An Electron-based desktop AI companion with a 3D avatar, local voice, persistent memory, and awareness of your system.

End goal: A fully local AI companion that integrates with your desktop and OS — always present, always aware, genuinely yours.

---

Requirements

• Node.js v20+ (v24 recommended via nvm)
• Python 3.10+
• Ollama — ollama.ai with the following models pulled:

  ollama pull llama3.2:3b
  ollama pull qwen3:8b
  ollama pull qwen2.5vl:7b
  

• API key — Anthropic (Claude) required. Others optional (ElevenLabs, Gemini, Groq, Mistral).

---

Install

git clone https://github.com/Raiyzs/nyxia.git
cd nyxia
npm install

Install Python dependencies for the TTS servers:

pip install --user kokoro-onnx sounddevice numpy flask

---

Run

./launch.sh

Or:

npm start

The launch script starts Ollama if it isn't running, then launches the app.

---

API Keys

Set your keys inside the app under Settings (gear icon). No .env file needed — keys are stored locally in your user config directory and never touch the project folder.

---

Features

• 3D animated avatar driven by emotion detection
• Local voice via Kokoro TTS (no internet required)
• Voice input via Whisper STT
• Persistent semantic memory (LanceDB + Kùzu graph)
• System awareness — clipboard, active windows, screen, files
• Agent loop with tool use (search, shell, browser, filesystem)
• Multi-provider support — Claude, Gemini, Groq, Mistral, Ollama
• PWA companion interface for phone