Arcanada-one/arcanada-llm-proxy

A small OpenAI-compatible shim over Arcanada's Model Connector that routes chat completion requests to two backends: Codex (for structured outputs when request.response_format.type=json_schema) and OpenRouter (for json_object/text or absent — per-token vendor-neutral fallback). On Codex 429 or auth errors it automatically falls back to OpenRouter and records the cause in response.x_arcanada.fallback_reason. Intended for internal infrastructure migration (CONN-0045) and to promote the LTM v3 benchmark proxy to a production service. Primary users are backend engineers and services that need OpenAI-compatible endpoints with controlled routing. Key features: - POST /v1/chat/completions OpenAI-compatible API - Local bind 127.0.0.1:4000, Vault-managed bearer auth - OpenTelemetry → Langfuse OTLP, span llm-proxy-request - FastAPI/Starlette/httpx, Python 3.12 - Docs (tutorials/how-to/reference/explanation), quick start, tests, MIT license

AI-related keywords found in this repo's description, topics, or README summary — grouped by category. Each badge links to the corresponding ranking detail page.