likhonsheikh/open-computer-use-agent

Open Computer Use Agent

An open-source alternative to OpenAI's Operator and Anthropic's Computer Use.

Features

• —🖥️ Full Linux desktop (Xfce) running in the browser
• —🖱️ Mouse control (click, double-click, right-click)
• —⌨️ Keyboard input (typing, key combinations)
• —📷 Real-time screenshots
• —🌐 Firefox ESR browser included

How It Works

This Space runs a virtual X11 desktop using:

• —Xvfb: Virtual framebuffer for headless display
• —Xfce4: Lightweight desktop environment
• —xdotool: Mouse and keyboard automation
• —Gradio: Web UI for control

Usage

1. 1.Click "Take Screenshot" to see the current desktop
2. 2.Use the action controls to interact:
3. 3.Enter X,Y coordinates and click
4. 4.Type text
5. 5.Press keyboard shortcuts
6. 6.Scroll up/down

Architecture

┌─────────────────────────────────┐
│  HuggingFace Spaces Container   │
│  ┌───────────┐  ┌────────────┐  │
│  │   Xvfb    │◄─│   Xfce4    │  │
│  │   :99     │  │  Desktop   │  │
│  └─────┬─────┘  └────────────┘  │
│        │                        │
│  ┌─────▼─────┐  ┌────────────┐  │
│  │  xdotool  │◄─│  Gradio    │  │
│  │  control  │  │  UI :7860  │  │
│  └───────────┘  └────────────┘  │
└─────────────────────────────────┘

License

MIT