How Saywind turns speech into text on your Mac

1. You hold the key

   Saywind listens only while you hold your dictation key, or, after a double-tap for hands-free, until you tap it again. Between dictations a built-in, wired or USB microphone is kept ready but muted, so your first word isn’t lost.

2. Parakeet writes it down

   Parakeet Ultra, a speech model built on NVIDIA’s Parakeet, turns your speech into text on your Mac’s own chips. It knows 25 languages and works out which one you’re speaking.

3. Saywind tidies it

   Fillers and stumbles come out, and your Snippets, your Dictionary spellings and, if you choose it, British spelling go in. On macOS 26 or later with Apple Intelligence, Smart clean-up can also ask Apple’s on-device model where you changed your mind. The model never writes your text: it points at words to take out, and what’s pasted is rebuilt from what you said.

4. It lands at your cursor

   Saywind pastes into the app in front, then puts your clipboard back. It never presses Return. In a password field, the text waits on the clipboard instead.

5. It learns from your fixes

   In most apps, Saywind then watches the text box it pasted into, for up to a minute. If you fix a name it got wrong there, it remembers the spelling. You can switch this off in Settings › Learning.

6. Your history stays on your Mac

   History keeps your most recent 2,000 dictations on your Mac, and Insights is worked out from them. Saywind doesn’t keep recordings of your voice.

What goes online

Nothing you say. The privacy policy lists each time Saywind goes online, and what it sends.

Built on

Parakeet Ultra, by Moondream, is a post-trained version of Parakeet TDT 0.6B v3, by NVIDIA. FluidInference converted it to Core ML, and Saywind runs it with FluidAudio, also by FluidInference. Smart clean-up uses Apple’s on-device model. The acknowledgements list the credits and licences that ship with the app.