Release Notes
Added
- OpenAI-compatible JSON error format: All error responses (400, 502, 403) now
return {"error": {"message": ..., "type": ..., "code": ...}} with
Content-Type: application/json. Official OpenAI SDKs can parse errors correctly.
- CORS headers:
Access-Control-Allow-Origin: * on all public endpoints
(/v1/chat/completions, /v1/embeddings, /v1/models). Enables browser-based
OpenAI SDKs (JavaScript, Vercel AI SDK) to call the proxy directly.
x-request-id correlation: Every response now includes an oxllm-<hex>
x-request-id header, visible in both success and error responses. The ID is
also attached to all log lines and OTel spans (proxy.request_id attribute)
for end-to-end request tracing.- Provider status OTel gauge:
llm_proxy.provider.status (0=Healthy, 1=Cooldown,
2=Tripped) now emits on every circuit state transition. Previously the gauge was
defined but never populated.
- Unit & integration tests: 7 new tests covering CORS preflight, JSON error
format parsing, and x-request-id presence on both success and error responses.
Changed
localhost_only middleware returns JSON error body instead of empty 403.- CORS layer registered globally, before all per-route middleware.
- All upstream-failure
warn! log lines now include request_id field.
Documentation
- Added CORS support,
x-request-id, and JSON error format to README features list.
- Updated architecture doc with CORS subsection, request correlation docs, and
updated middleware diagram.
Install oxllm 0.1.11
Install prebuilt binaries via shell script
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/planetf1/oxllm/releases/download/v0.1.11/oxllm-installer.sh | sh
Install prebuilt binaries via Homebrew
brew install planetf1/tap/oxllm
Download oxllm 0.1.11
