Protocols & Infrastructure

What Is Zero Data Retention (ZDR)?

A setting under which an AI service answers a request and then discards it: the question and answer are not written to history, logs or traces.

Most AI assistants keep a record of every conversation: the question, the answer, the sources, often a debug trace. That record helps with analytics, but it is also personal data that must be secured and deleted on request. Zero data retention removes it: the answer streams back and nothing about the exchange is kept.

The term is used at two levels. Model providers offer zero-retention terms for their APIs, and the details differ by provider. Applications built on those models need their own setting, because their database, logs and traces are a separate copy. A team that wants nothing kept needs both.

The trade-off is insight. With nothing stored there is no history to review, no rating to attach, and no record to mine for knowledge gaps. That is why it is usually a setting, not the default.

In Kelu

How Kelu uses this

Zero retention is a per-workspace setting in Kelu, on every plan. Answers still stream with citations, but no conversation, message or trace is stored, nothing goes into the answer cache, and no rating can be attached. Questions still count toward the monthly quota, as a daily count with no text. Indexed content is unaffected. Workspaces that want some history can keep it for 30, 90 or 365 days instead.

From definition to deployment

Connect a knowledge source and see cited AI answers on your own content in minutes — free, no credit card required.

Free to start · No credit card required