Hold a key and talk.

Potok sits in the macOS menu bar. You hold your hotkey, say the thing out loud, let go, and a second later the text is in the field where the cursor was blinking. An email, a chat, a note, a search box, a code editor. Not just the browser, any app you can type into.

Speech gets transcribed first, then the transcript goes through a language model that strips the noise and puts the punctuation back. What lands in the field is a sentence, not a stenogram.

I built it for myself, because talking is faster than typing. The first screen inside the app says I stopped typing. An exaggeration, but not a big one.

Speech cleanup. Real speech is full of filler, false starts and second runs at the same sentence. Potok takes that out, puts the punctuation back and hands over something readable. Four levels, from a raw transcript to a fully rebuilt sentence, and you can set a different one per app.

Your own dictionary. Names, brands and jargon the model is not allowed to correct into something that merely sounds similar. Edit the pasted text right after it lands and Potok notices the fix, then offers to remember the word.

Dictation history. Everything you dictate stays in the app and stays searchable. Final text and raw transcript sit next to each other, so you can see where cleanup went too far. History can be switched off, capped by age, or wiped completely.

Vibe speaking. A separate key for a different job. Ramble the idea out however it comes, and Potok hands back a structured prompt for Claude, ChatGPT or Cursor. The more you say, the better the prompt.

Six interface languages. English, Russian, Spanish, German, French and Brazilian Portuguese. Interface language and spoken language are separate things, dictation covers dozens of languages either way.

Audio goes to the cloud to get transcribed, there is no way around that today. It travels from your Mac straight to whichever service you put a key in for. Potok has no server of its own, so there is no middleman either: the recording never passes through me, or through anyone's infrastructure except the service you picked.

One more thing leaves the machine, context hints. If you grant screen recording, Potok attaches a shot of the active window to the cleanup request so it knows where you are dictating. That permission is optional and the hints have an off switch in settings.

Dictation history stays on your Mac, in the app's own folder, and is never uploaded. You can turn it off, cap how long it is kept, or erase all of it. Older months move into an encrypted archive instead of sitting on disk as readable text, and the key lives in the macOS keychain, where another program cannot take it silently, you get asked first. One honest caveat: the keychain travels in a Time Machine backup, so this is not a hard tie to a single machine.

No telemetry, no analytics inside. Potok reaches the network for two things: to process what you said, and to check whether a newer version exists.

macOS 13 Ventura or newer. The app lives in the menu bar and asks for two system permissions: the microphone, to hear you, and Accessibility, to put text into another app's window. macOS asks for both on first launch.

Your own Groq key. Speech-to-text runs on Groq, which needs a personal key. Signing up is free, no card required, and the first-run wizard walks you through it. The key stays on your Mac.

A warning on first launch. The app is signed with a self-signed certificate, Apple never checked it, so macOS will say it cannot verify the developer. On macOS 15 and newer: System Settings, Privacy and Security, Open Anyway near the bottom. On macOS 13 and 14 a right click on Potok in Applications and then Open is enough. You do it once, after that it launches like anything else.