Skip to main content

OpenAI-compatible API

Mother exposes OpenAI-compatible inference endpoints, so standard OpenAI clients, SDKs, and tools can talk to the models on your machine without changes.

Base URL​

http://<your-mother-host>:8000/openai/v1

On the machine itself, use localhost:8000.

What works​

  • GET /models — list available local models
  • Chat completions (POST /chat/completions), including streaming
  • Standard OpenAI request/response shapes; unknown request fields are ignored, so forward-compatible clients keep working

Authentication​

Every non-public call needs a Bearer token. Use an API key created in Settings (see API keys) or a login token issued by the auth API — both work as Authorization: Bearer <token>.

Model selection​

  • Send a canonical model name (for example openai/gpt-oss-20b) and Mother resolves the best ready build on your machine.
  • The response headers report what actually ran (X-Mother-Resolved-Variant-Id, X-Mother-Resolved-Model-Name), which is useful for clients that want to display or log the exact model.

Cloud provider models are selected through Genesis sessions and the native /model endpoint; the OpenAI-compatible model list stays tied to models served on the machine.

Where to go next​

  • API keys — creating and managing access.
  • Model selection details — variants, pinning, and precedence.