Offline dictation · macOS
You talk.
The text is already there.
Hold the right Command key, say what you have to say, let go. The text lands where your cursor was: in your email, your editor, a search field. No window to switch to, nothing to paste.
The recording never
leaves this machine.
There is no server for it to reach. Speech recognition runs locally on whisper.cpp large-v3-turbo, the same file you download once at install. Viskly works the same on a plane with no signal as it does at your desk.
No cloud
Not one network request between the microphone and the text.
No account
Nothing to sign up for, nowhere to log in. Install it and talk.
The full model
large-v3-turbo, not a trimmed-down mobile build.
Readable source
Every line of it, under the GPL. You do not have to take the claim on trust.
Three moves
One key you were not using anyway.
- 01right ⌘
Hold
The indicator appears near the bottom of the screen. It never takes focus, so your cursor stays exactly where it was.
- 02
Speak
Say a long sentence, correct yourself halfway through. Escape throws the recording away before anything reaches the screen.
- 03
Release
About half a second later the text is in the field your cursor was in. Your clipboard goes back to whatever it held.
Latency
About half a second.
A typical utterance of two or three sentences becomes text 500–700 ms after you release the key. It all runs on your own processor, so you are not waiting on a network or on a queue somewhere else in the world.
- Recording
- as long as you speak
- Speech recognition
- 500–700 ms
- Text insertion
- ~60 ms
Anywhere you
can type.
Viskly is not a separate window you switch to and copy out of. It pastes into whatever app you are already in.
- Mail and chat
- Code editors
- Form fields
- Notes and documents
- Your browser's address bar
- Terminal, with a caveat
Stop transcribing
your own thoughts.
Installing is one file and three system permissions the app walks you through the first time you open it.
- System
- macOS, Apple Silicon or Intel
- Disk space
- 574 MB for the model
- From source
- JDK 25 and Maven