Data Usage and Retention

Our data handling policy

We do not store customer prompts, inputs, or model outputs in our systems by default. All inference requests are processed ephemerally and transmitted directly to third party providers. We retain only limited operational metadata (such as request IDs, token counts, timestamps, and response codes) necessary for billing, reliability, and abuse prevention. Customer data is stored only if an individual customer explicitly requests temporary retention for debugging or support purposes. In such cases, retention is time-bounded and strictly limited to resolving the issue.

The Grid uses third-party providers to run models and return inference outputs. Every provider is configured with Zero Data Retention (ZDR), or similar “do not store prompts or outputs” control.

In some cases, this is their default setting, in others we have enabled it. In general, these vendors process prompts and outputs ephemerally—only for the duration required to fulfill the request—while logging limited usage metadata (e.g., token counts and request IDs) for billing, operational, and security purposes.

Below you can find a list of our providers, their policies, and any configurations we have enabled to ensure your privacy.

Provider

Data usage / retention policy

Notes / notable exceptions

Baseten _Docs_|_ToS_|_{Privacy Policy}