Potok · macOS · Built soloReady to download
Hold a key, talk, let go.
The text is already where your cursor was.
Why it exists
My voice notes were always longer than anything I typed, and I still typed the long stuff by hand. Potok closes that gap. It lives in the menu bar, the hotkey works system-wide, and the text goes straight into whatever field you were writing in.
The bilingual part is the whole reason it got built. If you write in two alphabets, the Fn key on a Mac is already spoken for by the layout switch, and every English dictation app puts its trigger right there. So the default here is Fn plus Option, and Fn on its own stays available for people who live in one language.
Text arrives through the clipboard and Cmd-V, because injecting it directly breaks in modern web fields. The recording strip sits at the bottom of the screen instead of up by the notch, where your eye rarely goes anyway.
What is inside
Your dictation history is encrypted. Transcripts of everything you have ever said no longer sit on disk as readable text. The key lives in the macOS keychain, is tied to that one machine, and never syncs anywhere.
A filter for endings you never said. The recognition model learned on YouTube and liked to append things like thanks for watching and see you next time. Nobody said them, so they get cut.
Models are picked in settings. Those values always existed. Changing them used to mean going through the terminal.
Cleanup comes in levels. The default is the light one. It fixes punctuation and leaves repetition alone, because repetition is how you actually talk and an app has no business editing that out.
Measurements
- Roughly two seconds pass between the end of a sentence and the text on screen. Measured on twenty of my own real dictations rather than a benchmark set.
- Cleanup costs more than the speech recognition does, 1.1 to 1.4 seconds against 0.84 to 0.90. I expected the opposite.
- Audio 22 times longer only takes twice as long to transcribe, 1.70 seconds against 0.84.
- I have never measured my own accuracy, so there is no percentage here. When I have my own number it will go in this spot.
Getting it
What you can download here is VerbaLite. It is a separate app, the small sibling of Potok: 201 kilobytes, the core job only, hold a key and talk, and it asks for your own key to the Groq speech service. The key is free and takes about two minutes to get at console.groq.com. Potok itself is being reworked right now, and its link goes in this spot once it is ready.
Needs macOS 13 or newer. Runs on both Apple Silicon and Intel.
The build is ad-hoc signed, so the first launch goes right-click on the app and Open. A double click will not get past Gatekeeper.
The app takes no money. I plan to charge for it later, and for now it runs on your key.