Zero-retention mode streams your file directly into the AI worker, processes it, returns the result, and forgets the source — nothing is written to disk on our side, even temporarily.
Enable it per-workspace at Settings → Privacy → Zero retention. Once enabled, every AI run from that workspace honors the setting; you do not need to remember to flip it on each time.
There is a small latency cost (typically 100–300 ms per request) because we cannot pre-cache the parsed document tree. For most documents the difference is invisible.