Agentastic turns speech into text anywhere on your Mac. There are two voice features: voice input into the launcher, and dictation that types into whatever app you're in.
Voice input (push-to-talk)
Speak your prompt instead of typing it:
- Hold
Space(default) for about half a second to start recording, release to transcribe into the launcher. - The transcription appears in the prompt field, where you can edit it or press Return to send.
Dictation (type anywhere)
Dictation drops formatted text straight into your current app (email, notes, chat, code):
- Hold the
Fn(Function) key for about half a second to start dictating, release to insert the text. - Under the hood this runs the dictation app (
/app:dictation), which is tuned for clean, formatted speech-to-text.
Both shortcuts are configurable in Settings → Preferences, and dictation has its own page in Settings → Apps.
Speech backends
Agentastic can transcribe with different engines. Choose one in settings:
| Backend | Where it runs | Notes |
|---|---|---|
| Apple (default) | On-device | Uses macOS's built-in speech recognition. No setup. |
| Local Whisper | On-device | Runs a local Whisper model for higher accuracy and full privacy. Falls back to Apple if the model isn't installed. |
| NVIDIA Parakeet | On-device | A fast local model. Falls back to Apple if model files are missing. |
On-device backends keep your audio on your Mac: nothing is sent to the cloud for transcription.
Tips
- Speak naturally; the dictation app cleans up filler and adds punctuation.
- Use voice input when you want to talk to the agent, and dictation when you want to talk through it into another app.
- Pair dictation with a model default in Models & Providers (the dictation role) to control which model formats your speech.
Related
- Launcher Modes: all the ways to drive the launcher
- Keyboard Shortcuts: change the voice and dictation keys