一、browser-use是什么

Browser Use 是一款开源Python库，专为大语言模型设计的智能浏览器工具，目的是让 AI 能够像人类一样自然地浏览和操作网页。它支持多标签页管理、视觉识别、内容提取，并能记录和重复执行特定动作。Browser Use 还支持开发者自定义动作，如保存数据到数据库，文件等。支持多种主流的大型语言模型，如 DeepSeek，GPT-4 和 Claude等，并支持同时运行多个任务，具备自我修正功能，从而提高任务执行的准确性和效率。

官网：https://browser-use.com/

项目网址 ：https://github.com/browser-use/browser-use

Browser-use采用的技术栈为：

• 1、Observation：页面解析层，采用DOM解析+截图辅助的非视觉+视觉方案。
  • DOM解析（HTML + XPath）：Browser-use通过底层框架（如Playwright）获取当前页面的完整HTML结构，并提取文本、元素属性等关键信息。
  • 截图辅助：在某些情况下（如验证码识别、动态图形验证），纯HTML解析可能无法直接获取信息，此时系统会自动或按需生成页面截图，并将截图作为辅助输入传递给视觉模型
• 2、Thought：核心决策层，分析Observation提供的页面信息并生成操作指令。
• 3、Action：指令执行层，微软开发的Playwright作为浏览器控制框架直接与浏览器交互完成自动化任务。Playwright作为新一代高性能UI自动化测试框架，提供低延迟、高稳定性的浏览器控制能力，支持快速页面加载和元素操作。

二、browser-use webui 主要功能

提供了全新的网页界面，简单好用，方便操作。
支持更多大语言模型，比如 Gemini、OpenAI、Azure 等，哦，还有最近爆火的国产大模型 DeepSeek，未来还会加更多。
支持用自己的浏览器，不用再反复登录，还能录屏。
定制了更智能的 Agent，通过优化后的提示让浏览器使用更高效。

使用场景

• 自动化任务：适合重复高频的浏览器操作任务，如表单填写，信息检索，文件下载
• 数据收集：适合爬取网络上的数据，如爬虫自动化测试：适合WEB UI
• 自动化测试，结合pytest轻松实现web自动化

三、使用教程

1.python 安装

python 官网： https://www.python.org/downloads/版本必须在 3.11 以上。

2、把项目clone下来

git clone https://github.com/browser-use/web-ui.git
cd web-ui

3、安装依赖

pip install browser-use
playwright install
pip install -r requirements.txt

4、配置环境

基于 .env.example 复制一个 .env 文件，并在 .env 文件中修改以下信息

# 路径 Chrome 浏览器路径（检查下自己的路径），例如
# Mac OS "/Applications/Google Chrome.app/Contents/MacOS/Google Chrome"
# Windows "C:\Program Files\Google\Chrome\Application\chrome.exe"
CHROME_PATH="/Applications/Google Chrome.app/Contents/MacOS/Google Chrome"# 浏览器的用户数据路径，例如
# Mac OS "/Users/<YourUsername>/Library/Application Support/Google/Chrome"
# Windows "C:\Users\<YourUsername>\AppData\Local\Google\Chrome\User Data"
CHROME_USER_DATA="/Users/<YourUsername>/Library/Application Support/Google/Chrome"# 还有一些大模型的 API Key 也要改
...

5、启动

执行如下命令启动

python webui.py --ip 127.0.0.1 --port 7788

启动成功如下所示：

浏览器访问 http://127.0.0.1:7788/，看到如下界面就成功了

6、配置

1.配置 Agent

注意，这里的 Use Vision，默认是选中状态，如果使用的 DeepSeek 不能勾选，因为 DeepSeek 不支持视觉输入，注意这里很多人踩坑，一定要注意。

2.配置要用的大模型

例如，下面我用的是 deepseek。

3.关于浏览器的一些设置

四、DeepSeek 的API获取

DeepSeek ：https://platform.deepseek.com/api_keys

五、界面Demo 演示

输入要执行的任务就可以点击 Run Agent 了

以下是运行时的项目日志输出，记录了执行步骤

执行的过程中也会打开浏览器和跳转到目标网站，按照区块一样对页面元素做标注。

在 Recodings 下会记录执行过程和反馈结果，还可以回看的。

六、代码示例

1.创建Agent

from langchain_openai import ChatOpenAI
from browser_use import Agent
import asyncioasync def main():agent = Agent(task="Go to Reddit, search for 'browser-use' in the search bar, click on the first post and return the first comment.",llm=ChatOpenAI(model="gpt-4o"),)result = await agent.run()print(result)asyncio.run(main())

如果没有openai-key的可以使用其他模型，下面以DeepSeek为例：

该文件在 browser-use/examples/deepseek.py

import asyncio
import osfrom dotenv import load_dotenv
from langchain_openai import ChatOpenAI
from pydantic import SecretStrfrom browser_use import Agent# dotenv
load_dotenv()api_key = os.getenv('DEEPSEEK_API_KEY', 'sk-xxxxxx')
if not api_key:raise ValueError('DEEPSEEK_API_KEY is not set')async def run_search():agent = Agent(task=("1. 在搜索框中输入抖音并搜索"'2. 点击搜索结果中的第一个链接''3. 关闭扫码登录'	'3. 返回第一个视频的内容'),llm=ChatOpenAI(base_url='https://api.deepseek.com/v1',model='deepseek-chat',api_key=SecretStr(api_key),),use_vision=False,)await agent.run()if __name__ == '__main__':asyncio.run(run_search())

运行结果如下：

七、实例展示

一、爬取基金数据

llm = ChatOpenAI(model='deepseek-chat',api_key='*************',base_url='https://api.deepseek.com',temperature=0
)asyncdefmain():agent = Agent(task="""1、导航到网址:https://fund.eastmoney.com/2、点击基金排行3、返回排行前10的基金数据，以json格式返回""",llm=llm,use_vision=False,)result = await agent.run()print(result.final_result())asyncio.run(main())

二、效果展示

1、导航到指定网址

2、点击基金排行

3、提取top 10的基金数据

DEBUG    [browser_use] --act Execution time: 0.00 seconds
INFO     [controller] 📄  Extracted from page
: ```json
{
"top_10_funds": [{"rank": 1,"fund_code": "018124","fund_name": "永赢先进制造智选混合发起A","date": "03-07","unit_net_value": "2.1654","cumulative_net_value": "2.1654","daily_growth_rate": "2.21%","1_week": "9.15%","1_month": "17.58%","3_months": "64.06%","6_months": "191.13%","1_year": "115.94%","since_inception": "116.54%","handling_fee": "0.15%"},{"rank": 2,"fund_code": "018125","fund_name": "永赢先进制造智选混合发起C","date": "03-07","unit_net_value": "2.1501","cumulative_net_value": "2.1501","daily_growth_rate": "2.21%","1_week": "9.15%","1_month": "17.54%","3_months": "63.89%","6_months": "190.55%","1_year": "115.07%","since_inception": "115.01%","handling_fee": "0.00%"},{"rank": 3,"fund_code": "016530","fund_name": "鹏华碳中和主题混合A","date": "03-07","unit_net_value": "1.7881","cumulative_net_value": "1.7881","daily_growth_rate": "3.00%","1_week": "10.23%","1_month": "22.92%","3_months": "68.07%","6_months": "178.39%","1_year": "104.00%","since_inception": "78.81%","handling_fee": "0.15%"},{"rank": 4,"fund_code": "016531","fund_name": "鹏华碳中和主题混合C","date": "03-07","unit_net_value": "1.7685","cumulative_net_value": "1.7685","daily_growth_rate": "3.00%","1_week": "10.21%","1_month": "22.86%","3_months": "67.81%","6_months": "177.59%","1_year": "102.79%","since_inception": "76.85%","handling_fee": "0.00%"},{"rank": 5,"fund_code": "001970","fund_name": "泰信鑫选灵活配置混合A","date": "03-07","unit_net_value": "1.3310","cumulative_net_value": "1.3310","daily_growth_rate": "-1.04%","1_week": "7.25%","1_month": "4.80%","3_months": "31.00%","6_months": "125.59%","1_year": "95.45%","since_inception": "33.10%","handling_fee": "0.15%"},{"rank": 6,"fund_code": "002580","fund_name": "泰信鑫选灵活配置混合C","date": "03-07","unit_net_value": "1.3220","cumulative_net_value": "1.3220","daily_growth_rate": "-0.97%","1_week": "7.31%","1_month": "4.84%","3_months": "31.15%","6_months": "125.60%","1_year": "95.27%","since_inception": "31.67%","handling_fee": "0.00%"},{"rank": 7,"fund_code": "016295","fund_name": "新华利率债债券E","date": "03-07","unit_net_value": "1.7977","cumulative_net_value": "1.9906","daily_growth_rate": "-0.13%","1_week": "-0.06%","1_month": "-0.66%","3_months": "0.85%","6_months": "94.13%","1_year": "92.89%","since_inception": "99.00%","handling_fee": "0.00%"},{"rank": 8,"fund_code": "019457","fund_name": "平安先进制造主题股票发起A","date": "03-07","unit_net_value": "1.7593","cumulative_net_value": "1.7593","daily_growth_rate": "1.78%","1_week": "10.41%","1_month": "23.92%","3_months": "57.40%","6_months": "134.29%","1_year": "90.71%","since_inception": "75.93%","handling_fee": "0.15%"},{"rank": 9,"fund_code": "019458","fund_name": "平安先进制造主题股票发起C","date": "03-07","unit_net_value": "1.7452","cumulative_net_value": "1.7452","daily_growth_rate": "1.78%","1_week": "10.40%","1_month": "23.87%","3_months": "57.17%","6_months": "133.60%","1_year": "89.59%","since_inception": "74.52%","handling_fee": "0.00%"},{"rank": 10,"fund_code": "007713","fund_name": "华富科技动能混合A","date": "03-07","unit_net_value": "1.4600","cumulative_net_value": "1.5100","daily_growth_rate": "1.47%","1_week": "8.19%","1_month": "18.94%","3_months": "47.07%","6_months": "135.41%","1_year": "89.14%","since_inception": "51.93%","handling_fee": "0.15%"}]
}

二、结合pytest实现页面自动化测试
异步执行需要安装插件pytest-asyncio

pip install pytest-asyncio

@pytest.mark.asyncio
@pytest.mark.parametrize("username,password,expected", [("kevin@xxxx.com", "a123456", "kevin"),("kevin@xxxx.com", "123456", "账号密码输入错误")])
asyncdeftest_login(username, password, expected):agent = Agent(task=f"""1、导航到网址:https://xxxxxxx.com2、输入用户名:{username}, 密码:{password}3、点击登录按钮4、验证是否登录成功,登录成功返回{expected}""",llm=llm,use_vision=False,)result = await agent.run()assert expected in str(result.final_result())

三、总结

Browser Use 深度集成大语言模型（LLM），通过语义理解与视觉决策链实现零硬编码自动化，彻底颠覆传统脚本开发模式。AI 自动解析页面结构、动态生成操作路径，无需人工编写 XPath/CSS 定位器，开发效率提升 5 倍以上，尤其擅长处理动态验证、反爬策略及多步骤交互场景，成为金融数据抓取、跨平台测试的新一代智能引擎。