# vortex@sandbox - AI Agent Desktop Control

> Containerized Linux desktop for AI agents via WebSocket VNC

## Quick Connect

WebSocket URL: wss://megharudushi-vnc-browser-agent.hf.space/websockify
Protocol: RFB 3.8 (VNC)
Resolution: 1280x720
Auth: None required

## Python Example

```python
import asyncio, struct, websockets

async def control():
    ws = await websockets.connect("wss://megharudushi-vnc-browser-agent.hf.space/websockify", subprotocols=["binary"])
    await ws.recv(); await ws.send(b"RFB 003.008\n")  # Handshake
    await ws.recv(); await ws.send(bytes([1]))        # No auth
    await ws.recv(); await ws.send(bytes([1]))        # Shared
    await ws.recv()                                    # Server init

    # Click at (640, 360)
    await ws.send(struct.pack(">BBHH", 5, 1, 640, 360))  # Press
    await ws.send(struct.pack(">BBHH", 5, 0, 640, 360))  # Release

    # Type text
    for c in "hello":
        await ws.send(struct.pack(">BBHI", 4, 1, 0, ord(c)))  # Key down
        await ws.send(struct.pack(">BBHI", 4, 0, 0, ord(c)))  # Key up

asyncio.run(control())
```

## Commands

Mouse: struct.pack(">BBHH", 5, button_mask, x, y)
Key: struct.pack(">BBHI", 4, down, 0, keycode)
Screenshot: struct.pack(">BBHHHH", 3, 0, 0, 0, 1280, 720)

## See llms-full.txt for complete API reference