> 2025/2/6 用 Deep Research 整理 https://chatgpt.com/share/67a4b86c-fe8c-8008-a5e3-a41dd8d35aac ## 長文寫作 * gpt-author https://github.com/mshumer/gpt-author * claude-journalist * https://github.com/mshumer/ai-journalist * https://twitter.com/mattshumer_/status/1772286375817011259 (2024/3/25) * [[STORM]] ## 新聞、研究收集 * https://blog.langchain.dev/exploring-uxs-besides-chat-with-research-assistant/ (2023/11) * https://x.com/Saboo_Shubham_/status/1815575181978746888 (2024/7) - GPT Researcher - https://gptr.dev/ - https://docs.gptr.dev/ - https://github.com/assafelovic/gpt-researcher ⭐ - https://blog.langchain.dev/gpt-researcher-x-langchain/ - 新版 https://twitter.com/LangChainAI/status/1788669909410644459 - https://blog.langchain.dev/how-to-build-the-ultimate-ai-automation-with-multi-agent-collaboration/ - GPT Newspaper - https://github.com/assafelovic/gpt-newspaper * [ ] claude-researcher * https://twitter.com/mattshumer_/status/1776341679617745126 * fork https://twitter.com/BinxNet/status/1776381322707304884 * https://github.com/Binxly/ai-researcher * [ ] gemini-youtube-researcher * https://twitter.com/mattshumer_/status/1777857512512819585 - [ ] Llama-Researcher https://github.com/rsrohan99/llama-researcher - LlamaIndex workflows * [[STORM]] - mshumer OpenDeepResearcher https://github.com/mshumer/OpenDeepResearcher - https://x.com/mattshumer_/status/1886558939434664404 (2025/2/4) - OpenRouter, SerpAPI , Jina API, Python notebook - 概念就 [[Iterative RAG]] - 用 prompt 改寫 user query 產生出四個 search query - query 用 SerpAPI 找出 links - links 用 Jina 抓出網頁內容 - 用 prompt 截取出相關的內容 - 根據目前累積的內容，用 prompt 判斷是否還需要 query (迴圈跑直到 ok 或超過設定上限) - 最後總結報告 - 2+1 prompts - 單迴圈，但有實作平行發送 search query 來加速 - Deeper Seeker https://github.com/HarshJ23/Deeper-Seeker - https://x.com/karminski3/status/1886563448840183965 (2025/2/4) - Python, Exa API - 概念也是 [[Iterative RAG]] - 不過使用 messages 的方式很巧妙，把 exa result 累加在 assistant message，然後用 prompt 產生出下一步的搜尋 query - 直到次數上限 or 無搜尋結果 or 無 query 後，最後產生報告 - just 2 prompt - 有改 2.0 版本更複雜了 - [ ] Open Deep Research https://github.com/nickscamara/open-deep-research - Firecrawl, Next.js AI SDK - [ ] open-deep-research https://github.com/btahir/open-deep-research - Next.js - [ ] Open Deep Research https://github.com/dzhng/deep-research#open-deep-research - https://x.com/dzhng/status/1886603396578484630 (2025/2/4) - Node.js, Firecrawl, OpenAI API - jina-ai 出的 node-DeepResearch https://github.com/jina-ai/node-DeepResearch ⭐ - Node.js - Brave search, jina readers, gemini - 單迴圈， prompt 類似 ReAct 寫法，提供工具挑選下一步: search, answer, reflect, visit - hugginface 的 DeepResearch - https://huggingface.co/blog/open-deep-research - https://github.com/huggingface/smolagents/tree/gaia-submission-r1/examples/open_deep_research - https://x.com/AymericRoucher/status/1886846796061688028 (2025/2/5) - 使用 smolagents 的 CodeAgent 做法 https://arxiv.org/abs/2402.01030 - https://huggingface.co/docs/smolagents/examples/multiagents * r1-reasoning-rag https://medium.com/@denizaskin/reasoning-recursive-retrieval-with-deepseek-r1-tavily-and-langgraph-8d0a32d2df27 (2025/2) * https://github.com/deansaco/r1-reasoning-rag/tree/main * 用 LangGraph 做的，檢索後 判斷是否有 missing information，如果有 miss 就再次檢索 ## Langchain - [ ] News TL;DR using Langgraph - https://github.com/NirDiamant/GenAI_Agents/blob/main/all_agents_tutorials/news_tldr_langgraph.ipynb - https://x.com/NirDiamantAI/status/1869763369596596395 - [ ] Tavily’s Company Researcher - https://x.com/LangChainAI/status/1875595554786349164 (2025/1/5) - https://blog.tavily.com/companyresearcher/ - [ ] Ollama Deep Researcher (2024/12/12) ⭐ - https://x.com/LangChainAI/status/1866879366317346941 - https://github.com/langchain-ai/ollama-deep-researcher - 用到 https://arxiv.org/abs/2410.04343v1 - 產生 search query -> 寫 summary -> reflect -> (跑固定次數的 loop 後) -> 總結 summary * langchain 的 Structured Report Generation (2025/1/6) * https://blog.langchain.dev/structured-report-generation-blueprint/ * https://github.com/langchain-ai/langchain-nvidia/blob/main/cookbook/structured_report_generation.ipynb ## Llamaindex * Research Paper Report Generating Agent 研究論文報告生成代理人 (2024/12/27) * https://colab.research.google.com/github/run-llama/llamacloud-demo/blob/main/examples/report_generation/research_paper_report_generation.ipynb#scrollTo=0w1c_cWgyYto * https://x.com/llama_index/status/1872322023151833335 (2024/12/27) * Document Research Assistant for Blog Creation (2025/1/8) * https://x.com/jerryjliu0/status/1876807888112665011 * https://www.llamaindex.ai/blog/document-research-assistant-for-blog-creation-with-nvidia-nim-microservices * https://docs.llamaindex.ai/en/stable/examples/agent/nvidia_document_research_assistant_for_blog_creation/