What privacy-first voice dictation actually means

“We take your privacy seriously” is a sentence, not an architecture. If a dictation tool streams your voice to a server, the real questions are: what may be stored, who decides, and can the answer be enforced rather than promised?

In Vocalis, Privacy Mode and Cloud Sync are independent controls. Privacy Mode governs whether raw voice data may persist anywhere; Cloud Sync governs whether your dictionary and settings follow you across devices. You can sync your vocabulary while guaranteeing no audio is ever stored — they are different promises, so they are different switches.

Every persistence decision runs through a single policy engine that classifies data (audio, transcripts, context, metadata) and answers one question: may this be stored, given this user's and this organization's settings? No code path writes voice data without passing that check. Screenshots captured for context are stripped at ingest and never stored at all.

For teams, an org admin can enforce privacy settings across every seat. Compliance stops being a training problem and becomes a configuration — the tool refuses to store what policy forbids, regardless of what an individual user clicks.