Arch Eleven - Local AI Assistant System

A modular, local AI assistant system designed for performance and flexibility.

Features

• ✅ Modular Architecture: Backend (Node.js) and Voice Server (Python) separated for stability.
• ✅ Local LLM Support: Works seamlessly with Ollama, llama.cpp, and other local inference servers.
• ✅ Real-time Audio: Low-latency Text-to-Speech (TTS) using EdgeTTS with sentence-level streaming.
• ✅ Synchronized Animation: Canvas-based bubble interface that reacts to AI state and speech.
• ✅ Unrestricted File Access: AI can read/write files anywhere in the project (Sandbox removed).
• ✅ Conversation Memory: Persistent JSON-based memory storage.
• ✅ No Frontend Frameworks: Pure HTML/CSS/JS for maximum performance and easy modification.

System Architecture

arch-eleven/
├── backend/                 # Main Node.js Server
│   ├── server.js           # Express + WebSocket Server
│   ├── config/             # JSON Configurations
│   ├── state/              # State Machine (Idle/Thinking/Speaking)
│   ├── router/             # Intent Router
│   ├── llm/                # LLM Abstraction Layer
│   ├── voice/              # TTS Inteface
│   └── tools/              # File & Code Tools
│
├── voice/                   # Python Voice Server
│   └── server.py           # TTS Engine (EdgeTTS)
│
├── frontend/                # User Interface
│   ├── ui.js               # WebSocket Client & Audio Queue
│   └── bubble.js           # Canvas Animation
│
└── data/                   # Persistent Storage

Installation

Prerequisites

• Node.js 16+
• Python 3.10+
• Local LLM (e.g., Ollama or llama.cpp)

Setup

1. Install Backend Dependencies:

   cd backend
   npm install

2. Install Voice Server Dependencies:

   cd ..
   pip install -r requirements.txt

3. Configure LLM: Update backend/config/models.json to point to your local LLM endpoint (default is Ollama at http://localhost:11434).

Running the System

You need to run two separate terminals:

Terminal 1: Voice Server

python voice/server.py
# Runs on http://localhost:8000

Terminal 2: Backend & UI

cd backend
npm start
# Runs on http://localhost:3113

Open your browser to http://localhost:3113.

Configuration

Voice Settings

Edit backend/config/voices.json to change:

• Voice Persona: Select different EdgeTTS voices (e.g., en-US-AriaNeural, en-GB-SoniaNeural).
• Speed/Pitch: Adjust audio characteristics.

Valid Voices (EdgeTTS)

Common high-quality voices:

• en-US-AriaNeural (Default, Female)
• en-US-GuyNeural (Male)
• en-GB-SoniaNeural (British Female)
• en-AU-NatashaNeural (Australian Female)

Troubleshooting

Audio Stops or Stutters

• Check EdgeTTS: If logs show 403 Forbidden, update the library:

  pip install -U edge-tts

• Browser Autoplay: Click anywhere on the page to ensure the browser allows audio playback.
• Single Instance: Ensure you don't have multiple backend processes running on port 3113.

LLM Not Responding

• Ensure Ollama/llama.cpp is running.
• Verify the endpoint URL in backend/config/models.json.
• Check if the model name in config matches your installed model (e.g., phi3:mini).

License

MIT License. Free to use and modify.