A tiny Rust menu bar app for macOS. Hold Fn, speak, paste. Microphone is only active while you hold the key — you'll see the yellow mic indicator in macOS as confirmation.

If you find FnKey useful, please star the repo — it helps others discover it.

1. Download from Releases:

   • Apple Silicon (M1/M2/M3): FnKey-arm64.zip
   • Intel: FnKey-x64.zip

2. Unzip and move to Applications:

   unzip FnKey-arm64.zip mv FnKey.app /Applications/

3. Set your API key(s):

   mkdir -p ~/.config/fnkey # Deepgram (streaming, recommended) — $200 free credit echo 'your-deepgram-key' > ~/.config/fnkey/deepgram_key # Groq (batch fallback if no Deepgram key) echo 'your-groq-key' > ~/.config/fnkey/api_key

   Get keys at console.deepgram.com and console.groq.com

4. Launch:

   open /Applications/FnKey.app

5. Grant permissions in System Settings → Privacy & Security:

   Permission	Purpose	How to Grant
   Input Monitoring	Detect Fn key press	Add FnKey.app via + button
   Microphone	Record voice	Prompted on first use, or add manually
   Accessibility	Auto-paste text	Add FnKey.app via + button

   Build script preserves permissions across rebuilds when using ./build-app.sh.

• Hold Fn and speak → transcription
• Release to transcribe and paste
• Click menu bar icon (○) to toggle Press Return after paste (sends Return key after pasting)
• Click menu bar icon (○) → Edit Keywords… to add custom vocabulary (opens in default text editor)
• Click menu bar icon (○) → Quit to exit

The icon changes: ○ (idle) → ● (recording)

If both keys are configured, Deepgram streaming is preferred.

./build-app.sh

This builds and installs directly to /Applications/FnKey.app, preserving permissions across rebuilds.

Note: If cargo isn't found, run with login shell: /bin/bash -l -c './build-app.sh'

• Real-time streaming - Audio streams to Deepgram as you speak (no waiting)
• Deepgram Nova-3 - Latest model with smart formatting and punctuation
• Groq fallback - Whisper large-v3 batch mode if Deepgram unavailable
• Audio enhancement - DC offset removal, high-pass filter, peak normalization (Groq mode)
• Auto sample rate - Uses device's native sample rate, resamples to 16kHz for Deepgram
• Non-blocking - WebSocket connects in background, never freezes the app
• Auto-return mode - Optional Return keypress after paste (toggle in menu bar)

Add words the transcription engine often gets wrong (proper nouns, technical terms):

# Edit via menu bar → "Edit Keywords…", or directly: cat > ~/.config/fnkey/keywords << 'EOF' # One term per line Anthropic Claude cron job Kubernetes EOF

Keywords are sent as keyterm to Deepgram (Nova-3) and as prompt hints to Groq/Whisper. Reloaded each recording session.

• No-speech detection - Use verbose_json response format and check no_speech_prob to skip silent recordings
• Backend toggle - Menu bar option to switch between Deepgram and Groq

• Falls back to Option key if Fn not detected after 5s

Slight recording delay: There's a brief moment when you start speaking before audio capture begins. This is a deliberate tradeoff — eliminating this delay would require the microphone to be always active, showing the yellow indicator constantly. The current design prioritizes privacy: the microphone only activates when you press the Fn key.