Skip to main content
Build agents that can interact with browsers and desktop applications through screenshots, clicks, and keyboard input.

Quick Start

1

Simple Usage

2

With Configuration

Configuration

Available Tools

The computer use module provides these tools:

With Playwright (Browser)

Human Approval

By default, all actions require human approval:

Custom Tool Implementations

Agent Example

Safety Best Practices

  1. Always require approval - Enable requireApproval: true
  2. Limit capabilities - Only enable needed features
  3. Sandbox execution - Run in isolated environments
  4. Log all actions - Track what the agent does
  5. Set timeouts - Prevent runaway actions

Environment Variables

Computer Use CLI

Terminal automation

Tool Approval

Human-in-the-loop approvals

Image Agent

Vision capabilities