Beta
Local Dictation for macOS

Speak. Words appear. Nothing leaves your Mac.

Hold a hotkey, talk, release — on-device Whisper (Apple MLX) types straight into whatever app you're in. No cloud, no accounts, no audio uploads. Ever.

100% on-device ASR 0 network calls in the dictation path MIT licensed GitHub

Built for Indian Speech

"pandrah lakh rupaye""₹15,00,000"
"do crore tees lakh""₹2,30,00,000"
"kal meeting hai danda""कल meeting है।"

Numbers roll over lakh/crore correctly, currency becomes ₹, and the danda (।) punctuates Hindi — all locally.

Indian names, tech jargon, and Hinglish handled by a local post-processing layer — no cloud correction service.

How It Works

01

Hold the hotkey

02

Speak naturally

03

Release — text lands in the focused app

Works in any macOS app that accepts typing: mail, docs, chat, code.

Get It

Beta — build from source in about 5 minutes.

git clone https://github.com/kalpitt/VivoType cd VivoType && ./clients/mac/build_app.sh

Requires macOS with Python 3.11+ · Apple Silicon recommended (MLX-accelerated). First launch downloads the model (~466 MB) once, locally.

Honest Limits

macOS-only today (the core engine is deliberately platform-independent — a Windows path exists but isn't built). English + Hinglish post-processing is the focus; other languages work via Whisper's multilingual models. Dictation, not meeting transcription.

For Engineers

169 green tests. The Hindi-English cleanup layer shipped as a ten-commit validated autonomy run — every commit individually tested and logged.

169/169 tests CI 10-commit autonomy window — merge report