MyDeviceAI Desktop

Transform your computer into an AI inference hub. Connect from any device via peer-to-peer networking to run AI models locally on your desktop.

Overview

MyDeviceAI Desktop is a cross-platform Electron application that enables remote devices (ios & android via the MyDeviceAI app) to leverage your desktop's computing power for AI inference. Using WebRTC peer-to-peer connections, devices can send prompts and receive AI-generated responses without relying on cloud services.

Key Features

• Local AI Inference: Run GGUF format models locally using llama.cpp
• Peer-to-Peer Networking: Direct WebRTC connections with devices via Cloudflare Workers signaling
• Model Management: Search, download, and switch between AI models from Hugging Face
• Cross-Platform: Supports macOS, Linux, and Windows
• Privacy-Focused: All inference happens locally on your machine
• Streaming Responses: Real-time token streaming for responsive AI interactions
• Modern UI: Clean, dark-themed interface with live connection monitoring

Quick Start

Prerequisites

• Node.js v20 or higher
• 10+ GB free disk space (for AI models)
• Stable internet connection (for initial model download)

Installation

From Release

Download the latest installer for your platform:

• macOS: .zip archive (ARM/M-series Macs only)
  • Important: After extracting, run xattr -c mydeviceai-desktop.app to remove quarantine attributes
  • Without this step, macOS will report the app as damaged and prevent it from opening
  • The app will be code-signed in future releases
• Linux: .deb package (Ubuntu/Debian)
• Windows: .exe installer

From Source

# Clone the repository
git clone https://github.com/navedmerchant/MyDeviceAI-Desktop.git
cd MyDeviceAI-Desktop

# Install dependencies
npm install

# Create environment configuration
cp src/Env.example.ts src/Env.ts
# Edit src/Env.ts with your P2P signaling server URL

# Start development server
npm start

First Run

1. Launch the application
2. The app will automatically download llama.cpp for your platform
3. Default AI model (Qwen3-4B, ~2.5GB) will be downloaded
4. Once setup completes, you'll see your Room ID
5. Use this Room ID to connect from other devices

Usage

Connecting Devices

1. Note your Room ID displayed in the app
2. On your mobile device or other computer, use the companion app
3. Enter the Room ID to establish a peer-to-peer connection
4. Send prompts and receive AI-generated responses

Managing Models

• Active Model: Displayed in the status bar
• Download Models: Search and download from Hugging Face
• Configure Parameters: Adjust temperature, top-p, max tokens, etc.
• Switch Models: Stop current model and load a different one

Room Management

• Current Room ID: Shown at the top of the interface
• Regenerate Room: Click to create a new Room ID (disconnects current peers)

Architecture

Tech Stack

• Frontend: TypeScript, Electron, HTML/CSS
• Backend: Node.js, Electron main process
• AI Runtime: llama.cpp (bundled)
• Networking: WebRTC, Cloudflare Workers (signaling)
• Build System: Webpack, Electron Forge

Project Structure

src/
├── index.ts           # Main process entry point
├── renderer.ts        # UI logic and P2P client
├── preload.ts        # IPC bridge (security layer)
├── llamaSetup.ts     # llama.cpp management
├── modelManager.ts   # Model download and lifecycle
├── p2pcf/            # P2P networking library
├── index.html        # Main window template
└── index.css         # Application styling

P2P Protocol

Communication uses WebRTC data channels with JSON messages:

• hello: Initial peer handshake
• version_negotiate: Protocol version exchange
• prompt: AI completion request
• tokens: Streaming response chunks
• model_info: Current model metadata

Development

Build Commands

# Development mode with hot reload
npm start

# Run linter
npm run lint

# Package application
npm run package

# Create platform installers
npm run make

# Publish release
npm run publish

Platform-Specific Builds

The CI/CD pipeline automatically builds for:

• Linux: DEB package (Ubuntu/Debian) - uses Ubuntu-compiled llama.cpp
• macOS: ZIP distribution
• Windows: Squirrel installer

Builds are triggered on git tags matching v* pattern.

Configuration

Environment Configuration:

The application requires a P2P signaling server URL to be configured in src/Env.ts.

P2P Signaling Server Deployment Options:

You can deploy the P2P CF signaling server using one of the following methods:

Option 1: Deploy on Cloudflare Workers

1. Use the official P2PCF worker implementation: p2pcf/worker.js
2. Deploy to Cloudflare Workers following their deployment guide
3. Update src/Env.ts with your Cloudflare Worker URL

Option 2: Deploy on Railway

1. Use the standalone signaling server: p2pcf-signalling
2. Follow the deployment steps in the repository to deploy on Railway
3. Update src/Env.ts with your Railway deployment URL

Model Parameters (configurable per model):

• Temperature
• Top-p, Top-k
• Maximum tokens
• Context window size
• GPU layers (for acceleration)

Security

The application implements several security measures:

• Electron Fuses: Code integrity validation
• Preload Sandboxing: Minimal IPC surface
• Content Security Policy: Restricted resource loading
• No Remote Code: All code loaded from ASAR bundle
• Local Inference: No data sent to external servers

Troubleshooting

macOS: App is damaged and can't be opened

On macOS, you may see an error stating the app is damaged. This occurs due to Gatekeeper quarantine attributes on unsigned apps.

Solution:

xattr -c mydeviceai-desktop.app

This removes the quarantine attribute and allows the app to run. Future releases will include proper code signing to eliminate this step.

Note: The macOS build is compiled for ARM architecture (Apple Silicon/M-series chips) only. Intel Macs are not currently supported.

llama.cpp fails to start

• Check logs in the collapsible "Logs" panel
• Ensure no other process is using the assigned port
• Verify model file integrity in the models directory

Peers cannot connect

• Verify Room ID is correct
• Check firewall settings (WebRTC requires UDP)
• Ensure STUN/TURN servers are accessible

Model download fails

• Check available disk space (models are 2-10GB)
• Verify internet connection
• Try a different Hugging Face model

Contributing

Contributions are welcome! Please:

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Submit a pull request

License

MIT

Author

Naved Merchant (naved.merchant@gmail.com)

Acknowledgments

• llama.cpp - High-performance LLM inference
• Electron - Cross-platform desktop framework
• P2PCF - Peer-to-peer communication library
• Hugging Face - Model hosting and distribution

---

Note: This is an early-stage project. Features and APIs may change in future releases.