Local proxy

atlas serve runs a local, OpenAI-compatible API. Any tool that can talk to OpenAI can use your connected AIs and the same fallback chain as the chat.

atlas serve                  # http://127.0.0.1:4141/v1
atlas serve --port 8080

On start it prints the URL and the token:

OPENAI_BASE_URL=http://127.0.0.1:4141/v1
OPENAI_API_KEY=atlas-...

The token is generated on the first run and saved as serveToken in ~/.the-atlas/config.json. The server listens on 127.0.0.1 only. New keys you add while it runs are picked up without a restart.

Endpoints

Method and path	Auth	What it does
`GET /health`	no	`{"status":"ok"}`
`GET /v1/models`	token	`auto` plus `provider/model` for each connected AI
`POST /v1/chat/completions`	token	Chat completion, streaming or not

Send the token as Authorization: Bearer <token>. /models and /chat/completions also work without the /v1 prefix.

Choosing the model

`model` value	Route
`auto`, `atlas` or empty	Your default provider (auto fallback unless you set one with `atlas use`)
`groq/llama-3.3-70b-versatile`	That provider and model, no fallback
`openrouter/openrouter/free`	OpenRouter’s free router
anything else	Auto fallback with each provider’s default model

The response headers X-Atlas-Provider and X-Atlas-Model say who answered. If a provider fails before answering (rate limit, outage, bad key), the next one is tried.

Example

curl http://127.0.0.1:4141/v1/chat/completions \
  -H "Authorization: Bearer $ATLAS_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"model":"auto","messages":[{"role":"user","content":"hi"}]}'

Most agents read these variables:

export OPENAI_BASE_URL=http://127.0.0.1:4141/v1
export OPENAI_API_KEY=atlas-...

atlas run <agent> --free and Tab in the agents picker start the proxy in the background and set the variables for you. See AI agents launcher.