e2b

Files (1)

• SKILL.md

SKILL.md (first lines)

# E2B — Cloud Desktop Sandbox & Computer Use

Secure cloud Linux desktops powered by [E2B Desktop Sandbox](https://e2b.dev) with autonomous computer-use capabilities, Browser Use integration, and LLM-driven visual control.

## Quick Start

```bash
# Set up environment
export E2B_API_KEY="your-e2b-api-key"
export BROWSERUSE_API_KEY="your-browseruse-api-key"  # Optional: for cloud browser sessions

# Via daemon commands
/sandbox new                    # Spin up a code sandbox
/desktop new                    # Spin up a desktop sandbox (with display)
/desktop browse https://example.com  # Open browser in desktop sandbox
/desktop screenshot             # Capture current screen
/desktop click 512 384          # Click at coordinates
/desktop type "hello world"     # Type text
/desktop key Return             # Send key
/shell ls -la                   # Run shell command in sandbox
/sandbox kill                   # Terminate sandbox
```

## Architecture

```
┌─────────────────────────────────────────────────────────────────┐
│                    SolanaOS Daemon                               │
│                                                                  │
│  ┌──────────────────┐  ┌─────────────────┐  ┌───────────────┐  │
│  │  E2B Client       │  │  Desktop Client  │  │ Browser Agent │  │
│  │  (code sandbox)   │  │  (desktop VNC)   │  │ (autonomous)  │  │
│  │                   │  │                  │  │               │  │
│  │  • RunCode()      │  │  • Screenshot()  │  │ • Vision LLM  │  │
│  │  • RunCommand()   │  │  • Click()       │  │ • Action LLM  │  │
│  │  • ListSandboxes  │  │  • TypeText()    │  │ • Tool Exec   │  │
│  │  • KillSandbox()  │  │  • SendKey()     │  │ • OODA Loop   │  │
│  │                   │  │  • OpenBrowser() │  │               │  │
│  └──────┬───────────┘  │  • RunShell()    │  └──────┬────────┘  │
│         │              │  • Scroll()      │         │           │
│         │              │  • Stream()      │         │           │
│         │              └────────┬─────────┘         │           │
│         │                       │                    │           │
│  ┌──────┴───────────────────────┴────────────────────┴────────┐ │
│  │                    E2B REST API                             │ │
│  │              https://api.e2b.dev/sandboxes                 │ │
│  └────────────────────────────────────────────────────────────┘ │
│                                                                  │
│  ┌────────────────────────────────────────────────────────────┐ │
│  │                  Browser Use Integration                    │ │
│  │                                                             │ │
│  │  Cloud Providers:                                          │ │
│  │  • Browser Use API  (BROWSERUSE_API_KEY)                   │ │
│  │  • Browserbase      (BROWSERBASE_API_KEY)                  │ │
│  │                                                             │ │
│  │  Features:                                                  │ │
│  │  • CDP-based browser control                               │ │
│  │  • Cloud session management                                │ │
│  │  • Headed/headless modes                                   │ │
│  │  • Profile persistence                                     │ │
│  │  • Proxy support                                           │ │
│  └────────────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────┘
```

## Sandbox Types

### Code Sandbox (Default)

Standard E2B code interpreter — Python execution, file I/O, shell commands. No display.

```
/sandbox new                    # Create code sandbox
/run print("hello world")      # Execute Python
/shell pip install requests     # Install packages
/sandbox kill                   # Terminate
```

### Desktop Sandbox

Full Linux desktop with X11 display, VNC streaming, mouse/keyboard control.

```
/desktop new                    # Create desktop sandbox
/desktop screenshot             # Capture screen
/desktop click 100 200          # Click at (100, 200)
/desktop type "search query"    # Type text
/desktop key ctrl+a             # Send key combo
/desktop browse https://...     # Open browser
/desktop kill                   # Terminate
```

### Browser Use Cloud Session

Cloud-hosted browser via Browser Use or Browserbase APIs.

```
/browseruse connect             # Start cloud browser session
/browseruse open https://...    # Navigate
/browseruse close               # End session
```

## Autonomous Computer Use (Browser Agent)

The Browser Agent combines vision LLMs with desktop control for autonomous task completion.

### Agent Loop (OODA)

```
1. OBSERVE:  Take screenshot of desktop
2. ORIENT:   Send screenshot to vision LLM with objective
3. DECIDE:   LLM analyzes screen, determines next action
4. ACT:      Execute action (click, type, command, etc.)
5. REPEAT:   Until objective is complete or max steps reached
```

### Supported LLM Providers for Vision

The agent supports multiple vision LLM providers (configured via daemon):

From x402agent/SolanaOS, MIT.

More from x402agent/SolanaOS