🧠 Open Deep Research:500 行代码的 AI 研究员,把你的浏览器变成研究助手
🆚 你还在手动搜资料、切换 10 个 tab、复制粘贴写报告?Open Deep Research(18.9k★)用 500 行 TypeScript 实现了 OpenAI Deep Research 的核心逻辑——输入一个主题,它会自动生成搜索词、爬结果、提炼要点、递归深入,最后输出一份带来源的 Markdown 报告。
怎么用
# 克隆 + 安装
git clone git@github.com:dzhng/deep-research.git
cd deep-research
cp .env.example .env.local
# 填入 FIRECRAWL_KEY 和 OPENAI_KEY
npm install
# 跑起来
npm start
系统会问你研究什么、广度(推荐 3-10)、深度(推荐 1-5),然后自动干活。最终报告存为 report.md。
支持本地模型
# .env.local 里注释掉 OPENAI_KEY,换成:
OPENAI_ENDPOINT="http://localhost:1234/v1"
OPENAI_MODEL="your-model-name"
DeepSeek R1 也支持——设 FIREWORKS_KEY 即可自动切换。
并发控制
有付费版 Firecrawl 就加 CONCURRENCY_LIMIT=5 加速;免费版设 CONCURRENCY_LIMIT=1 避免限流。
换句话说——一个 npm start,少花半小时查资料。
🧠 Open Deep Research: A 500-Line AI Researcher That Turns Your Browser Into a Research Lab
Still manually searching 10 tabs and copy-pasting reports? Open Deep Research (18.9k★) recreates OpenAI's Deep Research in just 500 lines of TypeScript. Give it a topic — it generates search queries, scrapes results, extracts insights, recursively deep-dives, and outputs a sourced Markdown report.
Quick Start
npm install
# Set FIRECRAWL_KEY and OPENAI_KEY in .env.local
npm start
Configure breadth (3-10 recommended) and depth (1-5), answer follow-up questions, and the agent handles the rest. Final report → report.md.
Local LLMs & R1 Support
Comment out OPENAI_KEY, set OPENAI_ENDPOINT + OPENAI_MODEL for local inference. Set FIREWORKS_KEY to automatically use DeepSeek R1.
Set CONCURRENCY_LIMIT to tune crawl speed. One npm start, no more manual research.