AI浏览器应用：
    Automate browser tasks in plain text. 以纯文本自动执行浏览器任务。
    结合LLM，将基于浏览器的任务以Prompt形式输入，并调用Playwright的headless模式完成任务

Python版本：软件依赖关系较强，新版python匹配度低，建议使用 python3.12.X / python3.12.X版本，具体以实际要求为准

• Browser Use
  
  - 官网：https://browser-use.com/ ，      Browser Use - The AI browser agent
    Automate browser tasks in plain text. 以纯文本自动执行浏览器任务。
  - Doc：Introduction - Browser Use
  - 安装

  pip install browser-use

  - 使用

  from browser_use import Agent
  from langchain_openai import ChatOpenAI
  from dotenv import load_dotenv
  import asyncio
  
  load_dotenv()
  
  async def main():
      llm = ChatOpenAI(model="llama3.1:8b")  # Refer: https://docs.browser-use.com/customize/agent/supported-models
                                              # LLM Recommendations：
                                              # Best accuracy: O3
                                              # Fastest: llama4 on groq
                                              # Balanced: fast + cheap + clever: gemini-2.5-flash or gpt-4.1-mini
                                              # Required environment variables: .env
                                              # OPENAI_API_KEY=
  
      task = "search Browser-use on 百度"     # 设置需要在Browser上完成的任务
      agent = Agent(task=task, llm=llm)     
                                          
      await agent.run()
  
  if __name__ == "__main__":
      asyncio.run(main())
  
  >>> 建立连接
  INFO     [telemetry] Anonymized telemetry enabled. See https://docs.browser-use.com/development/telemetry for more information.
  INFO     [agent] 🧠 Starting an agent with main_model=llama3.1:8b +tools +vision +memory, planner_model=None, extraction_model=llama3.1:8b 
  ERROR    [agent] 
  
  ❌  LLM ChatOpenAI connection test failed. Check that OPENAI_API_KEY is set correctly in .env and that the LLM API account has sufficient funding.
  
  Request timed out.
  
  INFO     [mem0.vector_stores.faiss] Loaded FAISS index from /tmp/mem0_1536_faiss/mem0.faiss with 0 vectors
  INFO     [mem0.vector_stores.faiss] Loaded FAISS index from C:\Users\rolei\.mem0\migrations_faiss/mem0_migrations.faiss with 8 vectors
  INFO     [mem0.vector_stores.faiss] Inserted 1 vectors into collection mem0_migrations
  INFO     [agent] 🚀 Starting task: search Browser-use on 百度
  INFO     [agent] 📍 Step 1
  ERROR    [agent] ❌ Result failed 1/3 times:
   Connection error.
  INFO     [agent] 📍 Step 1
  
  
  >>> 访问失败
  INFO     [service] Using anonymized telemetry, see https://docs.browser-use.com/development/telemetry.
  INFO     [Agent] 🧠 Starting a browser-use version 0.6.1 with model=gpt-4.1-mini
  INFO     [Agent] [34m🚀 Task: search Browser-use on 百度[0m
  INFO     [utils] 📦 Downloading uBlock Origin extension...
  WARNING  [utils] ⚠️ Failed to setup uBlock Origin extension: Failed to download extension: <urlopen error [WinError 10060] 由于连接方在一段时间后没有正确答复或连接的主机没有反应，连接尝试失败。>
  INFO     [utils] 📦 Downloading I still don't care about cookies extension...
  

• PlayWright
  
  - 官网：PlayWright for Python Fast and reliable end-to-end testing for modern web apps | Playwright Python
  - Doc：Installation | Playwright Python
  - 安装

  pip install playwright   # install playwright
  
  playwright install       # install browser driver
                           # support browsers (chromium, firefox and webkit)
  playwright install chromium  # only install chrome browser driver
  playwright install --force chrome

  - 执行
    synchronous（同步）和
    同步：发送请求 --> 等待返回 --> 再发送下一个请求，只能按顺序执行，出错后既退出
    异步：发送请求 --> 不等待返回 --> 即可再发送下一个请求，可以不按顺序执行，只关注请求的执行
  - 样例，synchronous（同步），asynchronous（异步）参考官方Doc

  from playwright.sync_api import sync_playwright
  
  with sync_playwright() as p:            # start playwright, with method(refer with open())
                                          # or using Interactive mode (REPL)
                                          # 交互式模式(Read-Eval-Print Loop,简称REPL)
                                          # playwright = sync_playwright().start()
  
      browser = p.chromium.launch()       # Chrome浏览器：chromium, firefox and webkit
                                          # default headless=True
                                          # headless mode，无头模式，不显示流览器界面，后台运行，
  
      page = browser.new_page()
      page.goto("https://www.baidu.com")  # 访问网页
      print(page.title())                 # 输出网页标题
      browser.close()
                                          # playwright.stop(), stop playwright if using start()
  
  >>> 百度一下，你就知道

  - API，参考 https://playwright.dev/python/docs/api
  - Playwright命令

  playwright --help
  Usage: playwright [options] [command]
  
  Options:
    -V, --version                          output the version number
    -h, --help                             display help for command
  
  Commands:
    open [options] [url]                   open page in browser specified via -b, --browser
    codegen [options] [url]                open page and generate code for user actions
    install [options] [browser...]         ensure browsers necessary for this version of Playwright are installed
    uninstall [options]                    Removes browsers used by this installation of Playwright from the system
                                           (chromium, firefox, webkit, ffmpeg). This does not include branded channels.
    install-deps [options] [browser...]    install dependencies necessary to run browsers (will ask for sudo permissions)
    cr [options] [url]                     open page in Chromium
    ff [options] [url]                     open page in Firefox
    wk [options] [url]                     open page in WebKit
    screenshot [options] <url> <filename>  capture a page screenshot
    pdf [options] <url> <filename>         save page as pdf
    run-server [options]
    show-trace [options] [trace...]        show trace viewer
    help [command]                         display help for command

  - 脚本录制，参考 Generating tests | Playwright Python

  > playwright codegen https://baidu.com -o test.py  # 启动脚本录制并输出到文件
  
  
  C:\Python313\Scripts>playwright codegen --help
  Usage: playwright codegen [options] [url]
  
  open page and generate code for user actions
  
  Options:
    -o, --output <file name>             saves the generated script to a file
    --target <language>                  language to generate, one of javascript, playwright-test, python, python-async,
                                         python-pytest, csharp, csharp-mstest, csharp-nunit, java, java-junit (default:
                                         "python")
    --test-id-attribute <attributeName>  use the specified attribute to generate data test ID selectors
    -b, --browser <browserType>          browser to use, one of cr, chromium, ff, firefox, wk, webkit (default:
                                         "chromium")
    --block-service-workers              block service workers
    --channel <channel>                  Chromium distribution channel, "chrome", "chrome-beta", "msedge-dev", etc
    --color-scheme <scheme>              emulate preferred color scheme, "light" or "dark"
    --device <deviceName>                emulate device, for example  "iPhone 11"
    --geolocation <coordinates>          specify geolocation coordinates, for example "37.819722,-122.478611"
    --ignore-https-errors                ignore https errors
    --load-storage <filename>            load context storage state from the file, previously saved with --save-storage
    --lang <language>                    specify language / locale, for example "en-GB"
    --proxy-server <proxy>               specify proxy server, for example "http://myproxy:3128" or
                                         "socks5://myproxy:8080"
    --proxy-bypass <bypass>              comma-separated domains to bypass proxy, for example
                                         ".com,chromium.org,.domain.com"
    --save-har <filename>                save HAR file with all network activity at the end
    --save-har-glob <glob pattern>       filter entries in the HAR by matching url against this glob pattern
    --save-storage <filename>            save context storage state at the end, for later use with --load-storage
    --timezone <time zone>               time zone to emulate, for example "Europe/Rome"
    --timeout <timeout>                  timeout for Playwright actions in milliseconds, no timeout by default
    --user-agent <ua string>             specify user agent string
    --user-data-dir <directory>          use the specified user data directory instead of a new context
    --viewport-size <size>               specify browser viewport size in pixels, for example "1280, 720"
    -h, --help                           display help for command

  

  - lib: %Python_path%\Lib\site-packages\playwright\_impl
    > _locator.py： selector locator，元素定位方法
       * all_text_contents，selector识别不足，可以通过系统输出组件内所有selector，然后筛选
    > _frame.py：FrameLocator
  - Issue
    > 缺少 Google API 密钥，因此 Chrominum 的部分功能将无法使用
       快捷方式： %userprofile%\Local\ms-playwright\chromium-1181\chrome-win\chrome.exe --test-type=webdriver

• Browse Use Web-UI
  
  - 安装
    > python 3.11 / python 3.12
    > https://github.com/browser-use/web-ui.git -- 下载zip文件后解压
    > 安装依赖lib -- 注意下载lib与已下载lib版本之间的冲突

  cd %Web-UI%     # 进入Web-UI目录
  
  pip install -r requirements.txt  # install dependency lib
  
  ############## Web-UI folder list
  # .
  # ..
  # .dockerignore
  # .env
  # .env.example       -- AI environment setting
  # .github
  # .gitignore
  # .vscode
  # assets
  # docker-compose.yml
  # Dockerfile
  # LICENSE
  # README.md
  # requirements.txt    -- dependency lib
  # SECURITY.md
  # src
  # supervisord.conf
  # tests
  # webui.py            -- run Web-UI

  - 启动

  python webui.py --ip 127.0.0.1 --port 7788
  
  >>> No module named 'gradio'      # pip install gradio
  >>> No module named 'distutils'   # Distutils2 is the packaging library that supersedes Distutils. ‌Distutils, expired after Python 3.12，replace by setuptools
                                    # pip install distutils2
  >> No module named 'ConfigParser' # pip install ConfigParser