Do you store my inputs and outputs?

Baseten operates a zero data retention (ZDR) posture for synchronous inference: model inputs and outputs are not stored by default, and they are not used to train models. Model weights are loaded at deployment time and are not persisted by default.

The precise edges:

• Async inference temporarily stores request inputs until the request is processed, then deletes them. Async outputs are not stored.

• Weight caching for faster cold starts is optional, and cached weights can be erased on request.

For the full policy set, see the Baseten Trust Center.

Docs: Secure model inference