Transcription happens locally on your machine. No account, no cloud, no subscription. Your voice never leaves it.
Hold fn and speak, in any app, at your cursor. A small pill shows it's listening.
Let go. On-device models transcribe and clean up in about a second: fillers dropped, punctuation restored, self-corrections applied.
Polished text lands at your cursor, pasted for you. “meet monday no wait tuesday” becomes “Meet Tuesday.”
Two local models do the work: Whisper writes down what you said; Gemma 4 edits it the way you would. No audio is uploaded, ever — you can pull the network cable and it keeps working.
Fully offline. No account, no API keys, no telemetry. Airplane-mode dictation.
Fast. About a second from release to paste, thanks to on-device Metal + speculative decoding.
Code-switching. Mix English and 中文 mid-sentence — Gemma follows your language list word for word.
Spoken formatting. Say “make this a bulleted list” and get one. Formatting commands are applied, not transcribed.
Your voice, learned. Optional on-device training log builds a dataset of your dictations for a future personal fine-tune.
The current version is yours forever. Updates are $10 a year if you want them, and the app keeps working either way.
An Apple Silicon Mac (M1 or later) on macOS 14+. The models download once (~5 GB) and then live on your machine.
No. Transcription and cleanup run locally. The only network calls are the one-time model download and your purchase.
Whisper covers ~99 languages, one per take. For mid-sentence mixing (e.g. English + 中文), the Gemma pipeline follows your language list word by word.
Email within 30 days for a refund. No questions, no forms.