Your voice is a biometric. Voz treats it like one: transcribed on your own silicon, then discarded. Nothing to intercept, subpoena, or leak.
0
third-party requests on this page
online
this page works offline too — try airplane mode
0 KB
audio Voz has ever uploaded, all-time
The pipeline, in full
This is the entire journey your audio takes. There is no asterisk — this page and the app are built to the same rule.
$ voz audit --network
✓ model whisper base · on-device
✓ audio buffer in-memory only, discarded after use
✓ audio uploaded 0 bytes (all-time)
✓ text uploaded 0 bytes (all-time)
✓ telemetry none — there's nothing to send
✓ update check version number only · switch off in Settings
All clear. Your words stayed home.
The feature list
- No account
- No cloud processing
- No telemetry
- No subscription
- No training on your voice
- No waiting on a server
Everything above is a feature. A dictation app that needs none of it is faster, cheaper, and impossible to leak from — the absences are the product.
The one connection
Voz asks our update feed, about once a day, whether a newer version exists. That is the only time it uses the network, apart from one thing you have to ask for: switching on the optional local AI features downloads a model file once, and after that it runs on your Mac like everything else. We would rather name both than let you find them. The request carries a version number and nothing else — not your audio, not your text, not an account, not an identifier, because none of those exist to send. It is how a fix reaches you without you having to come back and check. You can switch it off in Settings ▸ Updates, and everything else keeps working exactly the same.
Even this website
The site you're reading practices what the app preaches. It loads no fonts, scripts, or pixels from anyone else's servers — the counter above is measured live, not typed in. It keeps working with your Wi-Fi off. And the mic demo on the home page reads your microphone's level locally for the waveform, records nothing, and deliberately avoids the browser speech-recognition API — because in most browsers, that ships audio to a server.