On-device dictation for macOS

Talk. Sonari types it.

Press a shortcut and talk. Sonari pastes clean, punctuated text into whatever app you're in, and it all runs on your Mac. Your voice never leaves it.

Coming soon See how it works

Apple Silicon · macOS 14+ · One-time purchase, never a subscription.

Transforms settings with Slack, Messages, and Telegram mapped to the Message transform, and Mail to Email

Match transforms to apps

The right style for wherever you type.

Dictation should sound different depending on where it lands. Attach an app to a transform and Sonari applies it automatically, going off whichever app was in front when you started talking. Slack, iMessage, and Telegram can share one casual style. Mail turns into a proper email. A terminal can stay untouched. Set it once and stop babysitting a mode switch.

The History window listing past dictations with their model, duration, and tags

Nothing is thrown away

Go back to anything you've said.

Every dictation is saved on your Mac, audio and all. Search your transcripts, play back the original, or copy the text again. Downloaded a sharper model since? Re-transcribe an old recording without saying a word. It all stays on your machine, and you can delete any of it whenever you want.

The Model tab with Parakeet v2 active and the Whisper models listed

Your engine, your download

Whisper or Parakeet. Your call.

Sonari runs two on-device engines. Parakeet is fast and genuinely accurate for English; WhisperKit is lighter if you want a smaller download. Pick one and it caches locally, so switching between the ones you've grabbed is instant. Every model runs on your Mac, which means dictation works on a plane, off Wi-Fi, anywhere.

The custom dictionary correcting clod code to Claude Code and son are to Sonari

For the words it keeps missing

Teach it your names and jargon.

Names, acronyms, the product you say fifty times a day: the words a speech model mangles the same way every time. Add them to your dictionary once and Sonari fixes them on every transcript, before any AI even runs. It's deterministic and instant, and it works whether or not Apple Intelligence is available.

Nothing leaves your machine

Your voice is yours. It never touches a server.

Recording, transcription, and the AI rewrites all happen on your Mac. There's no cloud to opt out of. No account, no API key, nothing to paste in. And if Apple Intelligence isn't available, Sonari inserts your transcript unchanged. It never falls back to a server.

Recordings + transcripts~/Library/Application Support/Sonari/
Speech modelsDownloaded once, cached locally
Network useOnly to download a model you chose

How it works

Press, talk, and it's typed.

Press your shortcut

The default is ⌘⇧Space; change it to anything. Tap to start and stop, or hold to talk and release when you're done.

Speak

The notch expands into a live waveform while it listens. Recording and transcription both happen on your Mac.

It's typed

Clean, punctuated text is pasted into the app you were already in. No window to switch to, nothing to copy.

Dictate anywhere on your Mac.

It runs on your Mac, and nothing leaves it. One-time purchase, never a subscription.

Coming soon

Apple Silicon · macOS 14+