Crix — AI Voice Assistant for Linux Desktop

Full documentation: See PROJECT.md for complete architecture, API details, and development guide.

---

Features

• 🎙️ Real-time voice interaction — Talk to Crix naturally
• ⌨️ Keyboard control — Type text, press shortcuts by voice
• 🖱️ Mouse control — Move, click, scroll anywhere
• 🪟 Window & Workspace — Switch workspaces, focus windows
• 🌐 Browser automation — Web tasks via natural language
• 🧠 Persistent memory — Remembers preferences across sessions
• 🔍 Live web search — Up-to-date information via Tavily
• 🔇 Noise cancellation — Intelligent audio filtering

---

Quick Start

1. Install System Dependencies

# Ubuntu/Debian
sudo apt install ydotool wl-clipboard grim xdotool tesseract-ocr portaudio19-dev

# Enable ydotool daemon
sudo systemctl enable --now ydotoold

2. Clone and Install

git clone https://github.com/Aerex0/Crix.git
cd Crix
uv sync

3. Configure

cp .env.example .env

Edit .env with your credentials:

LIVEKIT_URL=wss://your-project.livekit.cloud
LIVEKIT_API_KEY=your_api_key
LIVEKIT_API_SECRET=your_api_secret
TAVILY_API_KEY=your_tavily_key

# For browser automation (choose one):
OPENAI_API_KEY=sk-...     # Recommended
# ANTHROPIC_API_KEY=sk-ant-...
# GOOGLE_API_KEY=...
# OLLAMA_MODEL=llama3.1

4. Run

# With wake word (always-on) - say "hey livekit" to activate
nohup uv run src/agent.py dev > crix.log 2>&1 &

# Or without wake word (manual mode)
uv run src/agent.py console

---

Usage

Voice Commands

"Open Firefox"                    → Launch app
"Switch to workspace 3"           → Change desktop
"Type hello world"                → Type text
"Search Google for AI news"       → Browser search
"Check my Gmail"                  → Open Gmail via browser
"Add mouse to Amazon cart"        → Shopping automation
"What time is it?"                → Get current time

Memory Commands

"Remember that I prefer dark mode"    → Save preference
"What did I tell you about my name?"  → Search memory

---

Wake Word Setup

Crix listens for "hey livekit" by default. To use a custom wake word:

# Train custom model
uv run livekit-wakeword run configs/hey_crix.yaml

# Update .env
CRIX_WAKEWORD_MODEL_PATH=models/hey_crix.onnx

Run on Boot (Optional)

Create ~/.config/systemd/user/crix.service:

[Unit]
Description=Crix Voice Assistant

[Service]
ExecStart=/home/user/.local/bin/uv run /path/to/Crix/src/agent.py dev
WorkingDirectory=/path/to/Crix
Restart=on-failure

[Install]
WantedBy=default.target

Enable:

systemctl --user daemon-reload
systemctl --user enable --now crix

---

Configuration

Variable	Default	Description
CRIX_WAKEWORD_ENABLED	true	Enable wake word
CRIX_MEMORY_ENABLED	true	Enable memory
CRIX_CHROME_PROFILE	Default	Chrome profile

See PROJECT.md for full configuration reference.

---

Tech Stack

Component	Technology
Voice Framework	LiveKit Agents
STT	Deepgram Nova-3
LLM	OpenAI GPT-4.1 Mini
TTS	ElevenLabs Turbo
VAD	Silero
Input	ydotool
Browser	browser-use
Web Search	Tavily

---

Project Structure

crix/
├── src/
│   ├── agent.py              # Entry point
│   ├── tools.py              # Voice tools
│   ├── config.py             # Configuration
│   ├── wakeword/             # Wake word detection
│   └── backends/             # System backends
├── models/                   # Wake word models
├── memory/                   # Session memory
├── MEMORY.md                 # Long-term memory
├── PROJECT.md               # Full documentation
└── pyproject.toml

---

License

MIT — See LICENSE file.