LunarGate Gateway

Self-hosted AI gateway

Run one OpenAI-compatible endpoint in your infrastructure and route requests across multiple LLM providers with fallback, retries, caching, hot-reloadable config, and optional observability export.

Get started Browse runnable examples Read the configuration reference

Best entry path:

• start with Quickstart if you want the shortest install -> config -> run -> client flow
• jump to Examples overview if you learn better from runnable projects
• keep Configuration overview open when you begin editing real YAML

Get running fast
Install LunarGate, save one config, run it, and call it with a normal OpenAI client.

See real examples
Browse Python, Docker Compose, Node/Express, Streamlit, and lunargate/auto examples.

Learn autorouting
Understand how lunargate/auto and model-selection headers drive tier-based routing.

Track release changes See a human-readable changelog built from tagged gateway releases.

Why teams use it

• One endpoint for every app

Keep your app code on the OpenAI API shape and swap providers behind the gateway.

• Resilience built in

Route by headers, retry transient failures, and cascade to fallback targets automatically.

• Operated from config

Change providers, routing, retry/cache behavior, rate limits, model selection, and Dashboard export settings without rebuilding the binary.

• Observability without lock-in

Export metrics only by default, or opt into prompt and response sharing for request inspection.

The request path

Request -> Optional Inbound Auth (API key) -> Rate Limit -> Cache -> Route Match -> Load Balance
        -> Retry -> Circuit Breaker -> Provider Translation -> LLM Call
        -> Response Translation -> Metrics -> Optional Data Sharing -> Response

Where to actually start:

• use Quickstart for the shortest install -> config -> run -> client path
• use Docker Compose local stack if you want a container-first setup
• use Examples overview if you prefer runnable projects over one-page setup instructions

What is in scope today

• Chat Completions via POST /v1/chat/completions, with regular JSON and SSE streaming responses
• Responses via POST /v1/responses for JSON or SSE and GET /v1/responses for WebSocket mode
• Embeddings via POST /v1/embeddings
• Model listing and lookup via GET /v1/models and GET /v1/models/{model}
• Health and metrics endpoints
• Multi-provider routing and fallback
• In-memory rate limiting and caching
• Hot-reloadable YAML config
• Optional Dashboard observability export

Important security note

Warning

The gateway now supports basic inbound API-key authentication, but it is still safest to run it inside a trusted network or behind an auth-enforcing edge such as an API gateway, reverse proxy, or service mesh.

Documentation map

1. Start with Quickstart for the fastest install -> config -> run -> client flow.
2. Go to Examples overview for runnable Python, Node, Streamlit, and Docker Compose apps based on gateway-examples/.
3. Read lunargate/auto and autorouting if you want the gateway to choose model tiers from one stable client model.
4. Use Routing and fallback for route ordering, fallback chains, and load-balancing strategy.
5. Keep Configuration overview and the detailed config pages open while editing YAML.
6. Read Observability and data sharing before enabling prompt or response export.
7. Read Changelog if you want the release history translated from Git into human-readable notes.