Home 零对冲(ZeroHedge)每日HackerNews

Show HN: Cleverb.ee – 一个开源的、能够撰写带参考文献的研究报告的智能代理
Show HN: Cleverb.ee – open-source agent that writes a cited research report

原始链接: https://github.com/SureScaleAI/cleverbee

CleverBee是一个基于Python的科研代理，它利用大型语言模型（LLM）来自动化网络研究。它使用Playwright进行网页浏览、内容提取和清理（将HTML转换为Markdown），然后根据用户定义的研究主题总结研究结果。Chainlit提供了一个交互式用户界面。 其主要功能包括：支持模型上下文协议 (MCP) 用于外部工具；可配置的多LLM架构（使用Gemini模型进行规划、下一步分析和总结）；自动化网页浏览；集成的令牌跟踪和成本估算；支持Gemini、Claude和本地GGUF模型的模块化LLM客户端；LLM缓存提高性能并降低成本。 配置通过`config.yaml`文件管理。CleverBee完全支持macOS和Linux系统，在Windows系统上通过WSL有限支持。它采用AGPLv3许可证，贡献者需要同意贡献者许可协议 (CLA)。安装过程包括克隆代码库并运行提供的安装和运行脚本。

Cleverb.ee是由nickwatson设计的一个开源代理，它结合网络资源、PDF文件、Reddit和YouTube内容生成带引文的调研报告。它使用Gemini 2.5 Pro进行规划和写作，使用Gemini 2.5 Flash进行总结。该系统旨在生成平衡的、经过事实核查的、带有内嵌引文的报告，并强调数据来源的透明性。 一位用户请求了一个示例，结果生成了一份关于咖啡及其健康影响的报告。反馈指出了一些问题，例如引用了无关的内容（例如，浏览器访问失败的尝试），促使了修复。用户也讨论了来源可信度和偏见的问题，因为该工具包含了Reddit和YouTube等平台。Nickwatson强调了该工具评估来源和考虑潜在偏见的能力。对话转向了将此过程标记为“研究”的争论，并考虑了“文献综述”和“数字信息综合”作为替代方案。人们也提出了对固有AI偏见和过度依赖大型语言模型的担忧。
相关文章
  • LangManus：一个基于LangChain + LangGraph的开源Manus代理 2025-03-23
  • （评论） 2025-03-26
  • Show HN：Crawlee for Python – 一个网页抓取和浏览器自动化库 2024-07-11
  • Show HN：Hyperbrowser MCP 服务器——通过浏览器连接AI智能体到网络 2025-03-21
  • Emacs-copilot：Emacs 的大型语言模型代码完成 2024-01-01
    • 原文

    CleverBee Logo

    Website • Documentation • GitHub

    CleverBee is a powerful Python-based research agent using Large Language Models (LLMs) like Claude and Gemini, Playwright for web browsing, and Chainlit for an interactive UI. It performs research by browsing the web, extracting content (HTML), cleaning it, and summarizing findings based on user research topics.

    • 🌐 Interactive web UI via Chainlit
    • 🔧 MCP Tool Support: Integrates external tools via the Model Context Protocol (MCP)
    • 🧠 Multi-LLM Research: Uses distinct, configurable LLMs for different tasks:
      • Primary LLM: Gemini 2.5 Pro for planning and final report generation
      • Next Step LLM: Gemini 2.5 Flash for analyzing research progress and deciding next actions
      • Summarizer LLM: Gemini 2.0 Flash for intermediate web content summarization
    • 🌍 Automated Web Browsing: Utilizes Playwright for searching the web and extracting HTML content
    • 📊 Content Processing: Cleans HTML to Markdown before summarization
    • 📈 Integrated Token Tracking: Monitors token usage and estimates costs for LLM calls
    • ⚙️ Highly Configurable: Settings managed via config.yaml
    • 🚀 Modular LLM Clients: Supports different providers (Gemini, Claude, Local GGUF via llama-cpp-python)
    • 💾 LLM Caching: Employs NormalizingCache (SQLite-based) for improved performance and cost reduction

    🖥️ System Compatibility

    • macOS: Fully tested and supported, including both Intel and Apple Silicon (via Rosetta 2)
    • Linux: Fully supported, with NVIDIA GPU detection and optimization for local models
    • Windows: Limited support via Windows Subsystem for Linux (WSL)
    # Clone the repository
git clone https://github.com/SureScaleAI/cleverbee.git
cd cleverbee

# Run the setup script
bash setup.sh

# Start the application
bash run.sh

    For full documentation, visit our website: https://cleverb.ee/docs

    All major configuration is handled in config.yaml. See the documentation for detailed configuration options.

    This project is licensed under the GNU Affero General Public License, Version 3.0. See the LICENSE file for the full text.

    By contributing to this project, you agree that your contributions will be licensed under its AGPLv3 license, and you grant a copyright license to your contributions according to the terms of the Contributor License Agreement (CLA).

    联系我们 contact @ memedata.com