Repository Public
1[2 {3 "version": "1.0.0",4 "date": "2026-11-19",5 "title": "Initial release",6 "changes": [7 "OpenAI-compatible inference API: POST /v1/chat/completions, POST /v1/completions, GET /v1/models and GET /v1/models/{model}.",8 "Server-sent event streaming with keepalive comments, reasoning_content deltas, an optional usage chunk (stream_options.include_usage) and in-stream error events.",9 "Tool calling with tool_choice \"auto\" and \"none\"; each tool call is delivered whole in a single chunk.",10 "JSON output modes (response_format json_object and json_schema), validated server-side with one automatic repair retry on non-streaming requests.",11 "Bearer API keys (mlx_sk_…) stored hashed, scoped per application to models and endpoints, with rotation, expiry and revocation.",12 "Per-key limits: requests per minute, tokens per minute, daily and monthly token quotas, concurrency; x-ratelimit-* and Retry-After headers.",13 "Priority scheduling (admin, internal, client, background) with anti-starvation aging and a bounded queue that applies backpressure.",14 "OpenAI-style error envelope with stable error codes, and a request id on every response (x-request-id).",15 "Machine-readable OpenAPI 3.1 contract at /openapi.json, generated from the backend schemas.",16 "Public portal: documentation with search, interactive Playground on the real API, and a live status page."17 ]18 }19]20