Questions,
answered
The short version: it types what you say, and your voice is nobody's business.
No. Recording, transcription, and formatting all happen on-device, and the audio buffer is discarded the moment it becomes text. Flip on Airplane Mode and keep dictating — Voz can't tell the difference.
Once a day it asks our update feed whether a newer version exists, and shows a small notice in the corner if there is one. Unless you switch on the optional local AI features — which download a model file once, and then run it on your Mac — that is the only network request Voz makes. It sends a version number — not your audio, not your text, not an account or an identifier, because Voz doesn't have any of those. You can switch it off in Settings ▸ Updates, and dictation is unaffected either way. More on the privacy page.
There isn't much of one anymore. On-device Whisper-class models sit within a hair of cloud accuracy, and with no round trip, Voz is usually faster. The trade: the model lives on your disk — about 141 MB, inside a 149 MB download.
No — and you can check. It uses Web Audio to read the microphone's level for the waveform and nothing else: no recording, no transcription, no upload. It deliberately avoids the browser's speech-recognition API, because in most browsers that does ship your audio to a server. Open the network tab and watch nothing happen.
Any Apple Silicon Mac on macOS 13 Ventura or later is the sweet spot. The same universal build runs on Intel and works fine — transcription just takes a beat longer.
Sixteen languages, with automatic detection. Auto-detect listens to the opening audio, so on a noisy room or a very short clip it helps to set the language explicitly. Custom vocabulary works in every one of them.
$2.99, once. No subscription, no account, no trial to forget to cancel — you pay a few dollars and the app is yours, updates included. Details on the pricing page.