# browser-use

Open-source browser automation agent framework that makes websites accessible to AI agents.

## Summary
browser-use is an MIT-licensed Python project for connecting AI agents to browser actions, making it useful for teams prototyping web automation, browser agents, and task execution over real websites.


## Guide
browser-use is one of the most direct ways to test whether an AI agent can operate a website. Instead of starting with a broad agent platform, it focuses on the browser surface: navigation, actions, page state, and the messy reality of web workflows.

### What it is
browser-use is an open-source Python framework that makes websites accessible to AI agents. It gives developers a way to connect model decisions to browser actions so agents can interact with web pages rather than only describe what a user should do.

### Why it matters
If action agents are going to be useful, they need to work where real work happens. Browser-based systems are still the front door to many tools, CRMs, dashboards, stores, and internal apps. browser-use is useful because it isolates that problem and lets teams test it directly.

### How it works
A typical experiment starts with a narrow browser task, a model provider, and a controlled browser session. The developer then observes how the agent reads page state, selects actions, handles errors, and recovers when a page changes.


## Use Cases
- Automating web research: Use browser-use to test agents that collect information from pages, compare options, or navigate repeated research flows.
- Form and dashboard workflows: Evaluate whether an agent can complete repeatable web tasks that previously required manual clicking.
- Browser-agent safety testing: Use it to study what permissions, prompts, and guardrails are needed when an agent controls a browser.

## Alternatives
- Use OpenClaw when you need a broader action-agent workspace vs OpenClaw: browser-use is focused on browser automation. OpenClaw is more relevant when browser control is only one layer in a larger workflow runtime.

### Getting Started
- Start with the docs: https://docs.browser-use.com/
- Inspect the repository: https://github.com/browser-use/browser-use

### FAQ
- Is browser-use open source?
  - The GitHub repository is listed with an MIT license.
- Is browser-use the same as a full agent platform?
  - No. It is more focused on browser automation. You may still need memory, scheduling, permissions, and workflow orchestration around it.
- Who should try browser-use?
  - Developers who need to test whether web tasks can be represented as reliable AI agent workflows should start here.
## Why It Matters
browser-use matters because browser control is one of the most practical surfaces for action agents. Many workflows still happen inside websites, and this project gives builders a focused open-source framework for testing that layer.


## Best For
- Developers building agents that need to operate websites
- Teams prototyping browser automation before committing to a larger agent platform
- Researchers evaluating web task completion and browser action safety

## Not For
- Users who want a complete no-code automation product
- Teams that cannot manage credentials, browser sessions, and site-specific failure cases

## What It Actually Does
- Browser-first agent surface: browser-use focuses on making websites actionable for AI agents.
  - Why it matters: Most business workflows still cross web apps, forms, dashboards, and authenticated pages.
- Python developer workflow: The project is built for developers who want to compose browser automation into agent code.
  - Why it matters: A focused framework can be easier to adopt than a full agent operating environment.
- Strong fit for narrow automation tests: It is useful for validating whether a browser task can be represented as an agent workflow.
  - Why it matters: Narrow tests reveal site failures, credential needs, and reliability issues before a team scales automation.

## Typical Use Cases
- Website task automation: Use browser-use when an agent needs to navigate, click, type, inspect pages, or complete a web flow.
- Browser agent prototypes: Test browser automation as a standalone layer before adding memory, scheduling, or full workflow orchestration.
- Web QA and research workflows: Evaluate how agents behave across real web pages, changing layouts, and multi-step browser tasks.

## How It Compares
- Choose browser-use for browser automation primitives vs full action-agent platforms: OpenClaw and OpenHands cover broader agent workflows. browser-use is more focused when the core problem is browser operation itself.

## Command Line
### Install browser-use
Use the official package path, then follow the docs for model provider and browser setup.

```bash
pip install browser-use
```

## Facts
- Category: agents
- Resource type: agent
- Open source: yes
- License: MIT
- Last verified: 2026-04-19
- GitHub repo: browser-use/browser-use
- GitHub stars: 88493

## Capabilities
- browser-automation
- browser
- workflow-orchestration
- tool-calling

## Structured Use Case Tags
- browser-agent
- self-hosted-ai
- developer-workflow

## Getting Started
- Open the GitHub repository: https://github.com/browser-use/browser-use
- Read the documentation: https://docs.browser-use.com/
- Visit the project website: https://browser-use.com

## Links
- GitHub: https://github.com/browser-use/browser-use
- Homepage: https://browser-use.com
- Docs: https://docs.browser-use.com/

## Structured Outputs
- JSON: https://www.openagent.bot/agents/browser-use.json
- Markdown: https://www.openagent.bot/agents/browser-use.md
- Canonical: https://www.openagent.bot/agents/browser-use