# browser-use — Python Browser Agent Toolkit

> browser-use runs a Python agent that controls a real browser for web tasks. Use the repo’s uv quickstart, then run an Agent with your LLM provider.

## Install

Save as a script file and run:

# browser-use — Python Browser Agent Toolkit

> browser-use runs a Python agent that controls a real browser for web tasks. Use the repo’s uv quickstart, then run an Agent with your LLM provider.

## Quick Use

1. Install:
   ```bash
   uv init && uv add browser-use && uv sync
   ```
2. Run:
   ```bash
   python -c "from browser_use import Agent; print('browser-use import ok')"
   ```
3. Verify:
   - Run the repo’s first-agent example and confirm it can open pages and complete a simple task


---

## Intro

browser-use runs a Python agent that controls a real browser for web tasks. Use the repo’s uv quickstart, then run an Agent with your LLM provider.

- **Best for:** developers building web-capable agents who need a real browser loop (not just HTTP requests) for login flows, forms, and navigation
- **Works with:** Python >=3.11 (repo), uv/uvx, a supported LLM provider, local or cloud browser options
- **Setup time:** 12 minutes


### Quantitative Notes

- Python >=3.11 requirement (repo)
- Setup time ~12 minutes
- GitHub stars (verified): see Source & Thanks


---

## Practical Notes

Treat browser-use as an execution layer: your agent decides *what* to do, and browser-use reliably performs *how* to do it in a browser. Start with deterministic tasks (open URL → extract one number), then expand to multi-step flows (login → navigate → download). For production, add guardrails: domain allowlists, human approval for purchases, and logs of every action.

**Safety note:** Browser automation is powerful—add permissions, allowlists, and approval gates for destructive actions.

### FAQ

**Q: Do I need Browser Use Cloud?**
A: No. The README shows an optional cloud API key; you can run locally with your own setup.

**Q: Which LLMs work?**
A: The README lists multiple providers (e.g., Browser Use, Google, Anthropic). Pick one that fits your environment and budget.

**Q: How do I make runs stable?**
A: Use narrow tasks, add explicit success criteria, and keep browser state clean (fresh profiles for tests).

---

## Source & Thanks

> GitHub: https://github.com/browser-use/browser-use
> Owner avatar: https://avatars.githubusercontent.com/u/192012301?v=4
> License (SPDX): MIT
> GitHub stars (verified via `api.github.com/repos/browser-use/browser-use`): 93,399


---

<!-- ZH -->

# browser-use——Python 浏览器自动化 Agent 工具箱

> browser-use 让你用 Python 跑一个能真实操控浏览器的 Agent 来做网页任务。按仓库 uv quickstart 安装后，选择 LLM provider 并运行 Agent 即可。

## 快速使用

1. 安装：
   ```bash
   uv init && uv add browser-use && uv sync
   ```
2. 运行：
   ```bash
   python -c "from browser_use import Agent; print('browser-use import ok')"
   ```
3. 验证：
   - Run the repo’s first-agent example and confirm it can open pages and complete a simple task


---

## 简介

browser-use 让你用 Python 跑一个能真实操控浏览器的 Agent 来做网页任务。按仓库 uv quickstart 安装后，选择 LLM provider 并运行 Agent 即可。

- **适合谁（Best for）:** 要做“能上网”的 agent 的开发者，需要真实浏览器闭环（而不是纯 HTTP 请求）来处理登录、表单与导航
- **兼容工具（Works with）:** Python >=3.11（仓库）、uv/uvx、可用的 LLM provider、本地或云浏览器选项
- **安装时间（Setup time）:** 12 分钟


### 量化信息

- 要求 Python >=3.11（仓库）
- 装机约 12 分钟
- GitHub stars（已核验）：见「来源与感谢」


---

## 实战要点

把 browser-use 当作执行层：你的 agent 决定“做什么”，browser-use 负责在真实浏览器里“怎么做”。先从确定性任务开始（打开 URL → 抽取一个数字），再扩展到多步流程（登录 → 导航 → 下载）。上生产要加护栏：域名白名单、购买等动作强制人工确认、并记录每一步操作日志。

**安全提示：** 浏览器自动化能力很强——请加权限控制、域名白名单，以及破坏性动作的审批闸门。

### FAQ

**Q: 必须用 Browser Use Cloud 吗？**
A: 不必。README 提到云 API key 是可选项；你也可以本地运行自己的浏览器环境。

**Q: 哪些 LLM 能用？**
A: README 列出多个 provider（如 Browser Use、Google、Anthropic）；按你的环境与预算选择即可。

**Q: 如何让运行更稳定？**
A: 把任务写窄、加明确成功标准，并保持浏览器状态干净（测试用全新 profile）。

---

## 来源与感谢

> GitHub：https://github.com/browser-use/browser-use
> Owner avatar：https://avatars.githubusercontent.com/u/192012301?v=4
> 许可证（SPDX）：MIT
> GitHub stars（已通过 `api.github.com/repos/browser-use/browser-use` 核验）：93,399


---
Source: https://tokrepo.com/en/workflows/browser-use-python-browser-agent-toolkit
Author: Script Depot