Zero data retention

How it works

Zero data retention (ZDR) is an arrangement where the provider does not store your prompts and outputs after processing a request, beyond what is needed to serve it or to meet legal obligations. It is usually something you request or contract for, not the default.

Providers define it differently and often limit it to certain endpoints, models or features. Features that need stored data, such as file storage or server-side conversation state, may be excluded or work differently. This site shows whether a model is documented as eligible.

Check the list of eligible features, not just the model, before sending regulated data. ZDR covers retention, not where data is processed, so pair it with data residency when location matters.

Worked example

Illustrative: a health-records app has ZDR approved for the standard chat endpoint, then adds PDF uploads through a files endpoint that stores documents. Unless that endpoint is also covered, the uploads fall outside the arrangement.

Common mistakes

• 1Assuming ZDR covers every endpoint and feature, including file storage and batch jobs.
• 2Confusing 'not used for training' with 'not retained'; they are different commitments.
• 3Forgetting that logging and observability tools in your own stack may still store the data.