Awesome Infra for AI › LLM Gateways & Proxies

sina2266/Gozar

⭐ 131 Python added to this list on 2026-08-17 repository created 2026-07-12

Gozar is a self-hosted LLM gateway that presents a single OpenAI-compatible /v1 surface to applications while the operator keeps control of credentials, routing and limits. Each project, agent or backend service receives its own Gozar API key; behind that key the gateway routes requests through operator-managed upstream accounts and provider keys, so client code never carries a provider secret and never changes when the upstream does. It implements the familiar OpenAI shapes — /v1/chat/completions with SSE streaming, /v1/embeddings and /v1/models — which makes it a drop-in for the OpenAI SDK, LangChain, LangGraph, Postman, cURL and local agents. Upstream support covers API-key providers such as OpenAI and OpenRouter alongside subscription providers, with a device-code sign-in flow for Codex that avoids the usual broken localhost redirect. Routing is expressed as two-lane fallback chains: one chain identifier holds an LLM lane and an embeddings lane, and every node picks its own account, model and fallback policy. Saved routes are rechecked against current account status and model catalogs, so a removed model or an unavailable account surfaces as a chain health alert rather than a silent production failure, and chat and embedding catalogs are discovered and cached per account. Operators get usage limits, request traces and analytics covering request volume, token usage, per-token and per-account activity, and routing outcomes. Security defaults include encrypted credential storage, fail-closed operator authentication, secret-free logs and password-confirmed key reveal. The stack is FastAPI with a React and TypeScript console, deployed with Docker Compose, and the source is released under the PolyForm Noncommercial licence for self-hosted, non-commercial use.

https://github.com/sina2266/Gozar

llm-gatewayproxyopenai-compatibleroutingfallbackself-hostedapi-keys