Every endpoint is served from one gateway and expects a Bearer token:
The base URL carries the version. Resource paths are version-agnostic: /datasets, /sql/statements, /chat/completions.

​
Authentication

Send your API key, or an OAuth2 access token, as Authorization: Bearer <token>. See Authentication for how to obtain one.

​
Conventions

  • Ids are typed strings: res.project.…, res.dataset.…, res.connection.….
  • Listings return an envelope: {"object": "list", "data": [...]}.
  • Single objects are returned as-is.
  • Errors carry {"detail": ...} with the status codes listed in Errors.
  • Inference endpoints (/models, /chat/completions, /embeddings, /rerank) follow the OpenAI conventions, so any OpenAI-compatible client works with base_url set to the gateway.

​
Scope of this reference

The pages that follow document the endpoints exposed through the public gateway, the ones the Python library uses. The catalog has more (agents, flows, ontologies, orchestration, schedules, lineage); they are reachable from the Atlas interface and are being opened on the gateway progressively.

Inference

Models, chat completions, embeddings, rerank.

Projects

Projects and what they hold.

Datasets and files

Datasets, their files and data branches.

SQL and connections

Statements on datasets or connections, and the files behind a connection.