browser-use: The browser automation champion with 119,000 stars releases its stable version 1.0. AI-controlled browsers finally have a de facto standard.browser-use:11.9 万星的浏览器自动化王者发布 1.0 稳定版,AI 操控浏览器这件事终于有了事实标准
Enabling an AI Agent to open a browser, click buttons, fill out forms, and scrape data is something that browser-use has achieved as the most widely cited open-source implementation in the entire field – with 119,000 stars on GitHub. Over the past year, almost all "AI-controlled browser" projects have been built on top of it. The project has now released the 1.0.4 stable version: rewritten based on Pydantic AI, with a persistent Python REPL running directly on the native CDP protocol. It can take over your local Chrome and also supports cloud browsers, with both speed and stability improved.
让 AI Agent 自己开浏览器、点按钮、填表单、抓数据,这件事 browser-use 已经做成了整个领域引用最多的开源实现——GitHub 11.9 万星,一年来几乎所有「AI 操控浏览器」项目都在它之上搭建。现在项目发布 1.0.4 稳定版:基于 Pydantic AI 重写,一条持久的 Python REPL 直接跑在原生 CDP 协议上,可接管你本地的 Chrome,也支持云端浏览器,速度和稳定性都上了一个台阶。

Why is it the "de facto standard"?为什么它是「事实标准」
Traditional browser automation solutions (Selenium, Playwright) excel at executing hard-coded scripts, but what AI agents need is the ability to "understand the page and then decide where to click". browser-use connects these two layers: the lower layer uses native CDP to directly connect to the browser, while the upper layer structures the page state and passes it to the LLM for decision-making, forming a closed loop of "observation–decision–execution".
Its niche says it all: it ranks 12th among GitHub agent-themed projects with 119,826 stars, and countless workflows and tutorials use it as their default foundation. Learning it is equivalent to mastering the common paradigm of this entire category of tools.
The key feature of Version 1.0 is "Persistent REPL": The browser session remains active between multiple commands, eliminating the need to reopen the page for each step. This significantly speeds up the execution of multi-step tasks, which is particularly important for Agents that need to perform continuous operations across dozens of pages.
浏览器自动化老牌方案(Selenium、Playwright)擅长按写死的脚本执行,但 AI Agent 需要的是「看懂页面再决定点哪」——browser-use 把这两层接在一起:底层用原生 CDP 直连浏览器,上层把页面状态结构化后交给 LLM 决策,形成「观察—决策—执行」的闭环。
生态位说明一切:它以 119,826 星排在 GitHub agent 主题项目第 12 位, countless 工作流和教程默认以它为底座。你学会它,等于掌握了这一整类工具的通用范式。
1.0 版的关键词是「持久 REPL」:浏览器会话在多条命令之间保持存活,不再每步重开页面,多步任务的执行速度显著提升,这对需要连续操作几十个页面的 Agent 尤其重要。

Local Chrome or cloud browser本地 Chrome 还是云端浏览器
Local mode: Directly takes over the Chrome installed on your machine, using the logged-in session and cookies – suitable for operations that require a logged-in state (checking orders, managing the backend, auto-filling).
Cloud Mode: The project provides a hosted browser with a clean environment and concurrency-friendly features, making it suitable for bulk scraping and unattended tasks. Both modes use the same set of APIs, and switching between them requires changing only one line of configuration.
Pure Python, zero compilation: pip-installable and ready to use, fully supported on Windows/macOS/Linux. It is also one of the few browser agent libraries that maintains API stability from its 1.0 official release.
本地模式:直接接管你机器上已装的 Chrome,沿用已登录的会话和 Cookie——适合需要登录态的操作(查订单、管理后台、自动填报)。
云端模式:项目提供托管浏览器,干净环境、并发友好,适合批量抓取和无人值守任务。两种模式同一套 API,切换只改一行配置。
纯 Python、零编译:pip 装完即用,Windows / macOS / Linux 全支持,也是少数在 1.0 正式版就保持 API 稳定承诺的浏览器 Agent 库。

Installation and First Task安装与首个任务
After downloading the wheel package from this site, run pip install browser_use-1.0.4-py3-none-any.whl, or directly run pip install browser-use==1.0.4. Prepare an available large model API key.
You can get your first task up and running in less than ten lines: create an Agent, give it the instruction "Go to a certain website, find certain information, and summarise it", and then observe the entire process as it opens a browser, navigates through pages, and extracts the information.
Advanced usage: Integrate browser-use into Claude Code or your own workflow as a "hand". Once MCP is set up, the coding agent can also open the browser at any time to check online pages.
下载本站 wheel 包后执行 pip install browser_use-1.0.4-py3-none-any.whl,或直接 pip install browser-use==1.0.4;准备一个可用的大模型 API Key。
十行以内就能跑起第一个任务:创建 Agent,给它一句「去某个网站找到某个信息并汇总」,运行后观察它自己开浏览器、翻页、提取的全过程。
进阶玩法:把 browser-use 接进 Claude Code 或你自己的工作流里当「手」用——配好 MCP 后,写代码的 Agent 也能随时打开浏览器核实线上页面。

Problem Encountered遇到问题
Missing browser: Local mode requires Chrome or Chromium to be installed on the machine. If the path is incorrect, set the location of the browser executable according to the documentation. Model does not perform actions: Check the API key and model configuration. Some small models have a significantly weaker understanding of page structure; simply switch to the main model. The installation package accompanying this article can be found via the download button above.
报缺浏览器:本地模式需要机器上装有 Chrome 或 Chromium,路径异常时按文档设置浏览器可执行文件位置;模型不执行动作:检查 API Key 与模型配置,部分小模型对页面结构的理解明显偏弱,换主力模型即可。本文配套安装包见上方下载按钮。
⬇ Download · 点击下载:browser-use v1.0.4 官方 wheel 安装包
MIT 许可 · PyPI 官方 bdist_wheel · pip install 直接安装
阅读与点赞数据保存在你的浏览器本地,欢迎留下你的想法。
今日正能量学一点,用一点;今天种下的种子,会长成明天的能力。去免费下载专区 →广告
评论 文明发言,让讨论更有价值