Flowy

Minimal local speech-to-text for macOS.
Hold a global hotkey → speak → release → text appears at the cursor.
Fully offline. No telemetry. Optional LLM cleanup via local Ollama.

---

Features

Feature	Detail
Local only	Transcription uses macOS Speech Recognition — no API keys, no cloud
Push-to-talk	Hold a configurable global hotkey while speaking; text is injected on release
Custom dictionary	Word-substitution map applied after transcription
Ollama enhancement	Optional cleanup pass through a local LLM for punctuation & grammar
Audio device picker	Choose which microphone to record from
Output modes	Type into focused window, copy to clipboard, or both
History	Browse and copy your last 20 transcriptions
Autostart	Optional launch at login

---

Requirements

• macOS 13 Ventura or later (Apple Silicon)
• Xcode Command Line Tools

xcode-select --install

Optional — Ollama for AI enhancement:

brew install ollama
ollama pull llama3.2:3b
ollama serve

---

Build

# Build and copy Flowy.app to /Applications
make build

# Build, then relaunch the running app
make relaunch

# Package a distributable DMG
make dmg

---

macOS permissions

On first use the OS will prompt for:

1. Speech Recognition — required for transcription.
2. Microphone — required for recording.
3. Accessibility — required for keystroke injection.
   System Settings → Privacy & Security → Accessibility → enable Flowy.

If Accessibility is not granted, Flowy falls back to copying text to the clipboard.

---

Configuration

Config is written automatically on first launch.

Path
macOS	~/Library/Application Support/flowy/config.json

{
  "hotkey": "CmdOrCtrl+Shift+Space",
  "autostart": false,
  "dictionary": { "gonna": "going to", "wanna": "want to" },
  "inputDevice": null,
  "outputMode": "type",
  "maxRecordingSecs": 60,
  "historySize": 20,
  "ollamaEnabled": false,
  "ollamaEndpoint": "http://localhost:11434",
  "ollamaModel": "llama3.2:3b",
  "ollamaPrompt": "You are a transcription cleaner..."
}

---

Architecture

Pure Swift / AppKit + SwiftUI macOS app. No Electron, no Tauri, no web layer.

Global hotkey ──► AppModel (SwiftUI @Observable)
                      │
               AVAudioEngine recording
                      │
               SFSpeechRecognizer (on-device)
                      │
               custom dictionary substitution
                      │
               optional Ollama HTTP pass
                      │
               CGEvent keystroke injection  ──► focused app
               (fallback: NSPasteboard)

---

License

MIT