Add per-account model allowlists and denylists to frozen routing, account settings, and CPR import/export. Preserve existing scheduling and account state boundaries.
Closes#97
Co-authored-by: Codex <noreply@openai.com>
* feat(auth): add API key workspace
Add persistent API key sessions and enforce admin/key authorization boundaries. Expose key-scoped quotas, trends, health, and paginated usage records through server-projected data.
Share the application shell, system overview, and usage views between admin and key sessions, reorganize frontend domains around view ownership, and omit account, provider, and authentication details from the key workspace.
BREAKING CHANGE: Browser authentication now uses /api/auth/* and cpr_session; existing browser sessions must sign in again.
Validated frontend lint, type checking, production build, backend workspace tests, and live authenticated key views.
Co-authored-by: Codex <noreply@openai.com>
* feat(console): refine key workspace experience
Unify role-aware console routes and use mode/role in authentication contracts.
Reuse request state, stable pagination and performance projections across workspaces.
Refine quota cards, usage panels, empty states and key-aware theme previews.
Skip session restoration on login and retain form inputs until unmount.
Keep the full health timeline grid visible without samples.
Co-authored-by: Codex <noreply@openai.com>
* feat(key-usage): replace shared workspace with focused usage page
Preserve unified admin/key authentication, restore admin pages, and remove the previous client workspace implementation. Add a scoped usage page with metrics, trends, budgets, request logs, and section loading placeholders.
Co-authored-by: Codex <noreply@openai.com>
* feat(gateway): queue requests at key and account concurrency limits
Share one wait deadline across client admission and account selection.
Co-authored-by: Codex <noreply@openai.com>
---------
Co-authored-by: Zyy <zyycn@users.noreply.github.com>
Co-authored-by: Codex <noreply@openai.com>
* feat(auth): add API key workspace
Add persistent API key sessions and enforce admin/key authorization boundaries. Expose key-scoped quotas, trends, health, and paginated usage records through server-projected data.
Share the application shell, system overview, and usage views between admin and key sessions, reorganize frontend domains around view ownership, and omit account, provider, and authentication details from the key workspace.
BREAKING CHANGE: Browser authentication now uses /api/auth/* and cpr_session; existing browser sessions must sign in again.
Validated frontend lint, type checking, production build, backend workspace tests, and live authenticated key views.
Co-authored-by: Codex <noreply@openai.com>
* feat(console): refine key workspace experience
Unify role-aware console routes and use mode/role in authentication contracts.
Reuse request state, stable pagination and performance projections across workspaces.
Refine quota cards, usage panels, empty states and key-aware theme previews.
Skip session restoration on login and retain form inputs until unmount.
Keep the full health timeline grid visible without samples.
Co-authored-by: Codex <noreply@openai.com>
* feat(key-usage): replace shared workspace with focused usage page
Preserve unified admin/key authentication, restore admin pages, and remove the previous client workspace implementation. Add a scoped usage page with metrics, trends, budgets, request logs, and section loading placeholders.
Co-authored-by: Codex <noreply@openai.com>
---------
Co-authored-by: Zyy <zyycn@users.noreply.github.com>
Co-authored-by: Codex <noreply@openai.com>
Forward complete upstream model objects and client versions without rebuilding metadata. Keep account-scoped catalogs in a bounded, coalesced cache with explicit invalidation and preserve model aliases without mixing fields.
Cover service tiers, native metadata, account isolation, cache behavior, and failure handling with regression tests; document the catalog contract.
Co-authored-by: Codex <noreply@openai.com>
Settle only observed USD costs and treat missing costs as zero without
blocking client keys. Preserve known retry costs and idempotent settlement.
Remove reconciliation controls, endpoints, pending records, and obsolete
schema fields; migrate legacy unknown charges to zero without changing
recorded usage.
Display daily and weekly amounts with two decimal places while retaining
full precision for accounting and tooltips. Distinguish missing WebSocket
closing handshakes from TCP resets in connection diagnostics.
Validation: workspace tests with PostgreSQL/Redis, targeted regressions,
strict Clippy, rustfmt, frontend lint/build, and migration checks passed.
Co-authored-by: Codex <noreply@openai.com>
Share account settings across import, single-account editing, and batch editing.
Commit credentials and import settings in the same transaction. Improve modal
actions and fix the flash at the end of light/dark theme transitions.
Reauthenticate each execution against the current key policy so existing
WebSocket connections apply updated RPM, concurrency limits, and enabled state.
Add regression coverage for policy updates and unavailable snapshots.
Validation: all 1,950 backend tests, workspace Clippy, frontend lint/typecheck/build,
migration checks, Compose validation, and workflow lint passed. Completed 24 real
requests with a free account using gpt-5.6-luna to verify daily/weekly budgets,
RPM, concurrency, reconciliation, WebSocket policy updates, and budget persistence
across restarts and usage-log cleanup.
Co-authored-by: Codex <noreply@openai.com>
Return model_not_found with HTTP 404 when all scoped provider catalogs
confirm the requested or mapped model is absent. Explain missing mapping
targets separately and preserve unavailable-provider errors for circuit
blocks, empty account scopes, and unsupported operations.
Cover missing models, aliases, empty and unknown catalogs, provider
fallback, and streaming and buffered HTTP error responses.
Validation: gateway-core and gateway-api tests (541 passed), strict Clippy
for both crates with all targets and features, workspace rustfmt, and diff
whitespace checks passed.
Co-authored-by: Codex <noreply@openai.com>
Persist models declared by standalone image and search requests without
routing them through the text model catalog. Extract image token usage
before delivering the unchanged JSON response, preserving unknown counts
and upstream totals without double-counting image token details.
Validation: 1,474 related backend tests, workspace rustfmt and strict
Clippy passed. One authorized live gpt-image-2 generation reported 18 input,
229 output and 247 total tokens, matching the canonical usage exactly.
Search responses inspected in official source and production logs did not
report usage, so their token counts remain unknown.
Co-authored-by: Codex <noreply@openai.com>
Move Dashboard quota projection to the backend, share pagination request state, and fix independent usage loads while keeping trend switches free of loading overlays.
Stream billing facts within the query budget, share pending authorization and usage validation, and separate Core value ownership and response observations. Remove the unused theme helper and strengthen dependency checks.
Validated with workspace integration tests against isolated PostgreSQL and Redis, strict Clippy, Rust formatting, and frontend typecheck, lint and build.
Co-authored-by: Codex <noreply@openai.com>
- carry underlying redis/serialization error into native continuation logs
- attach provider/account to affinity write failures and timeouts
- log quota cycle summary counts and per-account transient causes
- include upstream status in catalog/quota error diagnostics
- record request_id/method/uri in HTTP trace spans for failed responses
Co-authored-by: Codex <noreply@openai.com>
Own RotationStrategy wire mapping in gateway-core and re-export it from gateway-admin, move ProviderCircuitPolicy into the execution port contract, and rename the Core coordination timeout without a Redis implementation leak.
Co-authored-by: Codex <noreply@openai.com>
Expose exact upstream responses to authenticated connection tests, reload the resulting account state, and classify usage-limit windows as resettable quota exhaustion. Restore non-quota availability after verified credential rotation and refresh quota after reauthorization.
Co-authored-by: Codex <noreply@openai.com>
Log direct Store failures without changing client-visible results, emit connection-test diagnostics before asynchronous persistence, and keep Provider errors intact when observation writes fail. Add a process-wide byte budget, cumulative queue statistics, ordered failure isolation, and bounded shutdown draining to the PostgreSQL observation worker.
Co-authored-by: Codex <noreply@openai.com>
Preserve raw request and response data across HTTP, SSE, and WebSocket adapters, including opaque headers and upstream error envelopes. Remove private payload and catalog gates, keep account switching behind the pre-delivery barrier, and make observation failures unable to replace client-visible protocol data.
Co-authored-by: Codex <noreply@openai.com>
reset_uncommitted_observations left provider_processing_ms set when
discarding a failed attempt's observations, leaking one attempt's upstream
processing time into the next; clear it with the other per-attempt fields.
finish_success fabricated an upstream_status_code of 200 when none was
observed — persist the absent status as NULL instead. Give the "no active
attempt" internal state its own EngineError::NoActiveAttempt instead of
overloading EmptyRoutingPlan, and map it to an internal gateway error.
Remove the unused free function commit_requirement.
Co-authored-by: Claude <noreply@anthropic.com>
Connection tests run through a transient execution store, so a failed probe
left no trace at all: nothing in model_requests, nothing in ops_events, and
nothing in the log at info level. The reason a probe failed — upstream status,
provider error code, send state — was only reachable by reproducing it.
Add a ProbeFailure port on ExecutionStore, defaulting to a no-op so existing
implementations are unaffected, and have the execution service report probe
failures to the persistent store while the probe itself keeps using the
transient one. The Postgres store logs a warning and appends an ops event with
no model_request_id.
Such events already fit the surrounding contracts: the ops error list left
joins model_requests, health metrics inner join it so probes stay out of
business failure rates, and retention already targets rows with a null
model_request_id.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Combine the model-request insert with the first attempt into one ExecutionStore call (default implementation keeps the sequential behavior; the Postgres store overrides it with a single INSERT), removing one database round trip from the pre-first-token critical path. Reuse a session-scoped deadline timer in the event pull loop instead of registering a new global timer per event, and attach continuation session state in place so a uniquely held payload is no longer force-deep-copied by clone-before-make_mut.
Co-authored-by: Claude <noreply@anthropic.com>
Preserve transient credential failures as runtime cooldowns, restore account-aware retry and observability facts, and align Grok model/tool adaptation with Codex clients.
Keep complete SSE events on the live path while retaining only protocol-required tool and continuation state.
Co-authored-by: Codex <noreply@openai.com>
Collapse routing, health, circuit state, account pools, and admin APIs onto the fixed OpenAI and xAI Provider boundary while keeping credential runtime state account-scoped.
Remove configurable provider-instance and routing-catalog persistence, use fixed official endpoints in provider adapters, align account routes and frontend contracts, and mirror provider tests under their source modules.
Co-authored-by: Codex <noreply@openai.com>
Remove unused full-response transports and unconnected domain reservations while keeping test-only setup inside integration tests.
Restore image-generation accounting and OpenAI WebSocket pool observations across the engine, store, and admin boundaries without changing the transparent OpenAI WS/SSE path.
Co-authored-by: Codex <noreply@openai.com>
Add typed compaction operations across the gateway, OpenAI, and xAI providers with provider-owned request, response, accounting, and observability contracts.
Align xAI custom apply_patch calls with the patch function argument and re-encode restored custom calls during provider replay.
Co-authored-by: Codex <noreply@openai.com>
Move control-plane use cases, host lifecycle, provider capabilities, execution orchestration, and worker contracts into their final owners while keeping the application bootstrap as a thin composition root.
Restore verified OpenAI transport, continuation, pooling, fingerprint, accounting, and observability semantics, and retain the validated xAI multi-account integration.
Co-authored-by: Codex <noreply@openai.com>