Flowa

Voice dictation for macOS. Press fn, speak, press fn again — your words appear wherever your cursor is.

Runs entirely on your Mac after setup. No subscription, no cloud transcription, no audio leaving your machine.

How it works

1. Press fn — recording starts, a small floating bar appears
2. Speak for as long as you want (any language, mid-sentence switching is fine)
3. Press fn again — Flowa transcribes and pastes the text into whatever app you were using

That's it.

Features

• Local transcription — after first setup, speech runs on-device
• Any language — 99 languages supported. Switch languages mid-session; Flowa picks it up automatically
• Works everywhere — dictates into any app: notes, email, chat, code editors, browsers
• Menu bar icon — stays running in the background, always ready
• Recent history — last 100 dictations saved locally, copyable from the Home screen
• Launch at login — optional, toggled from the app

Requirements

• macOS 14 or later
• Apple Silicon (M1 or later) recommended — runs on Intel but first-time preparation takes longer
• Internet once on first launch if you built from this repo (to fetch the ~1.5 GB speech engine)

Build from this repo

The speech engine is not committed to git (it is ~1.5 GB). Anyone can still use the repo:

git clone https://github.com/maxkongerskov/Flowa.git
cd Flowa
open Flowa.xcodeproj

1. In Xcode, select your Team under Signing & Capabilities
2. Build and run (⌘R)
3. On first launch, grant Microphone, Input Monitoring, and Accessibility
4. System Settings → Keyboard → set "Press 🌐 key to" → Do Nothing
5. Flowa will download the speech engine once (~1.5 GB from the WhisperKit CoreML hub), then prepare it for your Mac (~10 minutes). Keep the app open.

After that, Flowa works offline.

Optional: ship a fully offline app package

If you want the engine inside the .app (no first-run download for end users):

1. Place a complete copy at
   Flowa/Models/openai_whisper-large-v3-v20240930_turbo/
   (same layout as on Hugging Face)
2. Archive / notarize as usual — the model is copied into the app bundle
3. First launch only specializes CoreML for that Mac (still ~10 minutes, offline)

scripts/notarize.sh verifies the bundled model is present before export.

Dependencies

• WhisperKit — Swift package (Xcode package manager)
• Speech engine weights from argmaxinc/whisperkit-coreml (openai_whisper-large-v3-v20240930_turbo)

License

MIT — see LICENSE