The voice layer for your Mac
Or let it act.
Hold one key and talk. Your words land in whatever you're already in — Mail, Slack, your editor — transcribed on your Mac, so your voice never leaves it.
Dictate
Most people type around 40 words a minute and speak closer to 150. Dictation just hands those words back to you.
Thanks for the notes — I'll turn
Thanks for the notes — I'll turn them around by Friday, and I'll loop in Priya on the pricing section before it goes out.
Typical rates for English. In practice you pause, think and fix things — so call it ~3× on a good day.
One key · two modes
The same push-to-talk key drives both. Dictation works the moment you install it; Command is there when you want it.
Talk into any text field — email, chat, code, docs. agentaud.io types it the moment you stop, so the words land wherever your cursor already is.
Turn on Command mode and it stops typing and starts doing. Say what you want in plain English; a frontier model — running on your own subscription — works out the steps, runs them, and speaks the answer back.
rm, sudo, git push) wait for your OK.Command
No syntax to learn. It works out the command, runs it, and tells you what happened in a sentence — and you see every step it took.
osascript -e 'tell app "System Events" to count
(processes whose background only is false)'
Then it speaks the answer back, so you never look away from what you were doing.
Private & yours
Speech is transcribed locally, by a model that lives on your Mac. In Command mode only the text of your request reaches the reasoning model — the audio never leaves.
The app is free. Command mode uses the AI subscription you already have — no API key to paste, no second bill. Dictation needs no account at all.
Command mode is Guarded by default — anything risky stops and waits for your OK. Loosen it to Autopilot or tighten it to Strict whenever you like.
Download
One .dmg, self-contained. No Homebrew, no setup — drag it to Applications and hold the key.