[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-TIGER-AI-Lab--OpenResearcher":3,"tool-TIGER-AI-Lab--OpenResearcher":64},[4,17,27,35,43,56],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":16},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,3,"2026-04-05T11:01:52",[13,14,15],"开发框架","图像","Agent","ready",{"id":18,"name":19,"github_repo":20,"description_zh":21,"stars":22,"difficulty_score":23,"last_commit_at":24,"category_tags":25,"status":16},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",138956,2,"2026-04-05T11:33:21",[13,15,26],"语言模型",{"id":28,"name":29,"github_repo":30,"description_zh":31,"stars":32,"difficulty_score":23,"last_commit_at":33,"category_tags":34,"status":16},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",107662,"2026-04-03T11:11:01",[13,14,15],{"id":36,"name":37,"github_repo":38,"description_zh":39,"stars":40,"difficulty_score":23,"last_commit_at":41,"category_tags":42,"status":16},3704,"NextChat","ChatGPTNextWeb\u002FNextChat","NextChat 是一款轻量且极速的 AI 助手，旨在为用户提供流畅、跨平台的大模型交互体验。它完美解决了用户在多设备间切换时难以保持对话连续性，以及面对众多 AI 模型不知如何统一管理的痛点。无论是日常办公、学习辅助还是创意激发，NextChat 都能让用户随时随地通过网页、iOS、Android、Windows、MacOS 或 Linux 端无缝接入智能服务。\n\n这款工具非常适合普通用户、学生、职场人士以及需要私有化部署的企业团队使用。对于开发者而言，它也提供了便捷的自托管方案，支持一键部署到 Vercel 或 Zeabur 等平台。\n\nNextChat 的核心亮点在于其广泛的模型兼容性，原生支持 Claude、DeepSeek、GPT-4 及 Gemini Pro 等主流大模型，让用户在一个界面即可自由切换不同 AI 能力。此外，它还率先支持 MCP（Model Context Protocol）协议，增强了上下文处理能力。针对企业用户，NextChat 提供专业版解决方案，具备品牌定制、细粒度权限控制、内部知识库整合及安全审计等功能，满足公司对数据隐私和个性化管理的高标准要求。",87618,"2026-04-05T07:20:52",[13,26],{"id":44,"name":45,"github_repo":46,"description_zh":47,"stars":48,"difficulty_score":23,"last_commit_at":49,"category_tags":50,"status":16},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",84991,"2026-04-05T10:45:23",[14,51,52,53,15,54,26,13,55],"数据工具","视频","插件","其他","音频",{"id":57,"name":58,"github_repo":59,"description_zh":60,"stars":61,"difficulty_score":10,"last_commit_at":62,"category_tags":63,"status":16},3128,"ragflow","infiniflow\u002Fragflow","RAGFlow 是一款领先的开源检索增强生成（RAG）引擎，旨在为大语言模型构建更精准、可靠的上下文层。它巧妙地将前沿的 RAG 技术与智能体（Agent）能力相结合，不仅支持从各类文档中高效提取知识，还能让模型基于这些知识进行逻辑推理和任务执行。\n\n在大模型应用中，幻觉问题和知识滞后是常见痛点。RAGFlow 通过深度解析复杂文档结构（如表格、图表及混合排版），显著提升了信息检索的准确度，从而有效减少模型“胡编乱造”的现象，确保回答既有据可依又具备时效性。其内置的智能体机制更进一步，使系统不仅能回答问题，还能自主规划步骤解决复杂问题。\n\n这款工具特别适合开发者、企业技术团队以及 AI 研究人员使用。无论是希望快速搭建私有知识库问答系统，还是致力于探索大模型在垂直领域落地的创新者，都能从中受益。RAGFlow 提供了可视化的工作流编排界面和灵活的 API 接口，既降低了非算法背景用户的上手门槛，也满足了专业开发者对系统深度定制的需求。作为基于 Apache 2.0 协议开源的项目，它正成为连接通用大模型与行业专有知识之间的重要桥梁。",77062,"2026-04-04T04:44:48",[15,14,13,26,54],{"id":65,"github_repo":66,"name":67,"description_en":68,"description_zh":69,"ai_summary_zh":69,"readme_en":70,"readme_zh":71,"quickstart_zh":72,"use_case_zh":73,"hero_image_url":74,"owner_login":75,"owner_name":76,"owner_avatar_url":77,"owner_bio":78,"owner_company":79,"owner_location":79,"owner_email":80,"owner_twitter":79,"owner_website":79,"owner_url":81,"languages":82,"stars":91,"forks":92,"last_commit_at":93,"license":79,"difficulty_score":10,"env_os":94,"env_gpu":95,"env_ram":96,"env_deps":97,"category_tags":106,"github_topics":107,"view_count":10,"oss_zip_url":79,"oss_zip_packed_at":79,"status":16,"created_at":111,"updated_at":112,"faqs":113,"releases":119},598,"TIGER-AI-Lab\u002FOpenResearcher","OpenResearcher","OpenResearcher: A Fully Open Pipeline for Long-Horizon Deep Research Trajectory Synthesis","OpenResearcher 是一款完全开源的智能体大语言模型，专为长周期的深度研究场景设计。它致力于解决传统 AI 在面对复杂、多步骤信息检索和综合分析时容易迷失方向或无法持续跟进的痛点。通过自主规划研究轨迹，OpenResearcher 能够像专业研究员一样，逐步深入挖掘并整合海量信息。\n\nOpenResearcher 特别适合开发者、科研人员以及追求高效研究的进阶用户使用。其核心亮点在于采用了 30B 参数规模的架构，在 BrowseComp-Plus 基准测试中取得了 54.8% 的高准确率，性能表现甚至超越了部分主流闭源模型。值得一提的是，OpenResearcher 不仅提供了预训练模型和高质量数据集，还开放了完整的训练代码，已被 NVIDIA Nemotron 系列模型采纳。用户可以直接在 Hugging Face 上体验 Demo，也可以根据需求自行部署或微调，是构建自动化深度研究流程的理想选择。","\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_8af7fffeb758.png\" height=\"82\" style=\"vertical-align: middle;\">\n  \u003Cimg src=\".\u002Fassets\u002Fimgs\u002Fopenresearcher-title.svg\" height=\"66\" style=\"vertical-align: middle;\">\u003C\u002Fp>\n\n\u003Cdiv align=\"center\" style=\"line-height: 1; margin-top: 16px;\">\n\n\u003Cdiv align=\"center\">\n    \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fpapers\u002F2603.20278\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FarXiv-B31B1B?style=for-the-badge&logo=arXiv&logoColor=white\" alt=\"Blog\">\u003C\u002Fa>\n    \u003Ca href=\"https:\u002F\u002Fx.com\u002Fzhuofengli96475\u002Fstatus\u002F2036475211063648414\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FTwitter-000000?style=for-the-badge&logo=X&logoColor=white\" alt=\"Blog\">\u003C\u002Fa>\n    \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FTIGER-AI-Lab\u002FOpenResearcher\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FGithub-181717?style=for-the-badge&logo=github&logoColor=white\" alt=\"Blog\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FOpenResearcher\u002FOpenResearcher-Dataset\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FDataset-FFB7B2?style=for-the-badge&logo=huggingface&logoColor=ffffff\" alt=\"Dataset\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002FOpenResearcher\u002FOpenResearcher-30B-A3B\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FModel-FFD966?style=for-the-badge&logo=huggingface&logoColor=ffffff\" alt=\"Model\">\u003C\u002Fa>\n    \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fspaces\u002FOpenResearcher\u002FOpenResearcher\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FDemo-F97316.svg?style=for-the-badge&logo=gradio&logoColor=white\" alt=\"Demo\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fx.com\u002Fzhuofengli96475\u002Fstatus\u002F2021682952074097086\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FVideo-CA4245?style=for-the-badge&logo=youtube&logoColor=white\" alt=\"Video\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FOpenResearcher\u002FOpenResearcher-Eval-Logs\u002Ftree\u002Fmain\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FEval%20Logs-755BB4?style=for-the-badge&logo=google-sheets&logoColor=white\" alt=\"Eval Logs\">\u003C\u002Fa>\n\u003C\u002Fdiv>\n\n\u003Cdiv align=\"center\">\n  \u003Cimg src=\"assets\u002Fimgs\u002Fnvidia_banner.svg\" alt=\"Adopted by NVIDIA's Nemotron family of models!\">\n\u003C\u002Fdiv>\n\u003C\u002Fdiv>\n\u003Cbr>\n\u003Cp align=\"center\">\n  🤗 \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fcollections\u002FTIGER-Lab\u002Fopenresearcher\" target=\"_blank\">HuggingFace\u003C\u002Fa> ｜\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_0b0f1a7c995c.png\" width=\"14px\" style=\"display:inline;\"> \u003Ca href=\"https:\u002F\u002Fjoin.slack.com\u002Ft\u002Fopenresearcher\u002Fshared_invite\u002Fzt-3p0r32cky-PqtZkVjjWIAI14~XwcRMfQ\" target=\"_blank\">Slack\u003C\u002Fa> | \u003Cimg src=\"assets\u002Fimgs\u002Fwechat.svg\" width=\"14px\" style=\"display:inline;\"> \u003Ca href=\"assets\u002Fimgs\u002Fwechat_group.jpg\" target=\"_blank\">WeChat\u003C\u002Fa> \n\u003C\u002Fp>\n\n## 📣 News \n+ **[2026.3.24]** 🔥 **OpenResearcher [paper](https:\u002F\u002Fhuggingface.co\u002Fpapers\u002F2603.20278) is now available**, highlighting practical insights into deep research pipeline design. Already adopted by **NVIDIA's Nemotron family of models**!\n+ **[2026.2.25]** Honored to be among the **top 3 trending datasets** on 🤗 [Hugging Face](https:\u002F\u002Fhuggingface.co\u002Fdatasets) — now **11K+** downloads! \n+ **[2026.2.18]** The OpenResearcher training [code](https:\u002F\u002Fgithub.com\u002FTIGER-AI-Lab\u002FOpenResearcher?tab=readme-ov-file#-optional-train-your-own-openresearcher) is now available. Start training your own OpenResearcher!\n+ **[2026.2.14]** Excited to have our OpenResearcher demo [video](https:\u002F\u002Fx.com\u002Fzhuofengli96475\u002Fstatus\u002F2021682952074097086). Dive in and unlock the power of Deep Research today!\n+ **[2026.2.12]**  Excited to see **OpenResearcher** powering deep research trajectory generation in [**NVIDIA’s NeMo Data Designer**](https:\u002F\u002Fnvidia-nemo.github.io\u002FDataDesigner\u002Flatest\u002Fdevnotes\u002Fdeep-research-trajectories-with-nemo-data-designer-and-mcp-tool-use\u002F)!\n+ **[2026.2.10]** Our X [post](https:\u002F\u002Fx.com\u002FDongfuJiang\u002Fstatus\u002F2020946549422031040) received **1.2K+ likes**! Feel free to check out the post and join the discussion! \n\n## 💥 Introduction\n\n**OpenResearcher** is a fully open agentic large language model (30B-A3B) designed for **long-horizon deep research** scenarios. It achieves an impressive **54.8%** accuracy on [BrowseComp-Plus](https:\u002F\u002Fhuggingface.co\u002Fspaces\u002FTevatron\u002FBrowseComp-Plus), surpassing performance of `GPT-4.1`, `Claude-Opus-4`, `Gemini-2.5-Pro`, `DeepSeek-R1` and `Tongyi-DeepResearch`. We **fully open-source** the training and evaluation recipe—including data, model, training methodology, and evaluation framework for everyone to progress deep research.\n\u003Cdiv align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_1cb7576bd696.png\" alt=\"OpenResearcher Teaser\" width=\"100%\" style=\"max-width: 850px; border-radius: 8px; box-shadow: 0 4px 10px rgba(0,0,0,0.1);\">\n\u003C\u002Fdiv>\n\n\u003Cbr>\n\n## 🏆 Deep Research Benchmark Results\n\n\u003Cdiv align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_af15b42ac583.png\" alt=\"Deep Research Benchmark Results\" width=\"100%\">\n\u003C\u002Fdiv>\n\n\n## ✨ Features\n+ 🔑 **Fully Open-Source Recipe** — We fully open-source our 96K high-quality [DeepResearch trajectory dataset](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FOpenResearcher\u002FOpenResearcher-Dataset) with 100+ turns generated by GPT-OSS-120B with [native browser tools](https:\u002F\u002Fdocs.vllm.ai\u002Fprojects\u002Frecipes\u002Fen\u002Flatest\u002FOpenAI\u002FGPT-OSS.html#usage:~:text=Limitation%20section%20below.-,Tool%20Use,-%C2%B6), the leading [30B-A3B model](https:\u002F\u002Fhuggingface.co\u002FOpenResearcher\u002FOpenResearcher-30B-A3B) trained on it, [distillation recipe](https:\u002F\u002Farxiv.org\u002Fabs\u002F2603.20278), and a lightweight [DeepResearch evaluation framework](https:\u002F\u002Fgithub.com\u002FTIGER-AI-Lab\u002FOpenResearcher) to progress deep research.\n\n+ 💰 **Highly Scalable and Low-Cost** — We generate DeepResearch trajectories at massive scale using self-built retriever over a dedicated ~11B-token [corpus](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FOpenResearcher\u002FOpenResearcher-Corpus), eliminating the need for external Search APIs. This scalable retriever significantly reduces training costs.\n\n+ 🚀 **Remarkable Performance on Deep Research Benchmarks** — OpenResearcher demonstrates leading performance across a range of deep research benchmarks, including BrowseComp-Plus, BrowseComp, GAIA, xbench-DeepSearch.\n\n## 📋 Table of Contents\n\n- [🛠 Environment Setup](#-environment-setup)\n  - [Installation](#installation)\n  - [Deep Research Benchmarks Preparation](#deep-research-benchmarks-preparation)\n- [🔍 Configuration](#-configuration)\n- [🚀 Quick Start](#-quick-start)\n- [🔬 Benchmark OpenResearcher](#-benchmark-openresearcher)\n  - [Example 1: BrowseComp-Plus with Local Search Engine](#example-1-browsecomp-plus-with-local-search-engine)\n  - [Example 2: GAIA with Serper API (No Local Search Needed)](#example-2-gaia-with-serper-api-no-local-search-needed)\n  - [Evaluation](#evaluation)\n  - [Quick Commands](#quick-commands)\n- [🧪 (Optional) Train Your Own OpenResearcher](#-optional-train-your-own-openresearcher)\n- [🤝 Core Contributors](#-core-contributors)\n- [🎓 Advisors](#-advisors)\n- [🙏 Acknowledgements](#-acknowledgements)\n- [✨ Contributing](#-contributing)\n- [📚 Citation](#-citation)\n## 🛠 Environment Setup\nWe run this repo on the following setup:\n+ 8 * A100 80G Nvidia GPUs\n+ Linux operating system\n\nOther hardware setups can also work, but remember to modify the corresponding parameters.\n### Installation \n```bash\nsudo apt update \nsudo apt install -y openjdk-21-jdk\n\n# install uv\ncurl -LsSf https:\u002F\u002Fastral.sh\u002Fuv\u002Finstall.sh | sh\nuv venv --python 3.12\nsource .venv\u002Fbin\u002Factivate\n\n# install tevatron for BrowseComp-plus \ngit clone https:\u002F\u002Fgithub.com\u002Ftexttron\u002Ftevatron.git\ncd tevatron\nuv pip install -e .\ncd ..\n\n# install all dependencies automatically\nuv pip install -e .\n```\n\n### Deep Research Benchmarks Preparation\n\nRun the setup script to automatically download the **[BrowseComp-Plus](https:\u002F\u002Farxiv.org\u002Fabs\u002F2508.06600)** benchmark. Other benchmarks, including **[BrowseComp](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.12516)**, **[GAIA](https:\u002F\u002Farxiv.org\u002Fabs\u002F2311.12983)** and **[xbench-DeepResearch](https:\u002F\u002Fgithub.com\u002FTHUDM\u002Fxbench)**, will be set up automatically when they are first used.\n\n```bash\nbash setup.sh\n```\n\n**This script will:**\n- ✅ Verify Python 3.12 virtual environment and automatically install any missing dependencies\n- ✅ Downlaod BrowseComp-Plus dataset from HuggingFace and set up the directory structure\n\nFor more info about these deep research benchmarks, see [benchmarks.md](assets\u002Fdocs\u002Fbenchmarks.md) \n\n## 🔍 Configuration\n\nCopy the template and configure your API keys:\n\n```bash\ncp .env.template .env\n```\n\nEdit `.env`:\n```bash\n# Serper API (for web search when using browser_backend=serper)\nSERPER_API_KEY=your_key        # Get from: https:\u002F\u002Fserper.dev\u002F\n\n# OpenAI API (for evaluation scoring)\nOPENAI_API_KEY=your_key        # Get from: https:\u002F\u002Fplatform.openai.com\u002Fapi-keys\n```\n\n\n## 🚀 Quick Start\n**Prerequisites:** Install dependencies and configure API keys (see [Environment Setup](#-environment-setup) and [Configuration](#-configuration))\n\n1. **Deploy OpenResearcher-30B-A3B**:\n\n```bash\nbash scripts\u002Fstart_nemotron_servers.sh\n```\n\nThe complete vLLM server logs can be found in the `logs` directory.\n\n2. **Run your first task** (Before proceeding, check the logs in `logs` directory to ensure the vLLM server is deployed.)\n\n```python\nimport asyncio\nfrom deploy_agent import run_one, BrowserPool\nfrom utils.openai_generator import OpenAIAsyncGenerator\n\nasync def main():\n    # Initialize generator and browser\n    generator = OpenAIAsyncGenerator(\n        base_url=\"http:\u002F\u002Flocalhost:8001\u002Fv1\",\n        model_name=\"OpenResearcher\u002FOpenResearcher-30B-A3B\",\n        use_native_tools=True\n    )\n    browser_pool = BrowserPool(search_url=None, browser_backend=\"serper\")\n\n    # Run deep research\n    await run_one(\n        question=\"What is the latest news about OpenAI?\",\n        qid=\"quick_start\",\n        generator=generator,\n        browser_pool=browser_pool,\n    )\n\n    browser_pool.cleanup(\"quick_start\")\n\nif __name__ == \"__main__\":\n    asyncio.run(main())\n```\n\nThe deep research agent will automatically search the web, browse webpages, and extract relevant information. You'll see the final answer along with all intermediate reasoning steps.\n\n\n## 🔬 Benchmark OpenResearcher\nWe benchmark our OpenResearcher-30B-A3B using below deep research benchmarks: \n\n| Benchmark | Dataset Key | Size | Language | Search Backend | Description |\n|-----------|-------------|------|----------|----------------|-------------|\n| [BrowseComp-Plus](https:\u002F\u002Farxiv.org\u002Fabs\u002F2508.06600) | `browsecomp_plus` | 830 | EN | local | Deep-research benchmark from BrowseComp isolating retriever and LLM agent effects |\n| [BrowseComp](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.12516) | `browsecomp` | 1,266 | EN | serper | A Simple Yet Challenging Benchmark for Browsing Agents |\n| [GAIA-text](https:\u002F\u002Farxiv.org\u002Fabs\u002F2311.12983) | `gaia` | 103 | EN | serper | Text-only subset of GAIA benchmark (dev split) |\n| [xbench-DeepResearch](https:\u002F\u002Fgithub.com\u002FTHUDM\u002Fxbench) | `xbench` | 100 | ZH | serper | DeepSearch benchmark with encrypted test cases |\n\nFor more info about these deep research benchmarks, see [benchmarks.md](assets\u002Fdocs\u002Fbenchmarks.md) \n\n### Example 1: BrowseComp-Plus with Local Search Engine\n\nComplete evaluation using local dense search with browsecomp-plus [corpus](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FTevatron\u002Fbrowsecomp-plus-corpus) and [embeddings](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FTevatron\u002Fbrowsecomp-plus-indexes\u002Ftree\u002Fmain\u002Fqwen3-embedding-8b) (**note: only applicable for BrowseComp-Plus**):\n\n```bash\n# Terminal 1: Start local Dense search service on port 8000\n# Embedding model (Qwen3-Embedding-8B) will be deployed on GPUs 7\nbash scripts\u002Fstart_search_service.sh dense 8000\n\n# Terminal 2: Start vLLM servers (requires 4 GPUs)\n# TP=2, deploy 2 servers starting from port 8001 on GPUs 0,1,2,3\nbash scripts\u002Fstart_nemotron_servers.sh 2 8001 0,1,2,3\n\n# Terminal 3: Run agent\nbash run_agent.sh results\u002Fbrowsecomp_plus\u002FOpenResearcher_dense 8001 2 browsecomp_plus local OpenResearcher\u002FOpenResearcher-30B-A3B\n```\n\nWhat this does:\n- Deploys Dense retriever service on port 8000 as search engine\n- Launches 2 vLLM servers (ports 8001, 8002) with TP=2 across 4 GPUs\n- Runs deepresearch agent with load balancing across both servers\n\n### Example 2: GAIA with Serper API (No Local Search Needed)\n\nRun with Serper Google Search API (**note: applicable to all benchmarks except BrowseComp-Plus**):\n\n```bash\n# Terminal 1: Start vLLM servers (requires 4 GPUs)\nbash scripts\u002Fstart_nemotron_servers.sh 2 8001 0,1,2,3\n\n# Terminal 2: Run agent with serper search backend\nbash run_agent.sh results\u002Fgaia\u002FOpenResearcher_serper 8001 2 gaia serper OpenResearcher\u002FOpenResearcher-30B-A3B\n```\n\n**Browser Backend Options:**\n- `local` - Use local BM25\u002FDense search service (for BrowseComp-Plus)\n- `serper` - Use Serper Google Search API (for all other benchmarks)\n\nFor other parameters, refer to [parameter.md](assets\u002Fdocs\u002Fparameter.md).\n\n### Evaluation\n\nAfter running experiments, evaluate results:\n\n```bash\n# eval on browsecomp_plus\npython eval.py --input_dir results\u002Fbrowsecomp_plus_dense\u002FOpenResearcher_dense\n\n# eval on gaia\npython eval.py --input_dir results\u002Fgaia\u002FOpenResearcher_serper\n```\n\n### Quick Commands\n\n| Scenario | Command |\n|----------|---------|\n| BrowseComp-Plus (BM25) | `bash scripts\u002Fstart_search_service.sh bm25 8000` then `bash scripts\u002Fstart_nemotron_servers.sh 2 8001 0,1,2,3` then `bash run_agent.sh results\u002Fbrowsecomp-plus\u002FOpenResearcher_bm25 8001 2 browsecomp_plus local OpenResearcher\u002FOpenResearcher-30B-A3B` |\n| BrowseComp-Plus (Qwen3-8B Dense Embeddings) | `bash scripts\u002Fstart_search_service.sh dense 8000` then `bash scripts\u002Fstart_nemotron_servers.sh 2 8001 0,1,2,3` then `bash run_agent.sh results\u002Fbrowsecomp-plus\u002FOpenResearcher_dense 8001 2 browsecomp-plus local OpenResearcher\u002FOpenResearcher-30B-A3B` |\n| BrowseComp | `bash scripts\u002Fstart_nemotron_servers.sh 2 8001 0,1,2,3` then `bash run_agent.sh results\u002Fbrowsecomp 8001 2 browsecomp serper OpenResearcher\u002FOpenResearcher-30B-A3B` |\n| GAIA | `bash scripts\u002Fstart_nemotron_servers.sh 2 8001 0,1,2,3` then `bash run_agent.sh results\u002Fgaia 8001 2 gaia serper OpenResearcher\u002FOpenResearcher-30B-A3B` |\n| xbench-DeepResearch | `bash scripts\u002Fstart_nemotron_servers.sh 2 8001 0,1,2,3` then `bash run_agent.sh results\u002Fxbench 8001 2 xbench serper OpenResearcher\u002FOpenResearcher-30B-A3B` |\n\nFor script parameter explanation, refer to [parameter.md](assets\u002Fdocs\u002Fparameter.md).\n\n**Note:** Don't forget to evaluate your results using:  \n```bash\npython eval.py --input_dir [INPUT_DIR]\n```\n\n## 🧪 (Optional) Train Your Own OpenResearcher\n\nOur [OpenResearcher-30B-A3B](https:\u002F\u002Fhuggingface.co\u002FOpenResearcher\u002FOpenResearcher-30B-A3B) is trained using [Megatron-LM](https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM) on [openresearcher-dataset](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FOpenResearcher\u002FOpenResearcher-Dataset). To get started, clone the `openresearcher` branch of the Megatron-LM repository:\n```\ngit clone -b openresearcher https:\u002F\u002Fgithub.com\u002Fjdf-prog\u002FMegatron-LM.git\n```\nThen, follow the training instructions [here](https:\u002F\u002Fgithub.com\u002Fjdf-prog\u002FMegatron-LM\u002Ftree\u002Fopenresearcher\u002Fexamples\u002Fopenresearcher) to train your own OpenResearcher!\n\n## 🤝 Core Contributors\n\n\u003Ctable>\n\u003Ctr>\n    \u003Ctd align=\"center\">\n        \u003Ca href=\"https:\u002F\u002Fzhuofeng-li.github.io\u002F\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_9d0aa70f5ca3.png\" width=\"75px;\" alt=\"Zhuofeng Li\"\u002F>\n            \u003Cbr \u002F>\n            \u003Csub>\u003Cb>Zhuofeng Li\u003C\u002Fb>\u003C\u002Fsub>\n        \u003C\u002Fa>\n    \u003C\u002Ftd>\n        \u003Ctd align=\"center\">\n        \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fjdf-prog\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_51ff34d286e7.png\" width=\"75px;\" alt=\"Dongfu Jiang\"\u002F>\n            \u003Cbr \u002F>\n            \u003Csub>\u003Cb>Dongfu Jiang\u003C\u002Fb>\u003C\u002Fsub>\n        \u003C\u002Fa>\n    \u003C\u002Ftd>\n    \u003C\u002Ftd>\n        \u003Ctd align=\"center\">\n        \u003Ca href=\"https:\u002F\u002Fmxueguang.github.io\u002F\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_2a674cb4063e.jpg\" width=\"75px;\" alt=\"Xueguang\"\u002F>\n            \u003Cbr \u002F>\n            \u003Csub>\u003Cb>Xueguang Ma\u003C\u002Fb>\u003C\u002Fsub>\n        \u003C\u002Fa>\n    \u003C\u002Ftd>\n    \u003Ctd align=\"center\">\n        \u003Ca href=\"https:\u002F\u002Fisaacghx.github.io\u002Fabout\u002F\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_550e40ea0dd1.png\" width=\"75px;\" alt=\"Haoxiang Zhang\"\u002F>\n            \u003Cbr \u002F>\n            \u003Csub>\u003Cb>Haoxiang Zhang\u003C\u002Fb>\u003C\u002Fsub>\n        \u003C\u002Fa>\n    \u003C\u002Ftd>\n    \u003Ctd align=\"center\">\n        \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Ferenup\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_2a32ad203de7.png\" width=\"75px;\" alt=\"Ping Nie\"\u002F>\n            \u003Cbr \u002F>\n            \u003Csub>\u003Cb>Ping Nie\u003C\u002Fb>\u003C\u002Fsub>\n        \u003C\u002Fa>\n    \u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftable>\n\n## 🎓 Advisors\n\n\u003Ctable>\n\u003Ctr>\n      \u003Ctd align=\"center\">\n        \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fwenhuchen\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_1c524a3cdf75.png\" width=\"75px;\" alt=\"Wenhu Chen\"\u002F>\n            \u003Cbr \u002F>\n            \u003Csub>\u003Cb>Wenhu Chen\u003C\u002Fb>\u003C\u002Fsub>\n        \u003C\u002Fa>\n    \u003C\u002Ftd>\n    \u003Ctd align=\"center\">\n        \u003Ca href=\"https:\u002F\u002Fyuzhimanhua.github.io\u002F\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_349331e941f2.jpg\" width=\"75px;\" alt=\"Yu Zhang\"\u002F>\n            \u003Cbr \u002F>\n            \u003Csub>\u003Cb>Yu Zhang\u003C\u002Fb>\u003C\u002Fsub>\n        \u003C\u002Fa>\n    \u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftable>\n\n## 🙏 Acknowledgements\n\u003Cdiv align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_a6d53a348144.png\" alt=\"Deep Research Benchmark Results\" width=\"100%\">\n\u003C\u002Fdiv>\n\n\n## ✨ Contributing\nWe are truly looking forward to open-source contributions to OpenResearcher! If you’re interested in contributing, collaborating, or reporting issues, please feel free to open an issue or submit a pull request (PR). You can also reach us at [zhuofengli12345@gmail.com](mailto:zhuofengli12345@gmail.com).\n\nWe are also looking forward to your feedback and suggestions!\n\n##  📚 Citation\n\n```bibtex\n@article{li2026openresearcher,\n  title={{OpenResearcher: A Fully Open Pipeline for Long-Horizon Deep Research Trajectory Synthesis}},\n  author={Li, Zhuofeng and Jiang, Dongfu and Ma, Xueguang and Zhang, Haoxiang and Nie, Ping and Zhang, Yuyu and Zou, Kai and Xie, Jianwen and Zhang, Yu and Chen, Wenhu},\n  journal={arXiv preprint arXiv:2603.20278},\n  year={2026}\n}\n```\n\n\n## ⭐ Star History\n[![Star History Chart](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_8d97aedf14fd.png)](https:\u002F\u002Fwww.star-history.com\u002F#TIGER-AI-Lab\u002FOpenResearcher&type=date&legend=top-left)\n\n\u003Cp align=\"right\" style=\"font-size: 14px; margin-top: 20px;\">\n  \u003Ca href=\"#readme-top\" style=\"text-decoration: none; font-weight: bold;\">\n    ↑ Back to Top ↑\n  \u003C\u002Fa>\n\u003C\u002Fp>\n\n\n","\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_8af7fffeb758.png\" height=\"82\" style=\"vertical-align: middle;\">\n  \u003Cimg src=\".\u002Fassets\u002Fimgs\u002Fopenresearcher-title.svg\" height=\"66\" style=\"vertical-align: middle;\">\u003C\u002Fp>\n\n\u003Cdiv align=\"center\" style=\"line-height: 1; margin-top: 16px;\">\n\n\u003Cdiv align=\"center\">\n    \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fpapers\u002F2603.20278\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FarXiv-B31B1B?style=for-the-badge&logo=arXiv&logoColor=white\" alt=\"论文\">\u003C\u002Fa>\n    \u003Ca href=\"https:\u002F\u002Fx.com\u002Fzhuofengli96475\u002Fstatus\u002F2036475211063648414\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FTwitter-000000?style=for-the-badge&logo=X&logoColor=white\" alt=\"博客\">\u003C\u002Fa>\n    \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FTIGER-AI-Lab\u002FOpenResearcher\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FGithub-181717?style=for-the-badge&logo=github&logoColor=white\" alt=\"GitHub\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FOpenResearcher\u002FOpenResearcher-Dataset\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FDataset-FFB7B2?style=for-the-badge&logo=huggingface&logoColor=ffffff\" alt=\"数据集\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002FOpenResearcher\u002FOpenResearcher-30B-A3B\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FModel-FFD966?style=for-the-badge&logo=huggingface&logoColor=ffffff\" alt=\"模型\">\u003C\u002Fa>\n    \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fspaces\u002FOpenResearcher\u002FOpenResearcher\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FDemo-F97316.svg?style=for-the-badge&logo=gradio&logoColor=white\" alt=\"演示\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fx.com\u002Fzhuofengli96475\u002Fstatus\u002F2021682952074097086\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FVideo-CA4245?style=for-the-badge&logo=youtube&logoColor=white\" alt=\"视频\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FOpenResearcher\u002FOpenResearcher-Eval-Logs\u002Ftree\u002Fmain\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FEval%20Logs-755BB4?style=for-the-badge&logo=google-sheets&logoColor=white\" alt=\"评估日志\">\u003C\u002Fa>\n\u003C\u002Fdiv>\n\n\u003Cdiv align=\"center\">\n  \u003Cimg src=\"assets\u002Fimgs\u002Fnvidia_banner.svg\" alt=\"被 NVIDIA 的 Nemotron 系列模型采用！\">\n\u003C\u002Fdiv>\n\u003C\u002Fdiv>\n\u003Cbr>\n\u003Cp align=\"center\">\n  🤗 \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fcollections\u002FTIGER-Lab\u002Fopenresearcher\" target=\"_blank\">HuggingFace\u003C\u002Fa> ｜\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_0b0f1a7c995c.png\" width=\"14px\" style=\"display:inline;\"> \u003Ca href=\"https:\u002F\u002Fjoin.slack.com\u002Ft\u002Fopenresearcher\u002Fshared_invite\u002Fzt-3p0r32cky-PqtZkVjjWIAI14~XwcRMfQ\" target=\"_blank\">Slack\u003C\u002Fa> | \u003Cimg src=\"assets\u002Fimgs\u002Fwechat.svg\" width=\"14px\" style=\"display:inline;\"> \u003Ca href=\"assets\u002Fimgs\u002Fwechat_group.jpg\" target=\"_blank\">WeChat\u003C\u002Fa> \n\u003C\u002Fp>\n\n## 📣 新闻 \n+ **[2026.3.24]** 🔥 **OpenResearcher [论文](https:\u002F\u002Fhuggingface.co\u002Fpapers\u002F2603.20278) 现已发布**，强调了深度研究流程设计的实用见解。已被 **NVIDIA 的 Nemotron 系列模型**采用！\n+ **[2026.2.25]** 荣幸成为 🤗 [Hugging Face](https:\u002F\u002Fhuggingface.co\u002Fdatasets) 上**前 3 名趋势数据集**之一 —— 现在下载量达 **1.1 万+**！ \n+ **[2026.2.18]** OpenResearcher 训练 [代码](https:\u002F\u002Fgithub.com\u002FTIGER-AI-Lab\u002FOpenResearcher?tab=readme-ov-file#-optional-train-your-own-openresearcher) 现已可用。开始训练您自己的 OpenResearcher！\n+ **[2026.2.14]** 很高兴我们的 OpenResearcher 演示 [视频](https:\u002F\u002Fx.com\u002Fzhuofengli96475\u002Fstatus\u002F2021682952074097086) 上线。立即探索，解锁深度研究的强大力量！\n+ **[2026.2.12]** 很高兴看到 **OpenResearcher** 在 [**NVIDIA's NeMo Data Designer**](https:\u002F\u002Fnvidia-nemo.github.io\u002FDataDesigner\u002Flatest\u002Fdevnotes\u002Fdeep-research-trajectories-with-nemo-data-designer-and-mcp-tool-use\u002F) 中为深度研究轨迹生成提供动力！\n+ **[2026.2.10]** 我们的 X [帖子](https:\u002F\u002Fx.com\u002FDongfuJiang\u002Fstatus\u002F2020946549422031040) 获得了 **1.2K+ 点赞**! 欢迎查看帖子并参与讨论！ \n\n## 💥 简介\n\n**OpenResearcher** 是一个完全开源的 Agentic Large Language Model (智能体大语言模型) (30B-A3B)，专为**长周期深度研究**场景设计。它在 [BrowseComp-Plus](https:\u002F\u002Fhuggingface.co\u002Fspaces\u002FTevatron\u002FBrowseComp-Plus) 上取得了令人印象深刻的 **54.8%** 准确率，超越了 `GPT-4.1`, `Claude-Opus-4`, `Gemini-2.5-Pro`, `DeepSeek-R1` 和 `Tongyi-DeepResearch` 的性能。我们**完全开源**了训练和评估方案——包括数据、模型、训练方法和评估框架，供所有人推进深度研究。\n\u003Cdiv align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_1cb7576bd696.png\" alt=\"OpenResearcher 宣传图\" width=\"100%\" style=\"max-width: 850px; border-radius: 8px; box-shadow: 0 4px 10px rgba(0,0,0,0.1);\">\n\u003C\u002Fdiv>\n\n\u003Cbr>\n\n## 🏆 深度研究基准测试结果\n\n\u003Cdiv align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_af15b42ac583.png\" alt=\"深度研究基准测试结果\" width=\"100%\">\n\u003C\u002Fdiv>\n\n\n## ✨ 特性\n+ 🔑 **完全开源方案** — 我们完全开源了 96K 高质量 [DeepResearch 轨迹数据集](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FOpenResearcher\u002FOpenResearcher-Dataset)，该数据集包含由 GPT-OSS-120B 生成的 100+ 轮对话，使用了 [原生浏览器工具](https:\u002F\u002Fdocs.vllm.ai\u002Fprojects\u002Frecipes\u002Fen\u002Flatest\u002FOpenAI\u002FGPT-OSS.html#usage:~:text=Limitation%20section%20below.-,Tool%20Use,-%C2%B6)。基于此训练的领先 [30B-A3B 模型](https:\u002F\u002Fhuggingface.co\u002FOpenResearcher\u002FOpenResearcher-30B-A3B)、[蒸馏方案](https:\u002F\u002Farxiv.org\u002Fabs\u002F2603.20278) 以及轻量级 [DeepResearch 评估框架](https:\u002F\u002Fgithub.com\u002FTIGER-AI-Lab\u002FOpenResearcher) 均已开源，以推动深度研究发展。\n\n+ 💰 **高度可扩展且低成本** — 我们使用自研检索器在专用约 11B-token [语料库](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FOpenResearcher\u002FOpenResearcher-Corpus) 上大规模生成了 DeepResearch 轨迹，消除了对外部 Search API 的需求。这种可扩展的检索器显著降低了训练成本。\n\n+ 🚀 **在深度研究基准测试上表现卓越** — OpenResearcher 在一系列深度研究基准测试中展现出领先性能，包括 BrowseComp-Plus, BrowseComp, GAIA, xbench-DeepSearch。\n\n## 📋 目录\n\n- [🛠 环境设置](#-environment-setup)\n  - [安装](#installation)\n  - [深度研究基准准备](#deep-research-benchmarks-preparation)\n- [🔍 配置](#-configuration)\n- [🚀 快速开始](#-quick-start)\n- [🔬 评测 OpenResearcher](#-benchmark-openresearcher)\n  - [示例 1：使用本地搜索引擎的 BrowseComp-Plus](#example-1-browsecomp-plus-with-local-search-engine)\n  - [示例 2：使用 Serper API 的 GAIA（无需本地搜索）](#example-2-gaia-with-serper-api-no-local-search-needed)\n  - [评估](#evaluation)\n  - [快捷命令](#quick-commands)\n- [🧪 （可选）训练您自己的 OpenResearcher](#-optional-train-your-own-openresearcher)\n- [🤝 核心贡献者](#-core-contributors)\n- [🎓 顾问](#-advisors)\n- [🙏 致谢](#-acknowledgements)\n- [✨ 贡献](#-contributing)\n- [📚 引用](#-citation)\n## 🛠 环境设置\n我们在以下环境中运行此仓库：\n+ 8 * A100 80G Nvidia GPU\n+ Linux 操作系统\n\n其他硬件配置也可以工作，但请记住修改相应的参数。\n\n### 安装 \n```bash\nsudo apt update \nsudo apt install -y openjdk-21-jdk\n\n# install uv\ncurl -LsSf https:\u002F\u002Fastral.sh\u002Fuv\u002Finstall.sh | sh\nuv venv --python 3.12\nsource .venv\u002Fbin\u002Factivate\n\n# install tevatron for BrowseComp-plus \ngit clone https:\u002F\u002Fgithub.com\u002Ftexttron\u002Ftevatron.git\ncd tevatron\nuv pip install -e .\ncd ..\n\n# install all dependencies automatically\nuv pip install -e .\n```\n\n### 深度研究基准准备\n\n运行设置脚本以自动下载 **[BrowseComp-Plus](https:\u002F\u002Farxiv.org\u002Fabs\u002F2508.06600)** 基准。其他基准，包括 **[BrowseComp](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.12516)**、**[GAIA](https:\u002F\u002Farxiv.org\u002Fabs\u002F2311.12983)** 和 **[xbench-DeepResearch](https:\u002F\u002Fgithub.com\u002FTHUDM\u002Fxbench)**，将在首次使用时自动设置。\n\n```bash\nbash setup.sh\n```\n\n**此脚本将：**\n- ✅ 验证 Python 3.12 虚拟环境并自动安装任何缺失的依赖项\n- ✅ 从 HuggingFace 下载 BrowseComp-Plus 数据集并设置目录结构\n\n有关这些深度研究基准的更多信息，请参见 [benchmarks.md](assets\u002Fdocs\u002Fbenchmarks.md)。\n\n## 🔍 配置\n\n复制模板并配置您的 API (应用程序接口) 密钥：\n\n```bash\ncp .env.template .env\n```\n\n编辑 `.env`：\n```bash\n# Serper API (for web search when using browser_backend=serper)\nSERPER_API_KEY=your_key        # Get from: https:\u002F\u002Fserper.dev\u002F\n\n# OpenAI API (for evaluation scoring)\nOPENAI_API_KEY=your_key        # Get from: https:\u002F\u002Fplatform.openai.com\u002Fapi-keys\n```\n\n\n## 🚀 快速开始\n**前置条件：** 安装依赖项并配置 API 密钥（见 [环境设置](#-environment-setup) 和 [配置](#-configuration)）\n\n1. **部署 OpenResearcher-30B-A3B**：\n\n```bash\nbash scripts\u002Fstart_nemotron_servers.sh\n```\n\n完整的 vLLM (高性能推理服务器) 服务器日志可在 `logs` 目录中找到。\n\n2. **运行您的第一个任务**（在继续之前，请检查 `logs` 目录中的日志以确保 vLLM 服务器已部署。）\n\n```python\nimport asyncio\nfrom deploy_agent import run_one, BrowserPool\nfrom utils.openai_generator import OpenAIAsyncGenerator\n\nasync def main():\n    # Initialize generator and browser\n    generator = OpenAIAsyncGenerator(\n        base_url=\"http:\u002F\u002Flocalhost:8001\u002Fv1\",\n        model_name=\"OpenResearcher\u002FOpenResearcher-30B-A3B\",\n        use_native_tools=True\n    )\n    browser_pool = BrowserPool(search_url=None, browser_backend=\"serper\")\n\n    # Run deep research\n    await run_one(\n        question=\"What is the latest news about OpenAI?\",\n        qid=\"quick_start\",\n        generator=generator,\n        browser_pool=browser_pool,\n    )\n\n    browser_pool.cleanup(\"quick_start\")\n\nif __name__ == \"__main__\":\n    asyncio.run(main())\n```\n\n深度研究智能体 (agent) 将自动搜索网络、浏览网页并提取相关信息。您将看到最终答案以及所有中间推理步骤。\n\n\n## 🔬 OpenResearcher 基准测试\n我们使用以下深度研究基准对 OpenResearcher-30B-A3B 进行了基准测试： \n\n| 基准测试 | 数据集键 | 大小 | 语言 | 搜索后端 | 描述 |\n|-----------|-------------|------|----------|----------------|-------------|\n| [BrowseComp-Plus](https:\u002F\u002Farxiv.org\u002Fabs\u002F2508.06600) | `browsecomp_plus` | 830 | EN | local | 来自 BrowseComp 的深度研究基准，隔离了检索器 (retriever) 和 LLM (大语言模型) 智能体的影响 |\n| [BrowseComp](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.12516) | `browsecomp` | 1,266 | EN | serper | 面向浏览智能体的简单却具挑战性的基准 |\n| [GAIA-text](https:\u002F\u002Farxiv.org\u002Fabs\u002F2311.12983) | `gaia` | 103 | EN | serper | GAIA 基准的纯文本子集（开发集划分） |\n| [xbench-DeepResearch](https:\u002F\u002Fgithub.com\u002FTHUDM\u002Fxbench) | `xbench` | 100 | ZH | serper | 带有加密测试用例的 DeepSearch 基准 |\n\n有关这些深度研究基准的更多信息，请参见 [benchmarks.md](assets\u002Fdocs\u002Fbenchmarks.md)。 \n\n### 示例 1：使用本地搜索引擎的 BrowseComp-Plus\n\n使用本地稠密 (Dense) 搜索配合 browsecomp-plus [语料库 (corpus)](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FTevatron\u002Fbrowsecomp-plus-corpus) 和 [嵌入向量 (embeddings)](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FTevatron\u002Fbrowsecomp-plus-indexes\u002Ftree\u002Fmain\u002Fqwen3-embedding-8b) 进行完整评估（**注意：仅适用于 BrowseComp-Plus**）：\n\n```bash\n# Terminal 1: Start local Dense search service on port 8000\n# Embedding model (Qwen3-Embedding-8B) will be deployed on GPUs 7\nbash scripts\u002Fstart_search_service.sh dense 8000\n\n# Terminal 2: Start vLLM servers (requires 4 GPUs)\n# TP=2, deploy 2 servers starting from port 8001 on GPUs 0,1,2,3\nbash scripts\u002Fstart_nemotron_servers.sh 2 8001 0,1,2,3\n\n# Terminal 3: Run agent\nbash run_agent.sh results\u002Fbrowsecomp_plus\u002FOpenResearcher_dense 8001 2 browsecomp_plus local OpenResearcher\u002FOpenResearcher-30B-A3B\n```\n\n这做了什么：\n- 在端口 8000 上部署稠密检索器服务作为搜索引擎\n- 启动 2 个 vLLM 服务器（端口 8001, 8002），TP (张量并行)=2，跨 4 个 GPU 部署\n- 运行深度研究智能体，并在两个服务器之间进行负载均衡\n\n### 示例 2：使用 Serper API 的 GAIA（无需本地搜索）\n\n使用 Serper Google 搜索 API 运行（**注意：适用于除 BrowseComp-Plus 之外的所有基准**）：\n\n```bash\n# Terminal 1: Start vLLM servers (requires 4 GPUs)\nbash scripts\u002Fstart_nemotron_servers.sh 2 8001 0,1,2,3\n\n# Terminal 2: Run agent with serper search backend\nbash run_agent.sh results\u002Fgaia\u002FOpenResearcher_serper 8001 2 gaia serper OpenResearcher\u002FOpenResearcher-30B-A3B\n```\n\n**浏览器后端选项：**\n- `local` - 使用本地 BM25\u002F稠密搜索服务（用于 BrowseComp-Plus）\n- `serper` - 使用 Serper Google 搜索 API（用于其他所有基准）\n\n对于其他参数，请参考 [parameter.md](assets\u002Fdocs\u002Fparameter.md)。\n\n### 评估\n\n运行实验后，评估结果：\n\n```bash\n# eval on browsecomp_plus\npython eval.py --input_dir results\u002Fbrowsecomp_plus_dense\u002FOpenResearcher_dense\n\n# eval on gaia\npython eval.py --input_dir results\u002Fgaia\u002FOpenResearcher_serper\n```\n\n### 快速命令\n\n| 场景 | 命令 |\n|----------|---------|\n| BrowseComp-Plus (BM25) | `bash scripts\u002Fstart_search_service.sh bm25 8000` then `bash scripts\u002Fstart_nemotron_servers.sh 2 8001 0,1,2,3` then `bash run_agent.sh results\u002Fbrowsecomp-plus\u002FOpenResearcher_bm25 8001 2 browsecomp_plus local OpenResearcher\u002FOpenResearcher-30B-A3B` |\n| BrowseComp-Plus (Qwen3-8B 稠密嵌入 (Dense Embeddings)) | `bash scripts\u002Fstart_search_service.sh dense 8000` then `bash scripts\u002Fstart_nemotron_servers.sh 2 8001 0,1,2,3` then `bash run_agent.sh results\u002Fbrowsecomp-plus\u002FOpenResearcher_dense 8001 2 browsecomp-plus local OpenResearcher\u002FOpenResearcher-30B-A3B` |\n| BrowseComp | `bash scripts\u002Fstart_nemotron_servers.sh 2 8001 0,1,2,3` then `bash run_agent.sh results\u002Fbrowsecomp 8001 2 browsecomp serper OpenResearcher\u002FOpenResearcher-30B-A3B` |\n| GAIA | `bash scripts\u002Fstart_nemotron_servers.sh 2 8001 0,1,2,3` then `bash run_agent.sh results\u002Fgaia 8001 2 gaia serper OpenResearcher\u002FOpenResearcher-30B-A3B` |\n| xbench-DeepResearch | `bash scripts\u002Fstart_nemotron_servers.sh 2 8001 0,1,2,3` then `bash run_agent.sh results\u002Fxbench 8001 2 xbench serper OpenResearcher\u002FOpenResearcher-30B-A3B` |\n\n有关脚本参数说明，请参阅 [parameter.md](assets\u002Fdocs\u002Fparameter.md)。\n\n**注意：** 别忘了使用以下命令评估您的结果：  \n```bash\npython eval.py --input_dir [INPUT_DIR]\n```\n\n## 🧪（可选）训练您自己的 OpenResearcher\n\n我们的 [OpenResearcher-30B-A3B](https:\u002F\u002Fhuggingface.co\u002FOpenResearcher\u002FOpenResearcher-30B-A3B) 是基于 [openresearcher-dataset](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FOpenResearcher\u002FOpenResearcher-Dataset)，使用 [Megatron-LM](https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM) 训练的。要开始操作，请克隆 Megatron-LM 仓库的 openresearcher 分支：\n```\ngit clone -b openresearcher https:\u002F\u002Fgithub.com\u002Fjdf-prog\u002FMegatron-LM.git\n```\n然后，按照这里的训练指南来训练您自己的 OpenResearcher！\n\n## 🤝 核心贡献者\n\n\u003Ctable>\n\u003Ctr>\n    \u003Ctd align=\"center\">\n        \u003Ca href=\"https:\u002F\u002Fzhuofeng-li.github.io\u002F\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_9d0aa70f5ca3.png\" width=\"75px;\" alt=\"Zhuofeng Li\"\u002F>\n            \u003Cbr \u002F>\n            \u003Csub>\u003Cb>Zhuofeng Li\u003C\u002Fb>\u003C\u002Fsub>\n        \u003C\u002Fa>\n    \u003C\u002Ftd>\n        \u003Ctd align=\"center\">\n        \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fjdf-prog\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_51ff34d286e7.png\" width=\"75px;\" alt=\"Dongfu Jiang\"\u002F>\n            \u003Cbr \u002F>\n            \u003Csub>\u003Cb>Dongfu Jiang\u003C\u002Fb>\u003C\u002Fsub>\n        \u003C\u002Fa>\n    \u003C\u002Ftd>\n    \u003C\u002Ftd>\n        \u003Ctd align=\"center\">\n        \u003Ca href=\"https:\u002F\u002Fmxueguang.github.io\u002F\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_2a674cb4063e.jpg\" width=\"75px;\" alt=\"Xueguang\"\u002F>\n            \u003Cbr \u002F>\n            \u003Csub>\u003Cb>Xueguang Ma\u003C\u002Fb>\u003C\u002Fsub>\n        \u003C\u002Fa>\n    \u003C\u002Ftd>\n    \u003Ctd align=\"center\">\n        \u003Ca href=\"https:\u002F\u002Fisaacghx.github.io\u002Fabout\u002F\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_550e40ea0dd1.png\" width=\"75px;\" alt=\"Haoxiang Zhang\"\u002F>\n            \u003Cbr \u002F>\n            \u003Csub>\u003Cb>Haoxiang Zhang\u003C\u002Fb>\u003C\u002Fsub>\n        \u003C\u002Fa>\n    \u003C\u002Ftd>\n    \u003Ctd align=\"center\">\n        \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Ferenup\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_2a32ad203de7.png\" width=\"75px;\" alt=\"Ping Nie\"\u002F>\n            \u003Cbr \u002F>\n            \u003Csub>\u003Cb>Ping Nie\u003C\u002Fb>\u003C\u002Fsub>\n        \u003C\u002Fa>\n    \u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftable>\n\n## 🎓 导师\n\n\u003Ctable>\n\u003Ctr>\n      \u003Ctd align=\"center\">\n        \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fwenhuchen\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_1c524a3cdf75.png\" width=\"75px;\" alt=\"Wenhu Chen\"\u002F>\n            \u003Cbr \u002F>\n            \u003Csub>\u003Cb>Wenhu Chen\u003C\u002Fb>\u003C\u002Fsub>\n        \u003C\u002Fa>\n    \u003C\u002Ftd>\n    \u003Ctd align=\"center\">\n        \u003Ca href=\"https:\u002F\u002Fyuzhimanhua.github.io\u002F\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_349331e941f2.jpg\" width=\"75px;\" alt=\"Yu Zhang\"\u002F>\n            \u003Cbr \u002F>\n            \u003Csub>\u003Cb>Yu Zhang\u003C\u002Fb>\u003C\u002Fsub>\n        \u003C\u002Fa>\n    \u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftable>\n\n## 🙏 致谢\n\u003Cdiv align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_a6d53a348144.png\" alt=\"Deep Research Benchmark Results\" width=\"100%\">\n\u003C\u002Fdiv>\n\n\n## ✨ 贡献\n我们非常期待社区对 OpenResearcher 的开源贡献！如果您有兴趣参与贡献、协作或报告问题 (Issue)，请随时提交一个 Issue 或拉取请求 (PR)。您也可以通过 [zhuofengli12345@gmail.com](mailto:zhuofengli12345@gmail.com) 联系我们。\n\n我们也期待您的反馈和建议！\n\n##  📚 引用\n\n```bibtex\n@article{li2026openresearcher,\n  title={{OpenResearcher: A Fully Open Pipeline for Long-Horizon Deep Research Trajectory Synthesis}},\n  author={Li, Zhuofeng and Jiang, Dongfu and Ma, Xueguang and Zhang, Haoxiang and Nie, Ping and Zhang, Yuyu and Zou, Kai and Xie, Jianwen and Zhang, Yu and Chen, Wenhu},\n  journal={arXiv preprint arXiv:2603.20278},\n  year={2026}\n}\n```\n\n\n## ⭐ Star 历史记录\n[![Star History Chart](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_readme_8d97aedf14fd.png)](https:\u002F\u002Fwww.star-history.com\u002F#TIGER-AI-Lab\u002FOpenResearcher&type=date&legend=top-left)\n\n\u003Cp align=\"right\" style=\"font-size: 14px; margin-top: 20px;\">\n  \u003Ca href=\"#readme-top\" style=\"text-decoration: none; font-weight: bold;\">\n    ↑ Back to Top ↑\n  \u003C\u002Fa>\n\u003C\u002Fp>","# OpenResearcher 快速上手指南\n\n**OpenResearcher** 是一个专为长周期深度研究场景设计的开源智能体大语言模型（30B-A3B）。它支持自主搜索、浏览网页并提取信息，在多项深度研究基准测试中表现优异。\n\n## 1. 环境准备\n\n*   **操作系统**: Linux\n*   **硬件要求**: 推荐 8 * A100 80G Nvidia GPUs（其他配置需调整参数）\n*   **Python 版本**: 3.12+\n*   **依赖工具**: OpenJDK 21, uv (Python 包管理器)\n\n## 2. 安装与配置\n\n### 安装依赖\n请在终端执行以下命令初始化环境：\n\n```bash\nsudo apt update \nsudo apt install -y openjdk-21-jdk\n\n# 安装 uv\ncurl -LsSf https:\u002F\u002Fastral.sh\u002Fuv\u002Finstall.sh | sh\nuv venv --python 3.12\nsource .venv\u002Fbin\u002Factivate\n\n# 安装 tevatron (用于 BrowseComp-plus)\ngit clone https:\u002F\u002Fgithub.com\u002Ftexttron\u002Ftevatron.git\ncd tevatron\nuv pip install -e .\ncd ..\n\n# 安装所有项目依赖\nuv pip install -e .\n```\n\n### 准备数据集与基准\n运行设置脚本自动下载 Benchmark 数据集并创建目录结构：\n\n```bash\nbash setup.sh\n```\n\n### 配置环境变量\n复制模板文件并填入必要的 API Key：\n\n```bash\ncp .env.template .env\n```\n\n编辑 `.env` 文件，确保包含以下密钥（可从对应平台获取）：\n```bash\n# Serper API (用于网络搜索)\nSERPER_API_KEY=your_key\n\n# OpenAI API (用于评估打分)\nOPENAI_API_KEY=your_key\n```\n\n## 3. 快速运行示例\n\n### 启动服务\n首先部署 vLLM 服务器（请确保日志正常）：\n\n```bash\nbash scripts\u002Fstart_nemotron_servers.sh\n```\n\n### 运行任务\n创建一个 Python 脚本（例如 `quick_start.py`），内容如下：\n\n```python\nimport asyncio\nfrom deploy_agent import run_one, BrowserPool\nfrom utils.openai_generator import OpenAIAsyncGenerator\n\nasync def main():\n    # 初始化生成器和浏览器池\n    generator = OpenAIAsyncGenerator(\n        base_url=\"http:\u002F\u002Flocalhost:8001\u002Fv1\",\n        model_name=\"OpenResearcher\u002FOpenResearcher-30B-A3B\",\n        use_native_tools=True\n    )\n    browser_pool = BrowserPool(search_url=None, browser_backend=\"serper\")\n\n    # 运行深度研究任务\n    await run_one(\n        question=\"What is the latest news about OpenAI?\",\n        qid=\"quick_start\",\n        generator=generator,\n        browser_pool=browser_pool,\n    )\n\n    browser_pool.cleanup(\"quick_start\")\n\nif __name__ == \"__main__\":\n    asyncio.run(main())\n```\n\n执行脚本后，Agent 将自动搜索网络、浏览页面并输出最终答案及推理过程。","某科技公司的产品经理需要在三天内完成一份关于“生成式 AI 在医疗影像诊断领域最新进展”的深度行业分析报告。\n\n### 没有 OpenResearcher 时\n- 需要手动在 Google Scholar、PubMed 及多家科技媒体间反复切换，收集信息效率极低。\n- 难以追踪复杂的技术演进脉络，经常遗漏关键的学术论文或新兴竞品的最新动态。\n- 验证信息来源可靠性耗时费力，人工整理数据时容易出现格式不统一或引用错误。\n- 面对海量碎片化信息，很难快速提炼出有深度的核心观点，导致报告内容流于表面。\n\n### 使用 OpenResearcher 后\n- OpenResearcher 自动规划长周期研究路径，一次性跨平台抓取关键文献与实时新闻。\n- 智能梳理技术演进轨迹，精准定位到近半年的突破性研究成果并建立关联。\n- 内置验证机制确保引用来源可靠，直接输出结构清晰、逻辑连贯的调研初稿。\n- 大幅缩短调研周期，让团队能专注于策略制定而非基础资料搜集，提升决策效率。\n\nOpenResearcher 将深度调研从繁琐的手工劳动转变为高效的自动化流程，显著提升了专业报告产出的质量与速度。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FTIGER-AI-Lab_OpenResearcher_8af7fffe.png","TIGER-AI-Lab","TIGER Lab","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002FTIGER-AI-Lab_41df1eb9.jpg","Our lab is currently based in UWaterloo, focusing on Text and Image Generative Research",null,"wenhuchen@uwaterloo.ca","https:\u002F\u002Fgithub.com\u002FTIGER-AI-Lab",[83,87],{"name":84,"color":85,"percentage":86},"Python","#3572A5",88.1,{"name":88,"color":89,"percentage":90},"Shell","#89e051",11.9,618,65,"2026-04-05T07:30:20","Linux","需要 NVIDIA GPU，官方推荐 8 * A100 80G，CUDA 版本未说明","未说明",{"notes":98,"python":99,"dependencies":100},"1. 仅支持 Linux 系统；2. 必须安装 OpenJDK 21；3. 使用 uv 创建 Python 3.12 虚拟环境；4. 首次运行需执行 setup.sh 自动下载 BrowseComp-Plus 等数据集；5. 需配置 Serper 和 OpenAI API Key；6. 推理服务基于 vLLM 部署，需确保端口可用","3.12",[101,102,103,104,105],"uv","openjdk-21-jdk","tevatron","vllm","openai",[54,26,13],[108,109,110],"deep-research","llm","retrieval","2026-03-27T02:49:30.150509","2026-04-06T05:16:04.354016",[114],{"id":115,"question_zh":116,"answer_zh":117,"source_url":118},2455,"为什么在 LM Studio 中使用 OpenResearch 模型会生成完全随机的回复？它只能用于网络搜索吗？","为了获得最佳性能，请使用非量化版本的 OpenResearcher-30B-A3B 模型，并配合项目提供的配套架构（包含提示词、工具和智能体循环）。具体请参考 README 中的 🚀 Quick Start 部分获取配置指南。","https:\u002F\u002Fgithub.com\u002FTIGER-AI-Lab\u002FOpenResearcher\u002Fissues\u002F1",[]]