What runs where
| What | Where it happens | Needs the internet |
|---|---|---|
| Recording your voice | On your Mac | No |
| Turning speech into text | On your Mac | No |
| Your history and your vocabulary | Stored on your Mac | No |
| Downloading the speech model | Once, from models.ivolna.com or Hugging Face | Yes, the first time |
| Checking for updates | About once a day, updates.ivolna.com | Yes, but Volna works without it |
The two network requests carry no audio, no text, no history and no vocabulary. The details are on the privacy page.
The speech model
Volna runs Whisper large-v3-turbo, an open speech model, through whisper.cpp, right on your Mac's Apple Silicon. The app downloads the model once and checks every file against its known size and fingerprint before using it. After that, dictation works with Wi-Fi off.
Why offline matters
- Privacy. What you dictate can be a medical note, a contract or a password reset email. With Volna it never reaches a server.
- Work rules. Many companies do not allow sending work text to cloud services. Speech recognized on the Mac does not leave it.
- Anywhere. On a plane, on a train, in a basement office: no connection, same dictation.
- No meter running. Nothing is billed per minute or per word.
What you get
- One key in any app: the right ⌘, or ⌥Space.
- Text about a second after you stop speaking.
- English, Brazilian Portuguese, Spanish and nearly a hundred more languages; English terms inside a sentence are fine.
- Your own vocabulary, and corrections Volna can remember.
- A history of your dictations, kept on your Mac.