OpenAI-compatible API

Many models.
One orbit.

One endpoint and one API format for every model. Switch models by changing a single string, and turn thinking on or off for each request.

Models
--
With thinking
--
Endpoint
/v1

01 / Drop-in

Works with any OpenAI SDK. Just change base_url.

02 / Streaming

Server-sent events, tool calls, and reasoning output in reasoning_content.

03 / Thinking control

Add :think, :nothink or :high to a model name, or send reasoning_effort.

04 / One format

Every model returns the same response format, so you never need provider-specific code.

# Quick start

More examples are in the docs. To see every model, go to models.