Treat LM Studio, vLLM, Llamafile, Triton, Docker Model Runner,
XInference, and oobabooga as managed local providers with default
base URLs and passthrough model discovery.
Avoid sending empty bearer headers during validation, model discovery,
and execution so self-hosted endpoints work without API keys while
still honoring custom base URLs and localized dashboard hints.
Introduce a custom CLI tools card that generates OpenAI-compatible
env vars and JSON config, and add a cost overview tab with pricing
source visibility.
Add persisted custom eval suites with authenticated CRUD routes and
validation, plus new translator stream transformation tooling and
richer live monitor routing details.
Improve localization coverage across dashboard flows and add unit
tests for custom CLI config, pricing sources, eval suites, and
stream transformation.
Store eval executions with target metadata, expose aggregated scorecard
and recent run history endpoints, and return dashboard-ready eval data
including target options and API key metadata.
Also require management auth for eval read endpoints and preserve
per-case latency, errors, and output snippets so historical results are
more reliable and easier to inspect.
Unify audit access under the logs experience and add richer active
request visibility with sanitized client and provider payload previews.
Expose provider warnings from upstream responses in compliance audit
logs, surface learned rate-limit header data in health monitoring, and
add MCP cache stats and cache flush tools with test coverage.
Also add local provider catalog support for SD WebUI and ComfyUI,
include video generation in endpoint docs and UI, add eval run storage
migrations, and refresh docs and translations to match the expanded
feature set.
The provider registry used PascalCase header keys (e.g. "Anthropic-Version")
while the Claude Code client path in base.ts sets lowercase keys
("anthropic-version"). Since JS object keys are case-sensitive, both keys
coexist in the headers object. When fetch() sends them, HTTP treats them
as duplicates and concatenates the values ("2023-06-01, 2023-06-01"),
causing Anthropic's API to reject the request with a 400 error.
Normalize all Anthropic-specific header keys to lowercase to match the
convention used in executors and the upstream API.
- Stream release: wrap stream body with TransformStream to release
account semaphore only when stream is fully consumed (Thread 1)
- Key scope: remove model from semaphore key — was per-account-per-model,
now truly per-account to match PR motivation (Thread 2)
- Test location: move accountSemaphore.test.ts to tests/unit/ per style guide (Thread 3)
- Fix flaky timestamp assertions in semaphore tests
- DB 마이그레이션 028: provider_connections.max_concurrent 컬럼 추가
- AccountSemaphore: 계정별 FIFO 세마포어 (acquire/release/timeout/block)
- chatCore.ts: 요청 파이프라인에 선제적 cap enforcement 통합
- providers.ts: maxConround-trip round-trip 저장/조회, cleanNulls() 보정
- API: provider limits route에서 maxConcurrent GET/PUT 지원
- UI: provider 연결 상세 페이지 account native cap 입력 필드 + hint
- UI: ResilienceTab combo concurrency 라벨 구분 (combo vs account)
- i18n: en/ko 번역 키 추가
- schemas.ts: maxConcurrent 음수 검증 + null 허용
- 테스트: semaphore 6개, DB round-trip + validation 6개
Reject invalid API keys even when authentication is optional and add
OPTIONS handlers for batch and file routes to support CORS preflights.
Also recover orphaned batches stuck in finalizing after restarts,
include finalizing in pending batch queries, preserve multipart upload
content handling, and fetch remote vision images as data URIs for
Anthropic requests.
* feat(vision-bridge): add automatic image description fallback for non-vision models
Implements VisionBridgeGuardrail (priority 5) that intercepts image-bearing
requests to non-vision models, extracts descriptions via a configurable
vision model (default: gpt-4o-mini), and replaces images with text before
forwarding. Fails open on any error.
- Add VisionBridgeGuardrail class extending BaseGuardrail
- Add visionBridgeHelpers: extractImageParts, callVisionModel, replaceImageParts, resolveImageAsDataUri
- Add visionBridgeDefaults with configurable settings
- Register VisionBridgeGuardrail in GuardrailRegistry at priority 5
- Add 51 unit tests covering all spec scenarios (VB-S01 through VB-S10)
- Dependency injection for getSettings and callVisionModel (testable without SQLite)
Closesdiegosouzapw/OmniRoute#1424
* fix(vision-bridge): resolve Anthropic API, parallel processing, provider keys, structuredClone
