Your voice never leaves your Mac — literally

Every dictation product makes some version of the same promise: we respect your privacy. Almost all of them mean: your audio goes to our servers, gets transcribed there, and we promise to treat it nicely.

We wanted to make a different kind of claim — not a policy, a property. So Privt Voice runs the entire speech model on your Mac. When you hold Right Shift and talk, audio goes from your microphone into a transcription model executing on the Neural Engine, and text lands in whatever app you're using. The audio is discarded from memory. There's no server in that sentence because there is no server.

This became possible recently. Modern open speech models brought fast, accurate recognition down to a size that runs comfortably on Apple silicon, and CoreML executes them on the Neural Engine faster than a round-trip to any cloud API. On-device stopped being the compromise option — it's simply better, and it happens to be private.

Meetings too

The same engine transcribes calls: your microphone on one lane, the call app's audio on the other, labelled Me and Them. Capture is scoped to the meeting app alone — your music never enters a transcript — and the finished note is sealed into your encrypted vault straight from memory, never written to disk in the clear.

Free forever isn't a promotion; it's a consequence. Software with no servers has no marginal cost, so the local product costs nothing, always.