AI builder glossary
Zero data retention (ZDR)
A provider arrangement in which your API inputs and outputs aren't stored after processing, apart from narrow exceptions such as abuse prevention.
By default, model providers keep API data for a period under their published policies, for purposes such as abuse monitoring. Zero data retention is a stricter arrangement: the provider doesn't store prompts and responses once the request completes, apart from narrow exceptions such as legal obligations and certain safety signals. Customers in healthcare, finance, and legal work often require it in their contracts.
ZDR is usually not a self-service switch. Expect an enterprise agreement and an approval process. It also excludes features that store data by design, such as uploaded files kept until you delete them and other endpoints a provider lists as ineligible. If your feature depends on one of those, ZDR may not cover it.
The bigger misconception is that ZDR makes your product zero-retention. It covers only the provider. Your own logs, traces, caches, vector indexes, eval datasets, and third-party observability tools still hold user data, and that's where most retention questions end up. ZDR also means the provider can't look at a failed request to help you debug, so your own logging matters more. Offer it to customers who need it, and get your own data map right first.
A healthcare customer evaluating Deskmate requires ZDR. The team confirms that the endpoints it uses are covered by its provider agreement, disables file attachments for AI features in that workspace, and shortens its own trace retention for that customer to match.
Learn more in the course
Related terms
Go from definitions to doing
Learn to build AI-powered apps in 132 practical lessons.