (e.g. 401) still stops polling instead of being retried as transient. * docs(videos): translate the video_gen parameter and model tables in all locales CodeRabbit follow-up: the eight translated video-bridge guides carried the new parameter rows, the refusal note and the model-table headers in English. * fix(videos): reject colliding keyframe slots and over-long classic reference clips (CodeRabbit) xAI snaps keyframes to a 1/3-second grid and rejects two in one slot, and the classic model limits reference-to-video to 10 seconds. Both are now refused before the paid submit; 1.5 references keep 15 seconds. | 1 day ago | |
feat(gui): switch models on/off from Providers and an Overall tab in Models (#9008) * feat(gui): switch models on/off from Providers and an Overall tab in Models Providers > provider > Models lists every model, including off ones, each with an on/off switch writing PUT /api/model-visibility through the shared queue, owned by ProviderWorkspaceShell so a provider switch never cancels a write. Models gets an Overall tab (#models/overall): every model across providers in one searchable list with All/On/Off filters, sharing the catalog's rows, visibility rule, write queue and poll guard. * fix(gui): disable provider switches during Delete consent; flag stale Overall list Review follow-ups: switches render disabled while the Delete dialog holds the flight, and Overall shows the catalog's load-failure notice when a background refresh fails after a good load. Search drops the ReactNode model label. * test: register the Overall tab hash in the Models routing contract * docs(devlog): record fb-models outcome * fix(gui): keep the provider visibility waiter out of render closures react-doctor (react-compiler MemoDependencies) flagged refresh's useCallback for capturing a per-render settle function. Settle is now a module-level helper over the ref, so every callback closes only over stable refs. * docs: describe the provider Models tab count and the queued-change warning in every locale CodeRabbit review on #9008: the provider count paragraph now says it is the provider-list count and describes the Models tab's on/total header; fr, ja, ko, ru, tr and zh-tw gain the warning that unsent queued visibility changes are discarded on leaving, which en, es and zh-cn already carry. * docs: state the full on rule and the Providers leave warning in every locale CodeRabbit round 2 on #9008: a model is on when it is not disabled and it is native, its provider has no allowlist, or the allowlist includes it (the modelIncluded rule); leaving the Providers page mid-save can cancel the request and the next load shows whether it saved. * docs(tr): name the native-model exception without the local-server sense of yerel CodeRabbit round 3 on #9008. | 1 day ago | |
docs(fab00): record maintainer decisions (Fabric = opencodex subsystem, TS-native Supervisor); clean remaining ?N section-ref and diagram artifacts to pure ASCII | 3 months ago |
| Files | Last commit | Last update |
|---|---|---|
| 2 days ago | ||
| 5 days ago | ||
| 4 days ago | ||
| 4 days ago | ||
| 4 days ago | ||
| 4 days ago | ||
| 4 days ago | ||
| 4 days ago | ||
| 3 days ago | ||
| 1 day ago | ||
| 1 day ago | ||
| 1 day ago | ||
| 1 day ago | ||
| 3 months ago |