> ## Documentation Index
> Fetch the complete documentation index at: https://mintlify.com/vercel-labs/agent-browser/llms.txt
> Use this file to discover all available pages before exploring further.

# Introduction

> Headless browser automation CLI for AI agents. Fast Rust CLI with Node.js fallback.

## What is Agent Browser?

Agent Browser is a powerful headless browser automation tool designed specifically for AI agents. It provides a fast, reliable command-line interface for controlling browsers programmatically, with an architecture optimized for both performance and ease of use.

<CardGroup cols={2}>
  <Card title="Fast & Efficient" icon="bolt">
    Native Rust CLI with sub-millisecond parsing overhead for maximum performance.
  </Card>

  <Card title="AI-First Design" icon="brain">
    Built for AI agents with snapshot-ref workflow, JSON output, and optimal token usage.
  </Card>

  <Card title="Production Ready" icon="shield">
    Security features including authentication vault, domain allowlists, and action policies.
  </Card>

  <Card title="Cross-Platform" icon="globe">
    Works on macOS, Linux, and Windows with automatic fallback support.
  </Card>
</CardGroup>

## Key Features

### Snapshot-Ref Workflow

The optimal way to interact with web pages. Get a snapshot of interactive elements, then use refs to interact deterministically:

```bash theme={null}
agent-browser snapshot -i
# Output shows interactive elements with refs: @e1, @e2, @e3...

agent-browser click @e2
agent-browser fill @e3 "user@example.com"
```

### Fast Architecture

Agent Browser uses a client-daemon architecture:

1. **Rust CLI** - Fast native binary for parsing and communication
2. **Node.js Daemon** - Manages Playwright browser instance
3. **Automatic Fallback** - Uses Node.js directly if native binary unavailable

The daemon starts automatically and persists between commands for fast subsequent operations.

### Security Built-In

Opt-in security features for safe AI agent deployments:

* **Authentication Vault** - Store credentials encrypted, LLM never sees passwords
* **Domain Allowlist** - Restrict navigation to trusted domains
* **Action Policies** - Gate destructive actions with policy files
* **Content Boundaries** - Wrap page output to help LLMs distinguish trusted from untrusted content

### Multiple Platforms

Run browsers anywhere:

* **Local** - Chromium, Firefox, or WebKit via Playwright
* **iOS Simulator** - Real Mobile Safari testing on macOS
* **Cloud Providers** - Browserbase, Browser Use, Kernel
* **CDP Mode** - Connect to existing browsers via Chrome DevTools Protocol

## Get Started

<CardGroup cols={2}>
  <Card title="Installation" icon="download" href="/installation">
    Install Agent Browser globally, locally, or from source.
  </Card>

  <Card title="Quickstart" icon="rocket" href="/quickstart">
    Get to a working example in under 2 minutes.
  </Card>

  <Card title="Core Concepts" icon="book" href="/concepts/snapshot-refs">
    Learn the snapshot-ref workflow and other key concepts.
  </Card>

  <Card title="AI Agents" icon="robot" href="/integrations/ai-agents">
    Integrate with Claude, GPT, and other AI coding assistants.
  </Card>
</CardGroup>

## Use Cases

### AI Agent Automation

Perfect for AI agents that need to interact with web applications:

* E2E testing with natural language instructions
* Web scraping with intelligent navigation
* Form filling and data entry
* Visual regression testing

### Development & Testing

Streamline your development workflow:

* Automated testing of web applications
* Screenshot comparison testing
* Accessibility tree inspection
* Network request monitoring

### Production Deployments

Deploy AI agents that interact with the web:

* Serverless environments (Vercel, AWS Lambda)
* CI/CD pipelines
* Monitoring and alerting systems
* Automated workflows

<Note>
  **Open Source**: Agent Browser is Apache-2.0 licensed. Contributions welcome at [github.com/vercel-labs/agent-browser](https://github.com/vercel-labs/agent-browser)
</Note>