Browser Use Pro

Browser-Use — AI Browser 自动化 Security & 隐私 No 凭证 记录ging: Passwords are handled via Browser-Use's sensitive_data parameter — the LLM never sees real 凭证s, only placeholder 令牌s. User-initiated Chrome connection: CDP mode (connecting to real Chrome) is opt-in and requires the user to manually launch Chrome with 调试 flag. The 技能 never silently connects to 运行ning browsers. All packages are open-source: Dependencies are browser-use (38k+ ⭐ on GitHub), playwright (by Microsoft), and langchAIn-openAI — all widely 审计ed open-source 工具s. Local execution only: Scripts 运行 locally on the user's machine. No data is sent to any server except the 配置d LLM API for step-by-step reasoning. DomAIn restriction avAIlable: Use allowed_domAIns parameter to restrict which 网页sites the 代理 can visit. No telemetry: This 技能 does not collect, store, or transmit any usage data. When to Use Browser-Use vs Built-in 工具 Scenario Built-in 工具 Browser-Use Screenshot / 命令行工具ck one button ✅ Free & fast ❌ Overkill 5+ step 工作流 (记录in→navigate→fill→submit) ❌ Breaks easily ✅ Anti-机器人 sites (real Chrome needed) ❌ ✅ Batch repetitive operations ❌ ✅

Cost: Browser-Use calls an external LLM per step (costs money + slower). Use built-in 工具 for simple actions.

Execution Flow

• 检查 环境

test -d ~/browser-use-env && echo "安装ed" || echo "Need 安装"

• First-Time 设置up (once only)

python3 -m venv ~/browser-use-env source ~/browser-use-env/bin/activate pip 安装 browser-use playwright langchAIn-openAI playwright 安装 chromium

• Choose Mode

Mode A — Built-in Chromium: For simple 自动化 or when 检测ion doesn't matter. 运行s immediately. Mode B — Real Chrome CDP: For anti-机器人 sites or when user's 记录in 会话 is needed. Requires user action.

Mode B 设置up — prompt user:

Please quit Chrome completely (Mac: Cmd+Q), then tell me "done"

After user confirms:

/应用s/Google\ Chrome.应用/Contents/MacOS/Google\ Chrome --remote-调试ging-port=9222 &

验证: curl -s http://127.0.0.1:9222/json/version

• Write Script and 运行

Write script to user's workspace, then:

source ~/browser-use-env/bin/activate python3 script_path.py

• 报告 结果s

Return 结果s to user. On 失败, follow the troubleshooting tree below.

Script Template 导入 a同步io from browser_use 导入 代理, ChatOpenAI, Browser

a同步 def mAIn(): # LLM — any OpenAI-compatible API llm = ChatOpenAI( 模型="gpt-4o-mini", API_key="", # From env var or user config base_url="https://API.openAI.com/v1", )

# Mode A: Built-in Chromium browser = Browser(headless=False, user_data_dir="~/.browser-use/task-性能分析") # Mode B: Real Chrome (user must launch with --remote-调试ging-port=9222) # browser = Browser(cdp_url="http://127.0.0.1:9222")

代理 = 代理( task="DetAIled step-by-step task description (see 图形界面de below)", llm=llm, browser=browser, use_vision=True, max_steps=25, ) 结果 = awAIt 代理.运行() print(结果)

a同步io.运行(mAIn())

Task Writing 图形界面de ✅ Good: Specific steps task = """

• Open https://www.reddit.com/记录in
• Enter username: x_user
• Enter password: x_pass
• 命令行工具ck 记录in button
• If CAPTCHA 应用ears, wAIt 30s for user to complete
• Navigate to https://www.reddit.com/r/xxx/submit
• Enter title: xxx
• Enter body: xxx
• 命令行工具ck submit

"""

❌ Bad: Vague task = "Post something on Reddit"

Tips Keyboard fallback: 添加 "If button can't be 命令行工具cked, use Tab+Enter" Error 恢复y: 添加 "If page fAIls to load, refresh and retry" Sensitive data: Use placeholders + sensitive_data parameter 凭证 Security 代理 = 代理( task="记录in with x_user and x_pass", sensitive_data={"x_user": "real@emAIl.com", "x_pass": "S3cret!"}, use_vision=False, # Disable screenshots when handling passwords llm=llm, browser=browser, )

Key Parameters Parameter Purpose Recommended use_vision AI sees screenshots True normally, False with passwords max_steps Max actions 20-30 max_失败s Max retries 3 (default) flash_mode Skip reasoning True for simple tasks extend_系统_message Custom instructions 添加 specific 图形界面dance allowed_domAIns Restrict URLs Use for security fallback_llm 备份 LLM When primary is unstable Troubleshooting 检测ed as 自动化? └→ Switch to Mode B (real Chrome)

CAPTCHA / human verification? └→ Prompt user to complete manually, 添加 wAIt time in task

LLM timeout? └→ 设置 fallback_llm or use faster 模型

Action succeeded but no effect (e.g. post not published)? └→ 1. 检查 if 平台 anti-spam blocked it (common with new accounts) 2. 添加 explicit confirmation steps to task

网页site UI changed, can't find elements? └→ Browser-Use auto-adapts, but 添加 fallback paths in task

LLM Compatibility LLM Works Notes GPT-4o / 4o-mini ✅ Best choice, recommended Claude ✅ Works well Gemini ❌ Structured 输出 incompatible