Friday

A pi package that adds a dedicated communications side panel. All conversation is routed to a separate tmux pane with typewriter effect, keeping your main window clean for code, diffs, and command output. Optionally enable text-to-speech and hands-free voice input via wake word detection.

Installation

pi install npm:pi-friday

Or from git:

pi install git:github.com/dantetekanem/friday

To try without installing:

pi -e npm:pi-friday

Requirements

Required (the extension will not load without these):

• tmux — pi already requires this
• perl — pre-installed on macOS

Optional — Voice output (TTS):

• piper-tts — pip3 install piper-tts
• sox — brew install sox (provides the play command)
• A Piper voice model in ~/.local/share/piper-voices/ (see Voices below)

Without piper-tts and sox, voice output is disabled. The panel still works for text.

Optional — Voice input (wake word + transcription):

• openwakeword — pip3 install openwakeword
• faster-whisper — pip3 install faster-whisper
• pyaudio — pip3 install pyaudio (requires brew install portaudio)
• sounddevice + numpy — pip3 install sounddevice numpy

Without these, the wake word listener (/friday listen, Alt+L) is unavailable.

Shortcuts

Shortcut	Action
Alt+M	Toggle voice on/off
Alt+L	Toggle wake word listener on/off

Commands

Command	Action
/friday	Toggle the extension on/off
/friday voice	Toggle voice output
/friday listen	Toggle wake word listener
/friday settings	Show current configuration

The status bar shows active modes: FRIDAY, VOICE, DAEMON ON.

Wake Word

Say the configured wake word (default: "hey friday") to activate hands-free voice input. After detection, Friday records your speech, transcribes it locally with faster-whisper, and sends it as a message to pi.

Friday looks for custom wake word models in ~/.pi/agent/friday/. To set up the default "hey friday" wake word:

1. Visit openwakeword.com/library (free, requires sign-in)
2. Search for "hey friday"
3. Download the .onnx file to ~/.pi/agent/friday/
4. Set wakeWord.model in settings.json to the filename without .onnx

You can use any custom wake word — just download its .onnx model to the same directory.

Built-in models (no download needed): alexa, hey_mycroft, hey_jarvis, hey_rhasspy, timer, weather.

Voices

Friday uses Piper for text-to-speech. The default voice is en_GB-jenny_dioco-medium (British female).

Install the default voice:

mkdir -p ~/.local/share/piper-voices
cd ~/.local/share/piper-voices
curl -sL -o en_GB-jenny_dioco-medium.onnx \
  "https://huggingface.co/rhasspy/piper-voices/resolve/v1.0.0/en/en_GB/jenny_dioco/medium/en_GB-jenny_dioco-medium.onnx"
curl -sL -o en_GB-jenny_dioco-medium.onnx.json \
  "https://huggingface.co/rhasspy/piper-voices/resolve/v1.0.0/en/en_GB/jenny_dioco/medium/en_GB-jenny_dioco-medium.onnx.json"

Use a different voice:

Browse voices at rhasspy/piper/voices. Download the .onnx and .onnx.json files to ~/.local/share/piper-voices/, then set voice.model in settings.json to the model name (without .onnx).

Settings

Edit settings.json in this directory:

{
  "name": "Friday",
  "voice": {
    "enabled": false,
    "model": "en_GB-jenny_dioco-medium",
    "speed": 0.9
  },
  "wakeWord": {
    "enabled": false,
    "model": "hey_friday",
    "threshold": 0.3,
    "whisperModel": "tiny.en"
  },
  "typewriter": {
    "enabled": true
  },
  "panelWidth": 30
}

Setting	Description
name	Display name in status bar
voice.model	Piper voice model name (filename in ~/.local/share/piper-voices/ without .onnx)
voice.speed	Speech speed multiplier (0.9 = slightly slower)
wakeWord.model	openwakeword model name or custom .onnx filename without extension
wakeWord.threshold	Detection confidence (0.0-1.0, lower = more sensitive)
wakeWord.whisperModel	faster-whisper model size: tiny.en, base.en, small.en, medium.en
typewriter.enabled	Typewriter text effect in panel
panelWidth	Panel width as percentage of terminal

License

MIT