Zero data retention

Zero data retention is a policy under which a service does not keep the content you send it. The input is processed, a result is returned, and the content is discarded. It is not written to a database, not kept in logs, and not used to train models.

It answers a different question from data residency. Residency is about where data is processed; retention is about how long it exists afterwards. Sensitive workloads usually need both: processing in the right place, and nothing left behind once the job is done.

What it does and does not cover

Zero retention applies to the content of a request. Most services still keep some metadata needed to run the business, such as the time of a request, whether it succeeded, and how much to bill. A precise zero-retention promise says exactly which metadata is kept and for how long.

It also reduces risk rather than removing it: content still exists in memory while it is being processed, and it is still transmitted over the network, which is why encryption in transit matters too.

Zero data retention in Laya Studio

Laya Studio processes the text and questions you send in memory and discards them when the answer is returned. They are never written to a database or log, and never used to train anything. For billing and debugging, only request metadata (time, status, number of questions, latency) is kept, for 30 days, with daily totals kept for invoicing. All traffic uses TLS.

Because Laya is a decision model that returns probabilities rather than generated text, there is also no model-written text to store: the response is a set of labels and numbers. See Swiss data residency for the full list of commitments and patient message triage for a sensitive-data example.