- EndpointPageClient: full migration including cloud proxy modals,
endpoint sections (chat, embeddings, images, rerank, audio, moderations),
enable/disable cloud flows, sync steps, and status messages
- Expanded endpoint namespace from 10 to 55 keys in en.json and pt-BR.json
- CombosPage, ComboCard, ComboFormModal, TestResultsView: full migration
- Headers, model tags, metrics, strategy descriptions, validation
- Form labels, advanced settings, drag-and-drop model list, actions
- Expanded combos namespace from 19 to 65 keys in en.json and pt-BR.json
- ApiManagerPageClient: full migration including stats cards, table headers,
modals, permissions system, validation messages, and usage tips
- PermissionsModal: search, model selection, allow/restrict toggles
- Expanded apiManager namespace from 18 to 67 keys in en.json and pt-BR.json
isModelAllowedForKey() existed in src/lib/db/apiKeys.ts but was never
called anywhere. API keys with allowedModels restrictions could access
any model through any endpoint.
Changes:
- Add shared enforceApiKeyPolicy() middleware (model restriction + budget)
- Wire it into chat handler (replacing inline budget-only check)
- Wire it into all /v1/* endpoints: embeddings, images/generations,
audio/speech, audio/transcriptions, moderations, rerank
- Wire it into provider-specific endpoints: /v1/providers/[provider]/embeddings,
/v1/providers/[provider]/images/generations
The middleware checks:
1. Model restriction — if key has allowedModels, verify the model is permitted
2. Budget limit — if key has budget configured, verify it hasn't been exceeded
Fixes#130
- Wrap exchangeTokens calls with runWithProxyContext in both 'exchange' and 'poll-callback' actions
- Resolve proxy via provider-level → global → direct fallback chain
- Fixes region-restricted providers (e.g. OpenAI Codex) failing during OAuth setup with unsupported_country_region_territory error
- Fix URL parameter injection: apply encodeURIComponent on providerStorageAlias and providerId in all API calls
- Replace blocking alert() with non-blocking notify.error/notify.success toast notifications
- Add success feedback for model add and delete operations
- Improve error handling: use console.error consistently and add user-facing notifications for import failures
- Check DELETE response status before proceeding with alias removal
- Add deploy-vps.yml: auto-deploys via SSH after Docker Hub publish
- Remove API key management section from Endpoint page (now in API Manager)
- Remove unused Link import from EndpointPageClient.tsx
- Add stats summary cards (total keys, restricted, total requests, models available)
- Add per-key usage statistics (total requests, last used timestamp)
- Add copy button on each key row
- Add color-coded lock/unlock status icons per key
- Bump version to 1.4.0
- Connection tests now route through configured proxy (key → combo → provider → global → direct)
- Compatible providers show friendly labels (OAI-COMPAT, ANT-COMPAT) in request logger
- 26 new unit tests for error classification, token expiry, and display labels
- Version bump to 1.3.1
- Add GET/PATCH endpoints for retrieving and updating API key permissions
- Move API key management from Endpoint page to dedicated API Manager page
- Add allowed_models column to database schema for model-specific access
- Implement caching layer for improved API key validation performance
- Add modelsUrl field to RegistryEntry interface for custom models endpoints
- Configure kilocode provider with dedicated models URL (https://api.kilo.ai/api/openrouter/models)
- Expand kilocode model list from 8 to 26 models including free options:
- openrouter/free (Free Models Router)
- stepfun/step-3.5-flash:free
- arcee-ai/trinity-large-preview:free
- Additional Qwen, DeepSeek, Llama, Mistral, Grok, and Kimi models
- Update validateOpenAILikeProvider to accept custom modelsUrl parameter
- Fix models URL derivation for base URLs ending with /chat/completions
- Add kilocode config to PROVIDER_MODELS_CONFIG
* fix: auto-generate JWT_SECRET at startup if not provided
- Add ensureJwtSecret() in instrumentation.ts to generate random 64-char secret
- Mark JWT_SECRET as non-required in secretsValidator
- Remove fatal error log for missing JWT_SECRET in proxy.ts
Fixes login failure when running via CLI without JWT_SECRET env var.
* feat: improve auth flow and login page UX
- Add setupComplete to settings validation schema for proper persistence
- Support ?tab= query param in settings page for direct tab navigation
- Redesign login page with professional two-column layout
- Add context-aware states for onboarding/password setup flows
- Smooth animations and refined visual hierarchy
* fix: 404 page to use primary color theme (coral red) instead of purple
- fix(chatCore): stream defaults to false per OpenAI spec (body.stream === true)
closes#89
- fix(models): custom OpenAI-compatible providers now appear in /v1/models
Added raw providerId check against activeAliases
closes#90
- fix(OAuthModal): Google OAuth providers force localhost redirect URI
Targeted amber warning + redirect_uri_mismatch error guidance for remote deployments
closes#91
- docs: Add 'OAuth em Servidor Remoto' tutorial in README with 7-step Google Cloud guide
- docs: .env.example prominent warning block for Google OAuth remote setup
- chore: bump version 1.0.6 → 1.0.7
- docs: CHANGELOG entry for v1.0.7
JWT session tokens now last 30 days instead of 24 hours. The middleware
silently refreshes the token when it has less than 7 days remaining,
so active users never get locked out of the dashboard.
Auto-refresh fires on any /dashboard request and sets a fresh 30-day
cookie via Set-Cookie header — completely transparent to the user.
1. Combos page: added Toggle to each ComboCard — disable/enable combos
with optimistic UI update. Disabled combos show opacity-50.
2. /v1/models: disabled combos (isActive=false) are excluded from the
model list.
3. Providers page: toggle is now always visible instead of hover-only.
Previously, if no connections existed in the database, ALL provider
models (378+) were shown as a fallback. Now only combos and custom
models appear when no provider connections are configured.
Models are strictly filtered by activeAliases — only providers with
at least one active (enabled) connection show their models. This
ensures /v1/models reflects exactly what the user has set up.
