Sandboxing

Isolating code execution for security.

Overview

Sandboxing restricts what code can do when executed by agents:

• File system access
• Network access
• System commands
• Resource limits

Configuration

{
  sandbox: {
    enabled: true,
    type: "docker",          // docker, firejail, none
    limits: {
      memory: "512MB",
      cpu: "1",
      timeout: 30000
    }
  }
}

Sandbox Types

Docker (Recommended)

Uses Docker containers for isolation:

{
  sandbox: {
    type: "docker",
    image: "openclaw/sandbox:latest"
  }
}

Firejail (Linux)

Uses Firejail for lightweight sandboxing:

{
  sandbox: {
    type: "firejail",
    profile: "default"
  }
}

File System Restrictions

{
  sandbox: {
    filesystem: {
      readOnly: ["/"],
      readWrite: ["/tmp", "~/.config/openclaw/workspace"],
      hidden: ["/etc/passwd", "~/.ssh"]
    }
  }
}

Network Restrictions

{
  sandbox: {
    network: {
      enabled: false,        // Disable network
      allowedHosts: []       // Or whitelist hosts
    }
  }
}

Resource Limits

{
  sandbox: {
    limits: {
      memory: "512MB",
      cpu: "1",
      processes: 10,
      timeout: 30000         // ms
    }
  }
}

See Also

• Security
• Agent Workspace