Beta

The voice layer for your Mac

Give your agents
a voice.

Speak to do, or speak to write.

See both modes
Apple silicon· ~135 MB· Runs on your own AI subscription

One key · two modes

Speak to do, or speak to write.

The same push-to-talk key drives two things. Flip between them in a click.

Command

Speak, and a frontier model acts.

Say what you want in plain English. agentaud.io hands it to a frontier model — running on your own subscription — which plans the steps, runs the real commands on your Mac, and speaks the result back.

  • Opens apps, runs shell & AppleScript, sets reminders — real actions, not answers.
  • Watch it work: reasoning, each command, and its output stream live.
  • Risky steps (rm, sudo, git push) wait for your OK.
> set a reminder to call mom at 6pm Reminders → "Call mom" · today 6:00 PM Done — I'll nudge you at six.
Dictate

Speak, and it types.

Flip to Dictate and talk into any text field — email, chat, code, docs. agentaud.io types it for you the moment you stop, key by key, so it lands wherever your cursor is.

  • Works in every app — no plugins, no copy-paste.
  • Types verbatim via real keystrokes — never touches your clipboard.
  • On-device transcription — your voice never leaves the Mac.
Compose · Mail Thanks for the notes — I'll turn them around by Friday.

Dictate

You speak faster
than you type.

Most people type around 40 words a minute and speak closer to 150. Dictation just gives you those words back.

~3×
Typing
~40 wpm
Speaking
~150 wpm

Typical rates for English · your mileage varies with what you're writing.

Command · try saying

Talk to it like a person.

No syntax to learn. It works out the commands, picks the right app, and reports back in a sentence.

> What's my battery percentage and Wi-Fi network?
> Open Safari and go to Hacker News.
> How many files are on my Desktop?
> Set a reminder to call mom at 6pm.

Private & yours

On your Mac. On your terms.

🔒

Your voice stays on-device

agentaud.io transcribes locally. In Command mode, only the reasoning step talks to the model — your audio never leaves.

Runs on your subscription

The app is free. Command mode uses the AI subscription you already have — no API key, no second bill.

⚠︎

A leash you control

Choose Guarded, Strict, or Autopilot. Anything risky pauses for a spoken or clicked OK before it runs.

Download

Give your agents a voice.

One .dmg, self-contained. No Homebrew, no setup — drag it to Applications and hold the key.

  • Apple silicon Mac (M1 or later)
  • ~135 MB download · self-contained
  • Command mode signs in to your AI provider
  • Speech model fetched once on first run

# First launch (ad-hoc signed, so Gatekeeper asks once)
Right-click the app → Open, or run:
xattr -dr com.apple.quarantine /Applications/agentaud.app