airmux is a self-hosted LLM gateway. Applications send requests using the Chat Completions, Responses, or Messages API.
airmux authenticates the workspace, applies policy, selects a model and provider credential, translates the request,
and records usage.
The managed catalog currently includes 9 providers and 243 models.
You run the whole platform: a webapp, organizations and workspaces, managed provider credentials, live policy, and
persistent usage history with estimated cost.
Quickstart
Start the platform, send a real request, find its cost in the webapp, and enforce your first policy.
What airmux provides
- One inference origin for multiple upstream provider families
- Compatible Chat Completions, Responses, and Messages surfaces
- Workspace-scoped inference keys and permission-limited management keys
- Instance, organization, and workspace provider credentials
- Model, provider, price, budget, parameter, request, denial, and fallback policies
- Buffered and streaming translation with tool calls, structured output, reasoning, images, and PDF inputs when the selected model supports them
- Usage, estimated cost, latency, status, credential scope, and audit records
- Cached versioned configuration so gateways keep serving during a control-plane outage
Choose a path
Run airmux another way
The quickstart deploys the full platform, which is the recommended path. Two alternatives trade capability for
simplicity or scale:
- Gateway only serves inference from local files, with no Docker, Postgres, webapp, or usage history
- Separate services splits the webapp, control plane, and gateways so gateways scale and restart independently
airmux is pre-1.0. Configuration, APIs, and migrations may change before the first stable release.