Documentation
Docs
Learn how Arion works and how to configure each feature, from listeners and routing to the LLM and MCP gateways.
How to configure Arion
How it works
A listener accepts connections. Requests pass through an ordered filter chain before a route chooses a cluster of upstream endpoints. Arion reads Envoy configuration from a file or over xDS.
Browse the docsFour paths
Choose a starting point
- 01 / Migrate Replace Envoy or nginx Run the Envoy configuration you already have, or move an nginx reverse proxy across. Guide: Migrate from Envoy
- 02 / Gateway Build an API gateway Terminate TLS, authenticate with JWT, rate limit per user and route with retries. Guide: Build an API gateway
- 03 / MCP Build an MCP gateway Turn services into tools agents can call, with discovery and per-tool authorization. Guide: Build an MCP gateway
- 04 / Extend Write a filter Choose WASM, ext_proc or a native Rust HTTP filter and put it in the request path. Docs: Extend Arion
10 groups
Browse by goal
- Get started Install Arion, understand the data plane and proxy your first request. 4 pages
- Concepts Architecture, request flow, the filter chain, configuration and Envoy compatibility. 6 pages
- Traffic Listeners, routing, upstreams, resilience, rate limits and TCP proxying. 6 pages
- Security TLS and mTLS, secret rotation, JWT, RBAC, Cedar policies and CORS. 7 pages
- LLM Gateway One endpoint for several model providers, with failover, budgets and guardrails. Under construction. 1 page
- AI Inference Gateway Route to ready model server Pods with Kubernetes Gateway API, InferencePool, and an endpoint picker. 1 page
- MCP Gateway Tools mode and code mode, REST and OpenAPI backends, federation, search, authorization and TDS. 9 pages
- Extend WASM filters, the Rust SDK, ext_proc and native HTTP filters. 6 pages
- Observe Access logs, logging, metrics, tracing and the admin interface. 5 pages
- Operate Deployment, xDS control planes, performance tuning and troubleshooting. 4 pages