@CycleDecoded: 一个叫 WebRover 的开源 AI Agent,你给它一句话,它自己打开浏览器、识别网页元素、点击、翻页、抓数据、搞定任务,最后把你要的结果整理得明明白白。 开源协议:MIT License 定位:自主式 Web 自动化 AI 智能体…

X AI KOLs Timeline 工具

摘要

WebRover 是一个基于 MIT 协议的开源 AI Agent,通过自然语言即可驱动浏览器完成网页自动化、跨站数据抓取与深度研究,支持本地部署。

一个叫 WebRover 的开源 AI Agent,你给它一句话,它自己打开浏览器、识别网页元素、点击、翻页、抓数据、搞定任务,最后把你要的结果整理得明明白白。 开源协议:MIT License 定位:自主式 Web 自动化 AI 智能体 这玩意凭什么能炸? 纯自然语言操控:不用写复杂的 Puppeteer/Selenium 脚本,直接输入人话就能驱动浏览器。 自主元素识别:能像人类一样“看懂”复杂的动态 DOM 结构,自己找到输入框和点击按钮。 跨网页信息搜集:不仅能看单个页面,还会自动跳转导航,跨站聚合数据。 结构化结果输出:抓下来的散乱网页信息,直接帮你清洗提炼成干净的数据。 完全开源自由:直接拿去打通自己的工作流,本地或服务器部署都行,告别昂贵 SaaS 工具。 这相当于直接给你雇了个免费的 24 小时全自动数字实习生,以后那些繁琐的爬虫、市场调研、批量填表,让它自己去折腾就行了。 项目传送门:https://github.com/hrithikkoduri/WebRover…
查看原文
查看缓存全文

缓存时间: 2026/08/05 02:17

一个叫 WebRover 的开源 AI Agent,你给它一句话,它自己打开浏览器、识别网页元素、点击、翻页、抓数据、搞定任务,最后把你要的结果整理得明明白白。

开源协议:MIT License

定位:自主式 Web 自动化 AI 智能体

这玩意凭什么能炸?

纯自然语言操控:不用写复杂的 Puppeteer/Selenium 脚本,直接输入人话就能驱动浏览器。

自主元素识别:能像人类一样“看懂”复杂的动态 DOM 结构,自己找到输入框和点击按钮。

跨网页信息搜集:不仅能看单个页面,还会自动跳转导航,跨站聚合数据。

结构化结果输出:抓下来的散乱网页信息,直接帮你清洗提炼成干净的数据。

完全开源自由:直接拿去打通自己的工作流,本地或服务器部署都行,告别昂贵 SaaS 工具。

这相当于直接给你雇了个免费的 24 小时全自动数字实习生,以后那些繁琐的爬虫、市场调研、批量填表,让它自己去折腾就行了。

项目传送门:https://github.com/hrithikkoduri/WebRover…


hrithikkoduri/WebRover

Source: https://github.com/hrithikkoduri/WebRover

WebRover

Your AI Co-pilot for Web Navigation 🚀

Autonomous Web Agent | Task Automation | Information Retrieval | Deep Research

Overview

WebRover is an AI-powered web agent that combines autonomous browsing with advanced research capabilities. While maintaining its core ability to automate web tasks, version 2.0 introduces sophisticated research workflows including multi-source analysis, academic paper generation, and deep topic exploration. The system intelligently routes queries between task automation and research modes, providing a versatile tool for both quick actions and comprehensive research.

Motivation

While traditional web automation tools excel at task execution, and search engines help with information retrieval, there’s a growing need for tools that can handle both while specializing in deep research workflows. WebRover bridges this gap by offering task automation alongside intelligent research capabilities, with a particular focus on comprehensive information gathering, analysis, and synthesis. This dual-purpose approach aims to transform how we interact with web content, making both task execution and research more efficient and thorough.

Demo Video - Deep Research Agent

https://github.com/user-attachments/assets/325c6c55-9384-4939-a912-3b1d13635799

Watch as the WebRover Deep Research Agent explores a topic, gathers information, and generates an academic paper.

Demo Video - Task Agent

https://github.com/user-attachments/assets/95ae9afb-3fdf-47f8-857e-f6a1a0d94df5

Watch as the WebRover Task Agent navigates a website and performs a task.

Key Features

Agent Capabilities

  • Three specialized agents for different use cases (Task, Research, Deep Research)
  • Dynamic agent selection based on task complexity
  • Real-time agent state visualization
  • Streaming agent actions and thoughts

Browser Integration

  • Local browser instance for privacy and control
  • Multi-tab management
  • PDF document handling
  • Secure browsing sessions

User Interface

  • Modern chat interface with real-time updates
  • Interactive agent selection
  • Action streaming with visual feedback
  • Real-time page annotations and highlights

Output Options

  • Direct chat responses
  • One-click Google Docs export
  • PDF download functionality
  • Copy to clipboard support

Research Tools

  • Vector store for information retention
  • Multi-source verification
  • Academic paper generation
  • Reference management

Technical Features

  • State-of-the-art LLM integration (GPT-4o, o3-mini-high, Claude-3.5 sonnet)
  • RAG pipeline for enhanced responses
  • LangGraph for state management
  • Playwright for reliable web automation

Agent Types

1. Task Agent

A specialized automation agent for executing web-based tasks and workflows.

  • Custom action planning for multi-step tasks
  • Dynamic element interaction based on context
  • Real-time task progress monitoring

2. Research Agent

An information gathering specialist with smart content processing.

  • Intelligent source selection and validation
  • Adaptive search refinement
  • Single-pass comprehensive information gathering

3. Deep Research Agent (New! 🎉)

An advanced research agent that produces academic-quality content through systematic topic exploration.

  • Automatic topic decomposition and structured research
  • Independent subtopic exploration
  • Academic paper generation with proper citations
  • Cross-referenced bibliography compilation

Agent Architecture Diagrams

Deep Research Agent Flow

Deep Research Agent Architecture

Deep Research Agent’s workflow for comprehensive research and content generation

Research Agent Flow

Research Agent Architecture

Research Agent’s workflow for information gathering and synthesis

Task Agent Flow

Task Agent Architecture

Task Agent’s workflow for automating web interactions

Architecture

The system is built on a modern tech stack with three distinct agent types, each powered by:

  1. State Management

    • LangGraph for maintaining agent state
    • Handles complex navigation flows and decision making
    • Structured workflow management
  2. Browser Automation

    • Playwright for reliable web interaction
    • Custom element detection and interaction system
    • Automated navigation and content extraction
  3. Content Processing

    • RAG (Retrieval Augmented Generation) pipeline
    • Vector store integration for efficient information storage
    • PDF and webpage content extraction
    • Automatic content structuring and organization
  4. AI Decision Making

    • Multiple LLM integration (GPT-4, Claude)
    • Context-aware navigation
    • Self-review mechanisms
    • Structured output generation

Setup Instructions

Backend Setup

  1. Clone the repository

    git clone https://github.com/hrithikkoduri18/WebRover.git
    cd WebRover
    cd backend
    
  2. Install Poetry (if not already installed)

    Mac/Linux:

    curl -sSL https://install.python-poetry.org | python3 -
    

    Windows:

    (Invoke-WebRequest -Uri https://install.python-poetry.org -UseBasicParsing).Content | python -
    
  3. Set Python version for Poetry

    poetry env use python3.12
    
  4. Activate the Poetry shell: For Unix/Linux/MacOS:

    poetry shell
    # or manually
    source $(poetry env info --path)/bin/activate
    

    For Windows:

    poetry shell
    # or manually
    & (poetry env info --path)\Scripts\activate
    
  5. Install dependencies using Poetry:

    poetry install
    
  6. Set up environment variables in .env:

    OPENAI_API_KEY="your_openai_api_key"
    LANGCHAIN_API_KEY="your_langchain_api_key"
    LANGCHAIN_TRACING_V2="true"
    LANGCHAIN_ENDPOINT="https://api.smith.langchain.com"
    LANGCHAIN_PROJECT="your_project_name"
    ANTHROPIC_API_KEY="your_anthropic_api_key"
    
  7. Run the backend:

    Make sure you are in the backend folder

    uvicorn app.main:app --reload --port 8000 
    

    For Windows User:

    uvicorn app.main:app --port 8000
    
  8. Access the API at http://localhost:8000

Frontend Setup

  1. Open a new terminal and make sure you are in the WebRover folder:

    cd frontend
    
  2. Install dependencies:

    npm install
    
  3. Run the frontend:

    npm run dev
    
  4. Access the frontend at http://localhost:3000

For mac users:

Try running http://localhost:3000 on Safari browser. If you face any with connecting to browser, open terminal and run:

pkill -9 "Chrome"

and try again.

If you still face issues, try changing the websocket port from 9222 to 9223 in the webrover_browser.py file in the backend/Browser folder.

Contributing

  1. Fork the repository
  2. Create your feature branch (git checkout -b feature/AmazingFeature)
  3. Commit your changes (git commit -m 'Add some AmazingFeature')
  4. Push to the branch (git push origin feature/AmazingFeature)
  5. Open a Pull Request

License

This project is licensed under the MIT License - see the LICENSE file for details.


Made with ❤️ by @hrithikkoduri

相似文章

@CycleDecoded: 搞自动化和爬虫的兄弟可以把之前的方案扔了。 GitHub 上突然爆火了一个开源项目 Index(由 AI 开发者平台 Laminar 团队打造),直接把网页浏览器变成了可调用的 API。这玩意本质上是一个极度丝滑的“AI 浏览器 Agen…

X AI KOLs Timeline

Index 是一个由 Laminar 团队开发的开源 AI 浏览器 Agent,可将任何网站转化为可调用的 API,支持 Claude、Gemini 等视觉模型,在 WebVoyager 上准确率达 92%。

@QingQ77: 用自然语言描述需求,AI Agent 自动拆解步骤、调用工具完成开发、文件操作、浏览器操控等任务,同时提供完整的编辑器与终端功能。 https://github.com/Liuchun-oss/codelf-agent… Codelf 是…

X AI KOLs Timeline

Codelf 是一个开源的桌面 AI 助理,可用自然语言描述需求,自动拆解步骤并调用工具完成开发、文件操作、浏览器操控等任务,同时提供完整的编辑器与终端功能,支持 DeepSeek、Claude、ChatGPT 等模型,对国内网络友好,具备本地 RAG 知识库能力。