AI开源社区

菜单

欣淇
发布于 2026-05-11 / 0 阅读
0
0

GPT Researcher26.9k Stars Agent：

🔍 GPT Researcher：26.9k Stars 的开源深度研究 Agent，扔个问题它自己写完整报告

项目地址：assafelovic/gpt-researcher | ⭐ 26,969 | 🐍 Python | 作者：assaf_elovic

老实说，现在写研究型文章让我自己动手是不太可能了。之前每次要查个技术趋势或行业分析，得开十几个 tab，手动翻网页，再拼拼凑凑整出一篇报告，基本半天就没了。

GPT Researcher 就是冲着这个痛点来的——这是一个开源的自主研究 Agent，你扔一个问题给它，它自己规划问题、爬取信息、整理引用、生成报告，全程不用你插手。目前 GitHub 上已经 26.9k Stars，而且还在活跃更新。

怎么做到的？

核心架构其实不复杂："规划器"+"执行器"+"发布器" 三个角色协作。

  • 规划器（Planner）：收到你的问题后，自动拆解出一系列子问题
  • 执行器（Executor）：并行爬取每个子问题对应的信息源，支持 20+ 来源聚合
  • 发布器（Publisher）：把爬到的内容归纳整理，生成带引用的完整报告

    • 说白了就是把人类做研究的流程——提问、搜索、整理、写作——全交给 AI 自动化了。

    一行命令跑起来

    # 安装
pip install gpt-researcher

# 然后三行 Python 代码就能用
python -c "
from gpt_researcher import GPTResearcher
import asyncio

async def main():
    researcher = GPTResearcher(query='DeepSeek V4 对比 GPT-4o 的性能差异')
    await researcher.conduct_research()
    report = await researcher.write_report()
    print(report)

asyncio.run(main())
"

    就是这么简单。装完 pip install gpt-researcher，配好 OpenAI Key 和 Tavily Key 就能直接用。

    完整部署到 Web 界面

    如果你想要网页版，克隆仓库启动服务器就行：

    git clone https://github.com/assafelovic/gpt-researcher.git
cd gpt-researcher
pip install -r requirements.txt

# 配好环境变量
export OPENAI_API_KEY=sk-xxx
export TAVILY_API_KEY=tvly-xxx

# 启动
python -m uvicorn main:app --reload

    打开 http://localhost:8000，输入问题，等几分钟就出一份带引用来源的完整报告。

    支持 MCP 扩展数据源

    最近还上了 MCP（Model Context Protocol）支持，可以接入 GitHub、数据库、内部文档等自定义数据源：

    export RETRIEVER=tavily,mcp


    from gpt_researcher import GPTResearcher

researcher = GPTResearcher(
    query="2026年最值得关注的开源 AI 项目",
    mcp_configs=[{
        "name": "github",
        "command": "npx",
        "args": ["-y", "@modelcontextprotocol/server-github"]
    }]
)

    这样一来，研究报告不仅能搜网页，还能直接查 GitHub 仓库数据、读你内部的知识库。

    几个亮点

  • 报告最长能到 2000+ 字，带引用链接、图片、表格
  • 支持导出 PDF、Word 格式
  • 20+ 信息源并行抓取，结论更客观
  • 内置记忆和上下文保持，长研究不跑偏
  • 可以当 Claude Skill 装进对话里用：npx skills add assafelovic/gpt-researcher

    • 总结

  • 26.9k Stars，活跃维护，社区生态不错
  • 安装简单：pip install gpt-researcher，三行代码出报告
  • 架构清晰（Planner → Executor → Publisher），值得学习
  • 支持 MCP 扩展，能查 GitHub 和内部数据
  • 适合做技术调研、竞品分析、行业趋势报告

    • 🔍 GPT Researcher: 26.9k Stars Open Source Deep Research Agent — Throw It a Question, It Writes the Full Report

    Project: assafelovic/gpt-researcher | ⭐ 26,969 | 🐍 Python | Author: assaf_elovic

    Let's be honest — doing deep research manually in 2026 is a grind. Open 15 browser tabs, skim through articles, stitch together a report. Takes half a day minimum.

    GPT Researcher solves exactly this. It's an open-source autonomous research agent: give it a question, and it plans the research, crawls the web, aggregates sources, and writes a full report with citations — without you lifting a finger. It's sitting at 26.9k Stars on GitHub and still actively maintained.

    How It Works

    Three-agent architecture: Planner + Executor + Publisher.

  • Planner breaks your question into sub-questions
  • Executor crawls 20+ sources in parallel for each sub-question
  • Publisher compiles everything into a coherent report with citations

    • It automates the entire human research workflow — ask, search, organize, write.

    One-Command Setup

    pip install gpt-researcher


    from gpt_researcher import GPTResearcher
import asyncio

async def main():
    researcher = GPTResearcher(query="DeepSeek V4 vs GPT-4o performance comparison")
    await researcher.conduct_research()
    report = await researcher.write_report()
    print(report)

asyncio.run(main())

    That's it. Install the pip package, set your API keys, and you're ready.

    Web UI Mode

    git clone https://github.com/assafelovic/gpt-researcher.git
cd gpt-researcher
pip install -r requirements.txt
export OPENAI_API_KEY=sk-xxx
export TAVILY_API_KEY=tvly-xxx
python -m uvicorn main:app --reload

    Open http://localhost:8000, type your question, and get a full research report in minutes.

    MCP Integration

    Recently added MCP (Model Context Protocol) support for custom data sources:

    export RETRIEVER=tavily,mcp

    Now your research agent can query GitHub repos, databases, and internal docs alongside web search.

    Highlights

  • Reports up to 2,000+ words with citations, images, and tables
  • Export to PDF, Word
  • 20+ sources crawled in parallel for objective conclusions
  • Built-in memory and context for long research sessions
  • Can be used as a Claude Skill: npx skills add assafelovic/gpt-researcher

    • Summary

  • 26.9k Stars, actively maintained with growing community
  • Simple install: pip install gpt-researcher, 3 lines of code
  • Clean architecture (Planner → Executor → Publisher)
  • MCP support for GitHub, internal data sources
  • Perfect for tech research, competitive analysis, trend reports

    • CowAgent44k Stars...
    🤖 OpenAI Agents ...

    评论