[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-sentient-agi--OpenDeepSearch":3,"tool-sentient-agi--OpenDeepSearch":61},[4,18,26,36,44,53],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":17},4358,"openclaw","openclaw\u002Fopenclaw","OpenClaw 是一款专为个人打造的本地化 AI 助手，旨在让你在自己的设备上拥有完全可控的智能伙伴。它打破了传统 AI 助手局限于特定网页或应用的束缚，能够直接接入你日常使用的各类通讯渠道，包括微信、WhatsApp、Telegram、Discord、iMessage 等数十种平台。无论你在哪个聊天软件中发送消息，OpenClaw 都能即时响应，甚至支持在 macOS、iOS 和 Android 设备上进行语音交互，并提供实时的画布渲染功能供你操控。\n\n这款工具主要解决了用户对数据隐私、响应速度以及“始终在线”体验的需求。通过将 AI 部署在本地，用户无需依赖云端服务即可享受快速、私密的智能辅助，真正实现了“你的数据，你做主”。其独特的技术亮点在于强大的网关架构，将控制平面与核心助手分离，确保跨平台通信的流畅性与扩展性。\n\nOpenClaw 非常适合希望构建个性化工作流的技术爱好者、开发者，以及注重隐私保护且不愿被单一生态绑定的普通用户。只要具备基础的终端操作能力（支持 macOS、Linux 及 Windows WSL2），即可通过简单的命令行引导完成部署。如果你渴望拥有一个懂你",349277,3,"2026-04-06T06:32:30",[13,14,15,16],"Agent","开发框架","图像","数据工具","ready",{"id":19,"name":20,"github_repo":21,"description_zh":22,"stars":23,"difficulty_score":10,"last_commit_at":24,"category_tags":25,"status":17},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,"2026-04-05T11:01:52",[14,15,13],{"id":27,"name":28,"github_repo":29,"description_zh":30,"stars":31,"difficulty_score":32,"last_commit_at":33,"category_tags":34,"status":17},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",150037,2,"2026-04-10T23:33:47",[14,13,35],"语言模型",{"id":37,"name":38,"github_repo":39,"description_zh":40,"stars":41,"difficulty_score":32,"last_commit_at":42,"category_tags":43,"status":17},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。",108322,"2026-04-10T11:39:34",[14,15,13],{"id":45,"name":46,"github_repo":47,"description_zh":48,"stars":49,"difficulty_score":32,"last_commit_at":50,"category_tags":51,"status":17},6121,"gemini-cli","google-gemini\u002Fgemini-cli","gemini-cli 是一款由谷歌推出的开源 AI 命令行工具，它将强大的 Gemini 大模型能力直接集成到用户的终端环境中。对于习惯在命令行工作的开发者而言，它提供了一条从输入提示词到获取模型响应的最短路径，无需切换窗口即可享受智能辅助。\n\n这款工具主要解决了开发过程中频繁上下文切换的痛点，让用户能在熟悉的终端界面内直接完成代码理解、生成、调试以及自动化运维任务。无论是查询大型代码库、根据草图生成应用，还是执行复杂的 Git 操作，gemini-cli 都能通过自然语言指令高效处理。\n\n它特别适合广大软件工程师、DevOps 人员及技术研究人员使用。其核心亮点包括支持高达 100 万 token 的超长上下文窗口，具备出色的逻辑推理能力；内置 Google 搜索、文件操作及 Shell 命令执行等实用工具；更独特的是，它支持 MCP（模型上下文协议），允许用户灵活扩展自定义集成，连接如图像生成等外部能力。此外，个人谷歌账号即可享受免费的额度支持，且项目基于 Apache 2.0 协议完全开源，是提升终端工作效率的理想助手。",100752,"2026-04-10T01:20:03",[52,13,15,14],"插件",{"id":54,"name":55,"github_repo":56,"description_zh":57,"stars":58,"difficulty_score":10,"last_commit_at":59,"category_tags":60,"status":17},4487,"LLMs-from-scratch","rasbt\u002FLLMs-from-scratch","LLMs-from-scratch 是一个基于 PyTorch 的开源教育项目，旨在引导用户从零开始一步步构建一个类似 ChatGPT 的大型语言模型（LLM）。它不仅是同名技术著作的官方代码库，更提供了一套完整的实践方案，涵盖模型开发、预训练及微调的全过程。\n\n该项目主要解决了大模型领域“黑盒化”的学习痛点。许多开发者虽能调用现成模型，却难以深入理解其内部架构与训练机制。通过亲手编写每一行核心代码，用户能够透彻掌握 Transformer 架构、注意力机制等关键原理，从而真正理解大模型是如何“思考”的。此外，项目还包含了加载大型预训练权重进行微调的代码，帮助用户将理论知识延伸至实际应用。\n\nLLMs-from-scratch 特别适合希望深入底层原理的 AI 开发者、研究人员以及计算机专业的学生。对于不满足于仅使用 API，而是渴望探究模型构建细节的技术人员而言，这是极佳的学习资源。其独特的技术亮点在于“循序渐进”的教学设计：将复杂的系统工程拆解为清晰的步骤，配合详细的图表与示例，让构建一个虽小但功能完备的大模型变得触手可及。无论你是想夯实理论基础，还是为未来研发更大规模的模型做准备",90106,"2026-04-06T11:19:32",[35,15,13,14],{"id":62,"github_repo":63,"name":64,"description_en":65,"description_zh":66,"ai_summary_zh":66,"readme_en":67,"readme_zh":68,"quickstart_zh":69,"use_case_zh":70,"hero_image_url":71,"owner_login":72,"owner_name":72,"owner_avatar_url":73,"owner_bio":74,"owner_company":75,"owner_location":75,"owner_email":75,"owner_twitter":75,"owner_website":75,"owner_url":76,"languages":77,"stars":82,"forks":83,"last_commit_at":84,"license":85,"difficulty_score":10,"env_os":86,"env_gpu":87,"env_ram":88,"env_deps":89,"category_tags":98,"github_topics":75,"view_count":32,"oss_zip_url":75,"oss_zip_packed_at":75,"status":17,"created_at":99,"updated_at":100,"faqs":101,"releases":102},6489,"sentient-agi\u002FOpenDeepSearch","OpenDeepSearch","SOTA search powered LLM","OpenDeepSearch 是一款旨在让顶级搜索能力普惠大众的开源项目。它通过结合先进的开源推理大模型与智能搜索代理，重新定义了信息检索的方式。传统搜索引擎往往只能返回简单的链接列表，用户需要自行筛选和整合信息，而 OpenDeepSearch 解决了这一痛点：它能像人类专家一样进行深度思考，自主规划搜索路径、多次检索并综合多方信息，最终直接给出经过逻辑推导的深度答案。\n\n这款工具特别适合开发者、AI 研究人员以及需要处理复杂信息查询的专业人士使用。对于希望构建下一代智能搜索应用的技术团队，或是需要从海量数据中快速提取精准洞察的研究者，OpenDeepSearch 提供了强大的底层支持。其核心亮点在于“推理驱动”，不仅依赖关键词匹配，更利用推理模型的思维链（Chain-of-Thought）能力，主动拆解复杂问题，动态调整搜索策略，从而实现远超传统工具的准确率与深度。作为一个完全开源的项目，它打破了技术壁垒，让每个人都能自由部署、定制属于自己的高性能深度搜索系统，推动搜索技术向更智能、更透明的方向发展。","# 🔍OpenDeepSearch: Democratizing Search with Open-source Reasoning Models and Reasoning Agents 🚀\n\n\u003C!-- markdownlint-disable first-line-h1 -->\n\u003C!-- markdownlint-disable html -->\n\u003C!-- markdownlint-disable no-duplicate-header -->\n\n\u003Cdiv align=\"center\">\n    \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fsentient-agi_OpenDeepSearch_readme_42ee58c66f8e.png\" alt=\"alt text\" width=\"60%\"\u002F>\n\u003C\u002Fdiv>\n\n\u003Chr>\n\u003Cdiv align=\"center\" style=\"line-height: 1;\">\n  \u003Ca href=\"https:\u002F\u002Fsentient.xyz\u002F\" target=\"_blank\" style=\"margin: 2px;\">\n    \u003Cimg alt=\"Homepage\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FSentient-Homepage-%23EAEAEA?logo=data%3Aimage%2Fsvg%2Bxml%3Bbase64%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%2BPC9zdmc%2B&link=https%3A%2F%2Fhuggingface.co%2FSentientagi\" style=\"display: inline-block; vertical-align: middle;\"\u002F>\n  \u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fsentient-agi\" target=\"_blank\" style=\"margin: 2px;\">\n    \u003Cimg alt=\"GitHub\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FGithub-sentient_agi-181717?logo=github\" style=\"display: inline-block; vertical-align: middle;\"\u002F>\n  \u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002FSentientagi\" target=\"_blank\" style=\"margin: 2px;\">\n    \u003Cimg alt=\"Hugging Face\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F%F0%9F%A4%97%20Hugging%20Face-SentientAGI-ffc107?color=ffc107&logoColor=white\" style=\"display: inline-block; vertical-align: middle;\"\u002F>\n  \u003C\u002Fa>\n\u003C\u002Fdiv>\n\n\u003Cdiv align=\"center\" style=\"line-height: 1;\">\n  \u003Ca href=\"https:\u002F\u002Fdiscord.gg\u002Fsentientfoundation\" target=\"_blank\" style=\"margin: 2px;\">\n    \u003Cimg alt=\"Discord\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FDiscord-SentientAGI-7289da?logo=discord&logoColor=white&color=7289da\" style=\"display: inline-block; vertical-align: middle;\"\u002F>\n  \u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fx.com\u002FSentientAGI\" target=\"_blank\" style=\"margin: 2px;\">\n    \u003Cimg alt=\"Twitter Follow\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F-SentientAGI-grey?logo=x&link=https%3A%2F%2Fx.com%2FSentientAGI%2F\" style=\"display: inline-block; vertical-align: middle;\"\u002F>\n  \u003C\u002Fa>\n\u003C\u002Fdiv>\n\n\u003Ch4 align=\"center\">\n        \u003Ca href=\"https:\u002F\u002Farxiv.org\u002Fpdf\u002F2503.20201\"> Paper  \u003C\u002Fa>\n\u003C\u002Fh4>\n\n## Description 📝\n\nOpenDeepSearch is a lightweight yet powerful search tool designed for seamless integration with AI agents. It enables deep web search and retrieval, optimized for use with Hugging Face's **[SmolAgents](https:\u002F\u002Fgithub.com\u002Fhuggingface\u002Fsmolagents)** ecosystem.\n\n\u003Cdiv align=\"center\">\n    \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fsentient-agi_OpenDeepSearch_readme_440021dcb222.png\" alt=\"Evaluation Results\" width=\"80%\"\u002F>\n\u003C\u002Fdiv>\n\n- **Performance**: ODS performs on par with closed source search alternatives on single-hop queries such as [SimpleQA](https:\u002F\u002Fopenai.com\u002Findex\u002Fintroducing-simpleqa\u002F) 🔍.\n- **Advanced Capabilities**: ODS performs much better than closed source search alternatives on multi-hop queries such as [FRAMES bench](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fgoogle\u002Fframes-benchmark) 🚀.\n\n## Table of Contents 📑\n\n- [🔍OpenDeepSearch: Democratizing Search with Open-source Reasoning Models and Reasoning Agents 🚀](#opendeepsearch-democratizing-search-with-open-source-reasoning-models-and-reasoning-agents-)\n  - [Description 📝](#description-)\n  - [Table of Contents 📑](#table-of-contents-)\n  - [Features ✨](#features-)\n  - [Installation 📚](#installation-)\n  - [Setup](#setup)\n  - [Usage ️](#usage-️)\n    - [Using OpenDeepSearch Standalone 🔍](#using-opendeepsearch-standalone-)\n    - [Running the Gradio Demo 🖥️](#running-the-gradio-demo-️)\n    - [Integrating with SmolAgents \\& LiteLLM 🤖⚙️](#integrating-with-smolagents--litellm-️)\n      - [](#)\n    - [ReAct agent with math and search tools 🤖⚙️](#react-agent-with-math-and-search-tools-️)\n      - [](#-1)\n  - [Search Modes 🔄](#search-modes-)\n    - [Default Mode ⚡](#default-mode-)\n    - [Pro Mode 🔍](#pro-mode-)\n  - [Acknowledgments 💡](#acknowledgments-)\n  - [Citation](#citation)\n  - [Contact 📩](#contact-)\n\n## Features ✨\n\n- **Semantic Search** 🧠: Leverages **[Crawl4AI](https:\u002F\u002Fgithub.com\u002Funclecode\u002Fcrawl4ai)** and semantic search rerankers (such as [Qwen2-7B-instruct](https:\u002F\u002Fhuggingface.co\u002FAlibaba-NLP\u002Fgte-Qwen2-7B-instruct\u002Ftree\u002Fmain) and [Jina AI](https:\u002F\u002Fjina.ai\u002F)) to provide in-depth results\n- **Two Modes of Operation** ⚡:\n  - **Default Mode**: Quick and efficient search with minimal latency.\n  - **Pro Mode (Deep Search)**: More in-depth and accurate results at the cost of additional processing time.\n- **Optimized for AI Agents** 🤖: Works seamlessly with **SmolAgents** like `CodeAgent`.\n- **Fast and Lightweight** ⚡: Designed for speed and efficiency with minimal setup.\n- **Extensible** 🔌: Easily configurable to work with different models and APIs.\n\n## Installation 📚\n\nTo install OpenDeepSearch, run:\n\n```bash\npip install -e . #you can also use: uv pip install -e .\npip install -r requirements.txt #you can also use: uv pip install -r requirements.txt\n```\n\nNote: you must have `torch` installed.\nNote: using `uv` instead of regular `pip` makes life much easier!\n\n### Using PDM (Alternative Package Manager) 📦\n\nYou can also use PDM as an alternative package manager for OpenDeepSearch. PDM is a modern Python package and dependency manager supporting the latest PEP standards.\n\n```bash\n# Install PDM if you haven't already\ncurl -sSL https:\u002F\u002Fraw.githubusercontent.com\u002Fpdm-project\u002Fpdm\u002Fmain\u002Finstall-pdm.py | python3 -\n\n# Initialize a new PDM project\npdm init\n\n# Install OpenDeepSearch and its dependencies\npdm install\n\n# Activate the virtual environment\neval \"$(pdm venv activate)\"\n```\n\nPDM offers several advantages:\n- Lockfile support for reproducible installations\n- PEP 582 support (no virtual environment needed)\n- Fast dependency resolution\n- Built-in virtual environment management\n\n## Setup\n\n1. **Choose a Search Provider**:\n   - **Option 1: Serper.dev**: Get **free 2500 credits** and add your API key.\n     - Visit [serper.dev](https:\u002F\u002Fserper.dev) to create an account.\n     - Retrieve your API key and store it as an environment variable:\n\n     ```bash\n     export SERPER_API_KEY='your-api-key-here'\n     ```\n\n   - **Option 2: SearXNG**: Use a self-hosted or public SearXNG instance.\n     - Specify the SearXNG instance URL when initializing OpenDeepSearch.\n     - Optionally provide an API key if your instance requires authentication:\n\n     ```bash\n     export SEARXNG_INSTANCE_URL='https:\u002F\u002Fyour-searxng-instance.com'\n     export SEARXNG_API_KEY='your-api-key-here'  # Optional\n     ```\n\n2. **Choose a Reranking Solution**:\n   - **Quick Start with Jina**: Sign up at [Jina AI](https:\u002F\u002Fjina.ai\u002F) to get an API key for immediate use\n   - **Self-hosted Option**: Set up [Infinity Embeddings](https:\u002F\u002Fgithub.com\u002Fmichaelfeil\u002Finfinity) server locally with open source models such as [Qwen2-7B-instruct](https:\u002F\u002Fhuggingface.co\u002FAlibaba-NLP\u002Fgte-Qwen2-7B-instruct\u002Ftree\u002Fmain)\n   - For more details on reranking options, see our [Rerankers Guide](src\u002Fopendeepsearch\u002Franking_models\u002FREADME.md)\n\n3. **Set up LiteLLM Provider**:\n   - Choose a provider from the [supported list](https:\u002F\u002Fdocs.litellm.ai\u002Fdocs\u002Fproviders\u002F), including:\n     - OpenAI\n     - Anthropic\n     - Google (Gemini)\n     - OpenRouter\n     - HuggingFace\n     - Fireworks\n     - And many more!\n   - Set your chosen provider's API key as an environment variable:\n   ```bash\n   export \u003CPROVIDER>_API_KEY='your-api-key-here'  # e.g., OPENAI_API_KEY, ANTHROPIC_API_KEY\n   ```\n   - For OpenAI, you can also set a custom base URL (useful for self-hosted endpoints or proxies):\n   ```bash\n   export OPENAI_BASE_URL='https:\u002F\u002Fyour-custom-openai-endpoint.com'\n   ```\n   - You can set default LiteLLM model IDs for different tasks:\n   ```bash\n   # General default model (fallback for all tasks)\n   export LITELLM_MODEL_ID='openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001'\n\n   # Task-specific models\n   export LITELLM_SEARCH_MODEL_ID='openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001'  # For search tasks\n   export LITELLM_ORCHESTRATOR_MODEL_ID='openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001'  # For agent orchestration\n   export LITELLM_EVAL_MODEL_ID='gpt-4o-mini'  # For evaluation tasks\n   ```\n   - When initializing OpenDeepSearch, you can specify your chosen model using the provider's format (this will override the environment variables):\n   ```python\n   search_agent = OpenDeepSearchTool(model_name=\"provider\u002Fmodel-name\")  # e.g., \"anthropic\u002Fclaude-3-opus-20240229\", 'huggingface\u002Fmicrosoft\u002Fcodebert-base', 'openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001'\n   ```\n\n## Usage ️\n\nYou can use OpenDeepSearch independently or integrate it with **SmolAgents** for enhanced reasoning and code generation capabilities.\n\n### Using OpenDeepSearch Standalone 🔍\n\n```python\nfrom opendeepsearch import OpenDeepSearchTool\nimport os\n\n# Set environment variables for API keys\nos.environ[\"SERPER_API_KEY\"] = \"your-serper-api-key-here\"  # If using Serper\n# Or for SearXNG\n# os.environ[\"SEARXNG_INSTANCE_URL\"] = \"https:\u002F\u002Fyour-searxng-instance.com\"\n# os.environ[\"SEARXNG_API_KEY\"] = \"your-api-key-here\"  # Optional\n\nos.environ[\"OPENROUTER_API_KEY\"] = \"your-openrouter-api-key-here\"\nos.environ[\"JINA_API_KEY\"] = \"your-jina-api-key-here\"\n\n# Using Serper (default)\nsearch_agent = OpenDeepSearchTool(\n    model_name=\"openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001\",\n    reranker=\"jina\"\n)\n\n# Or using SearXNG\n# search_agent = OpenDeepSearchTool(\n#     model_name=\"openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001\",\n#     reranker=\"jina\",\n#     search_provider=\"searxng\",\n#     searxng_instance_url=\"https:\u002F\u002Fyour-searxng-instance.com\",\n#     searxng_api_key=\"your-api-key-here\"  # Optional\n# )\n\nif not search_agent.is_initialized:\n    search_agent.setup()\n    \nquery = \"Fastest land animal?\"\nresult = search_agent.forward(query)\nprint(result)\n```\n\n### Running the Gradio Demo 🖥️\n\nTo try out OpenDeepSearch with a user-friendly interface, simply run:\n\n```bash\npython gradio_demo.py\n```\n\nThis will launch a local web interface where you can test different search queries and modes interactively.\n\nYou can customize the demo with command-line arguments:\n\n```bash\n# Using Serper (default)\npython gradio_demo.py --model-name \"openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001\" --reranker \"jina\"\n\n# Using SearXNG\npython gradio_demo.py --model-name \"openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001\" --reranker \"jina\" \\\n  --search-provider \"searxng\" --searxng-instance \"https:\u002F\u002Fyour-searxng-instance.com\" \\\n  --searxng-api-key \"your-api-key-here\"  # Optional\n```\n\nAvailable options:\n- `--model-name`: LLM model to use for search\n- `--orchestrator-model`: LLM model for the agent orchestrator\n- `--reranker`: Reranker to use (`jina` or `infinity`)\n- `--search-provider`: Search provider to use (`serper` or `searxng`)\n- `--searxng-instance`: SearXNG instance URL (required if using `searxng`)\n- `--searxng-api-key`: SearXNG API key (optional)\n- `--serper-api-key`: Serper API key (optional, will use environment variable if not provided)\n- `--openai-base-url`: OpenAI API base URL (optional, will use OPENAI_BASE_URL env var if not provided)\n\n### Integrating with SmolAgents & LiteLLM 🤖⚙️\n\n####\n\n```python\nfrom opendeepsearch import OpenDeepSearchTool\nfrom smolagents import CodeAgent, LiteLLMModel\nimport os\n\n# Set environment variables for API keys\nos.environ[\"SERPER_API_KEY\"] = \"your-serper-api-key-here\"  # If using Serper\n# Or for SearXNG\n# os.environ[\"SEARXNG_INSTANCE_URL\"] = \"https:\u002F\u002Fyour-searxng-instance.com\"\n# os.environ[\"SEARXNG_API_KEY\"] = \"your-api-key-here\"  # Optional\n\nos.environ[\"OPENROUTER_API_KEY\"] = \"your-openrouter-api-key-here\"\nos.environ[\"JINA_API_KEY\"] = \"your-jina-api-key-here\"\n\n# Using Serper (default)\nsearch_agent = OpenDeepSearchTool(\n    model_name=\"openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001\",\n    reranker=\"jina\"\n)\n\n# Or using SearXNG\n# search_agent = OpenDeepSearchTool(\n#     model_name=\"openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001\",\n#     reranker=\"jina\",\n#     search_provider=\"searxng\",\n#     searxng_instance_url=\"https:\u002F\u002Fyour-searxng-instance.com\",\n#     searxng_api_key=\"your-api-key-here\"  # Optional\n# )\n\nmodel = LiteLLMModel(\n    \"openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001\",\n    temperature=0.2\n)\n\ncode_agent = CodeAgent(tools=[search_agent], model=model)\nquery = \"How long would a cheetah at full speed take to run the length of Pont Alexandre III?\"\nresult = code_agent.run(query)\n\nprint(result)\n```\n### ReAct agent with math and search tools 🤖⚙️\n\n####\n```python\nfrom opendeepsearch import OpenDeepSearchTool\nfrom opendeepsearch.wolfram_tool import WolframAlphaTool\nfrom opendeepsearch.prompts import REACT_PROMPT\nfrom smolagents import LiteLLMModel, ToolCallingAgent, Tool\nimport os\n\n# Set environment variables for API keys\nos.environ[\"SERPER_API_KEY\"] = \"your-serper-api-key-here\"\nos.environ[\"JINA_API_KEY\"] = \"your-jina-api-key-here\"\nos.environ[\"WOLFRAM_ALPHA_APP_ID\"] = \"your-wolfram-alpha-app-id-here\"\nos.environ[\"FIREWORKS_API_KEY\"] = \"your-fireworks-api-key-here\"\n\nmodel = LiteLLMModel(\n    \"fireworks_ai\u002Fllama-v3p1-70b-instruct\",  # Your Fireworks Deepseek model\n    temperature=0.7\n)\nsearch_agent = OpenDeepSearchTool(model_name=\"fireworks_ai\u002Fllama-v3p1-70b-instruct\", reranker=\"jina\") # Set reranker to \"jina\" or \"infinity\"\n\n# Initialize the Wolfram Alpha tool\nwolfram_tool = WolframAlphaTool(app_id=os.environ[\"WOLFRAM_ALPHA_APP_ID\"])\n\n# Initialize the React Agent with search and wolfram tools\nreact_agent = ToolCallingAgent(\n    tools=[search_agent, wolfram_tool],\n    model=model,\n    prompt_templates=REACT_PROMPT # Using REACT_PROMPT as system prompt\n)\n\n# Example query for the React Agent\nquery = \"What is the distance, in metres, between the Colosseum in Rome and the Rialto bridge in Venice\"\nresult = react_agent.run(query)\n\nprint(result)\n```\n\n## Search Modes 🔄\n\nOpenDeepSearch offers two distinct search modes to balance between speed and depth:\n\n### Default Mode ⚡\n- Uses SERP-based interaction for quick results\n- Minimal processing overhead\n- Ideal for single-hop, straightforward queries\n- Fast response times\n- Perfect for basic information retrieval\n\n### Pro Mode 🔍\n- Involves comprehensive web scraping\n- Implements semantic reranking of results\n- Includes advanced post-processing of data\n- Slightly longer processing time\n- Excels at:\n  - Multi-hop queries\n  - Complex search requirements\n  - Detailed information gathering\n  - Questions requiring cross-reference verification\n\n## Acknowledgments 💡\n\nOpenDeepSearch is built on the shoulders of great open-source projects:\n\n- **[SmolAgents](https:\u002F\u002Fhuggingface.co\u002Fdocs\u002Fsmolagents\u002Findex)** 🤗 – Powers the agent framework and reasoning capabilities.\n- **[Crawl4AI](https:\u002F\u002Fgithub.com\u002Funclecode\u002Fcrawl4ai)** 🕷️ – Provides data crawling support.\n- **[Infinity Embedding API](https:\u002F\u002Fgithub.com\u002Fmichaelfeil\u002Finfinity)** 🌍 – Powers semantic search capabilities.\n- **[LiteLLM](https:\u002F\u002Fwww.litellm.ai\u002F)** 🔥 – Used for efficient AI model integration.\n- **Various Open-Source Libraries** 📚 – Enhancing search and retrieval functionalities.\n\n## Citation\n\nIf you use `OpenDeepSearch` in your works, please cite it using the following BibTex entry:\n\n```\n@misc{alzubi2025opendeepsearchdemocratizing,\n      title={Open Deep Search: Democratizing Search with Open-source Reasoning Agents},\n      author={Salaheddin Alzubi and Creston Brooks and Purva Chiniya and Edoardo Contente and Chiara von Gerlach and Lucas Irwin and Yihan Jiang and Arda Kaz and Windsor Nguyen and Sewoong Oh and Himanshu Tyagi and Pramod Viswanath},\n      year={2025},\n      eprint={2503.20201},\n      archivePrefix={arXiv},\n      primaryClass={cs.LG},\n      url={https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.20201},\n}\n```\n\n\n## Contact 📩\n\nFor questions or collaborations, open an issue or reach out to the maintainers.\n","# 🔍OpenDeepSearch：以开源推理模型和推理代理 democratize 搜索 🚀\n\n\u003C!-- markdownlint-disable first-line-h1 -->\n\u003C!-- markdownlint-disable html -->\n\u003C!-- markdownlint-disable no-duplicate-header -->\n\n\u003Cdiv align=\"center\">\n    \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fsentient-agi_OpenDeepSearch_readme_42ee58c66f8e.png\" alt=\"alt text\" width=\"60%\"\u002F>\n\u003C\u002Fdiv>\n\n\u003Chr>\n\u003Cdiv align=\"center\" style=\"line-height: 1;\">\n  \u003Ca href=\"https:\u002F\u002Fsentient.xyz\u002F\" target=\"_blank\" style=\"margin: 2px;\">\n    \u003Cimg alt=\"Homepage\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FSentient-Homepage-%23EAEAEA?logo=data%3Aimage%2Fsvg%2Bxml%3Bbase64%2CPHN2ZyB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciIHdpZHRoPSIzNDEuMzMzIiBoZWlnaHQ9IjM0MS4zMzMiIHZlcnNpb249IjEuMCIgdmlld0JveD0iMCAwIDI1NiAyNTYiPjxwYXRoIGQ9Ik0xMzIuNSAyOC40Yy0xLjUgMi4yLTEuMiAzLjkgNC45IDI3LjIgMy41IDEzLjcgOC41IDMzIDExLjEgNDIuOSAyLjYgOS.9IDUuMyAxOC42IDYgMTkuNCAzLjIgMy4zIDExLjctLjggMTMuLTYuQuNS0xLjktMTuuu.1-LT72-LT19.7-LT78.6-LT1.2-LT3-LT7.5-LT6.9-LT11.3-LT6.9-LT1.6-LT0-LT3.1-LT9-LT4.1-LT2.4zk0x110-M30-c1.1L1.1-2L3.1-2L4.5z0x110-M30-c1.1L1.1-2L3.1-2L4.5z0x110-M30-c1.1L1.1-2L3.1-2L4.5z0x110-M30-c1.1L1.1-2L3.1-2L4.5z0x110-M30-c1.1L1.1-2L3.1-2L4.5z0x110-M30-c1.1L1.1-2L3.1-2L4.5z0x110-M30-c1.1L1.1-2L3.1-2L4.5z0x110-M30-c1.1L1.1-2L3.1-2L4.5z0x110-M30-c1.1L1.1-2L3.1-2L4.5z0x110-M30-c1.1L1.1-2L3.1-2L......# 🔍OpenDeepSearch：以开源推理模型和推理代理 democratize 搜索 🚀\n\n\u003C!-- markdownlint-disable first-line-h1 -->\n\u003C!-- markdownlint-disable html -->\n\u003C!-- markdownlint-disable no-duplicate-header -->\n\n\u003Cdiv align=\"center\">\n    \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fsentient-agi_OpenDeepSearch_readme_42ee58c66f8e.png\" alt=\"alt text\" width=\"60%\"\u002F>\n\u003C\u002Fdiv>\n\n\u003Chr>\n\u003Cdiv align=\"center\" style=\"line-height: 1;\">\n  \u003Ca href=\"https:\u002F\u002Fsentient.xyz\u002F\" target=\"_blank\" style=\"margin: 2px;\">\n    \u003Cimg alt=\"Homepage\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FSentient-Homepage-%23EAEAEA?logo=data%3Aimage%2Fsvg%2Bxml%3Bbase64%2CPHN2ZyB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciIHdpZHRoPSIzNDEuMzMzIiBoZWlnaHQ9IjM0MS4zMzMiIHZlcnNpb249IjEuMCIgdmlld0JveD0iMCAwIDI1NiAyNTYiPjxwYXRoIGQ9Ik0xMzIuNSAyOC40Yy0xLjUgMi4yLTEuMiAzLjkgNC45IDI3LjIgMy41IDEzLjcgOC41IDMzIDExLjEgNDIuOSAyLjYgOS.9IDUuMyAxOCu6IDYgMTkuNCAzLjIgMy4zIDExLjctLjggMTMuLTYuQu5-LjktMTuuISu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7-MTgu7......\n\n## 目录 📑\n\n- [🔍OpenDeepSearch：通过开源推理模型和推理代理 democratize 搜索 🚀](#opendeepsearch-democratizing-search-with-open-source-reasoning-models-and-reasoning-agents-)\n  - [描述 📝](#description-)\n  - [目录 📑](#table-of-contents-)\n  - [特性 ✨](#features-)\n  - [安装 📚](#installation-)\n  - [设置](#setup)\n  - [使用 ️](#usage-️)\n    - [独立使用 OpenDeepSearch 🔍](#using-opendeepsearch-standalone-)\n    - [运行 Gradio 演示 🖥️](#running-the-gradio-demo-️)\n    - [与 SmolAgents 和 LiteLLM 集成 🤖⚙️](#integrating-with-smolagents--litellm-️)\n      - [](#)\n    - [带有数学和搜索工具的 ReAct 代理 🤖⚙️](#react-agent-with-math-and-search-tools-️)\n      - [](#-1)\n  - [搜索模式 🔄](#search-modes-)\n    - [默认模式 ⚡](#default-mode-)\n    - [专业模式 🔍](#pro-mode-)\n  - [致谢 💡](#acknowledgments-)\n  - [引用](#citation)\n  - [联系方式 📩](#contact-)\n\n## 特性 ✨\n\n- **语义搜索** 🧠：利用 **[Crawl4AI](https:\u002F\u002Fgithub.com\u002Funclecode\u002Fcrawl4ai)** 和语义搜索重排序器（如 [Qwen2-7B-instruct](https:\u002F\u002Fhuggingface.co\u002FAlibaba-NLP\u002Fgte-Qwen2-7B-instruct\u002Ftree\u002Fmain) 和 [Jina AI](https:\u002F\u002Fjina.ai\u002F)），提供深入的结果\n- **两种运行模式** ⚡：\n  - **默认模式**：快速高效的搜索，延迟极低。\n  - **专业模式（深度搜索）**：以额外的处理时间为代价，获得更深入、更准确的结果。\n- **专为 AI 代理优化** 🤖：可与 `CodeAgent` 等 **SmolAgents** 无缝协作。\n- **快速轻量** ⚡：设计注重速度与效率，设置简单。\n- **可扩展** 🔌：易于配置，可与不同模型和 API 配合使用。\n\n## 安装 📚\n\n要安装 OpenDeepSearch，请运行：\n\n```bash\npip install -e . #你也可以使用：uv pip install -e .\npip install -r requirements.txt #你也可以使用：uv pip install -r requirements.txt\n```\n\n注意：必须安装 `torch`。\n注意：使用 `uv` 而不是普通的 `pip` 会让操作更加轻松！\n\n### 使用 PDM（替代包管理器）📦\n\n你也可以使用 PDM 作为 OpenDeepSearch 的替代包管理器。PDM 是一个现代化的 Python 包和依赖项管理器，支持最新的 PEP 标准。\n\n```bash\n# 如果尚未安装 PDM，请先安装\ncurl -sSL https:\u002F\u002Fraw.githubusercontent.com\u002Fpdm-project\u002Fpdm\u002Fmain\u002Finstall-pdm.py | python3 -\n\n# 初始化一个新的 PDM 项目\npdm init\n\n# 安装 OpenDeepSearch 及其依赖项\npdm install\n\n# 激活虚拟环境\neval \"$(pdm venv activate)\"\n```\n\nPDM 具有以下优势：\n- 锁文件支持，实现可重复的安装\n- 支持 PEP 582（无需虚拟环境）\n- 快速的依赖解析\n- 内置虚拟环境管理\n\n## 设置\n\n1. **选择搜索提供商**：\n   - **选项 1：Serper.dev**：获取 **免费 2500 积分**，并添加你的 API 密钥。\n     - 访问 [serper.dev](https:\u002F\u002Fserper.dev) 创建账户。\n     - 获取你的 API 密钥，并将其存储为环境变量：\n\n     ```bash\n     export SERPER_API_KEY='your-api-key-here'\n     ```\n\n   - **选项 2：SearXNG**：使用自托管或公共 SearXNG 实例。\n     - 在初始化 OpenDeepSearch 时指定 SearXNG 实例的 URL。\n     - 如果你的实例需要认证，可选择提供 API 密钥：\n\n     ```bash\n     export SEARXNG_INSTANCE_URL='https:\u002F\u002Fyour-searxng-instance.com'\n     export SEARXNG_API_KEY='your-api-key-here'  # 可选\n     ```\n\n2. **选择重排序解决方案**：\n   - **快速开始使用 Jina**：在 [Jina AI](https:\u002F\u002Fjina.ai\u002F) 注册以获取立即可用的 API 密钥\n   - **自托管选项**：在本地设置 [Infinity Embeddings](https:\u002F\u002Fgithub.com\u002Fmichaelfeil\u002Finfinity) 服务器，使用开源模型，例如 [Qwen2-7B-instruct](https:\u002F\u002Fhuggingface.co\u002FAlibaba-NLP\u002Fgte-Qwen2-7B-instruct\u002Ftree\u002Fmain)\n   - 更多关于重排序选项的详细信息，请参阅我们的 [重排序指南](src\u002Fopendeepsearch\u002Franking_models\u002FREADME.md)\n\n3. **设置 LiteLLM 提供商**：\n   - 从 [支持列表](https:\u002F\u002Fdocs.litellm.ai\u002Fdocs\u002Fproviders\u002F) 中选择一个提供商，包括：\n     - OpenAI\n     - Anthropic\n     - Google (Gemini)\n     - OpenRouter\n     - HuggingFace\n     - Fireworks\n     - 以及更多！\n   - 将你选择的提供商的 API 密钥设置为环境变量：\n   ```bash\n   export \u003CPROVIDER>_API_KEY='your-api-key-here'  # 例如，OPENAI_API_KEY、ANTHROPIC_API_KEY\n   ```\n   - 对于 OpenAI，你还可以设置自定义的基础 URL（适用于自托管端点或代理）：\n   ```bash\n   export OPENAI_BASE_URL='https:\u002F\u002Fyour-custom-openai-endpoint.com'\n   ```\n   - 你可以为不同任务设置默认的 LiteLLM 模型 ID：\n   ```bash\n   # 通用默认模型（所有任务的回退）\n   export LITELLM_MODEL_ID='openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001'\n\n   # 任务特定模型\n   export LITELLM_SEARCH_MODEL_ID='openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001'  # 用于搜索任务\n   export LITELLM_ORCHESTRATOR_MODEL_ID='openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001'  # 用于代理编排\n   export LITELLM_EVAL_MODEL_ID='gpt-4o-mini'  # 用于评估任务\n   ```\n   - 在初始化 OpenDeepSearch 时，你可以使用提供商的格式指定你选择的模型（这将覆盖环境变量）：\n   ```python\n   search_agent = OpenDeepSearchTool(model_name=\"provider\u002Fmodel-name\")  # 例如，“anthropic\u002Fclaude-3-opus-20240229”、“huggingface\u002Fmicrosoft\u002Fcodebert-base”、“openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001”\n   ```\n\n## 使用 ️\n\n你可以独立使用 OpenDeepSearch，也可以将其与 **SmolAgents** 集成，以增强推理和代码生成能力。\n\n### 独立使用 OpenDeepSearch 🔍\n\n```python\nfrom opendeepsearch import OpenDeepSearchTool\nimport os\n\n# 设置 API 密钥的环境变量\nos.environ[\"SERPER_API_KEY\"] = \"your-serper-api-key-here\"  # 如果使用 Serper\n# 或者对于 SearXNG\n# os.environ[\"SEARXNG_INSTANCE_URL\"] = \"https:\u002F\u002Fyour-searxng-instance.com\"\n# os.environ[\"SEARXNG_API_KEY\"] = \"your-api-key-here\"  # 可选\n\nos.environ[\"OPENROUTER_API_KEY\"] = \"your-openrouter-api-key-here\"\nos.environ[\"JINA_API_KEY\"] = \"your-jina-api-key-here\"\n\n# 使用 Serper（默认）\nsearch_agent = OpenDeepSearchTool(\n    model_name=\"openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001\",\n    reranker=\"jina\"\n)\n\n# 或者使用 SearXNG\n# search_agent = OpenDeepSearchTool(\n#     model_name=\"openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001\",\n#     reranker=\"jina\",\n#     search_provider=\"searxng\",\n#     searxng_instance_url=\"https:\u002F\u002Fyour-searxng-instance.com\",\n#     searxng_api_key=\"your-api-key-here\"  # 可选\n# )\n\nif not search_agent.is_initialized:\n    search_agent.setup()\n    \nquery = \"最快的陆地动物是什么？\"\nresult = search_agent.forward(query)\nprint(result)\n```\n\n### 运行 Gradio 演示 🖥️\n\n要通过友好的用户界面体验 OpenDeepSearch，只需运行以下命令：\n\n```bash\npython gradio_demo.py\n```\n\n这将启动一个本地 Web 界面，您可以在其中交互式地测试不同的搜索查询和模式。\n\n您可以通过命令行参数自定义演示：\n\n```bash\n# 使用 Serper（默认）\npython gradio_demo.py --model-name \"openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001\" --reranker \"jina\"\n\n# 使用 SearXNG\npython gradio_demo.py --model-name \"openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001\" --reranker \"jina\" \\\n  --search-provider \"searxng\" --searxng-instance \"https:\u002F\u002Fyour-searxng-instance.com\" \\\n  --searxng-api-key \"your-api-key-here\"  # 可选\n```\n\n可用选项：\n- `--model-name`: 用于搜索的 LLM 模型\n- `--orchestrator-model`: 用于代理编排器的 LLM 模型\n- `--reranker`: 要使用的重排序器（`jina` 或 `infinity`）\n- `--search-provider`: 要使用的搜索引擎（`serper` 或 `searxng`）\n- `--searxng-instance`: SearXNG 实例 URL（使用 `searxng` 时必填）\n- `--searxng-api-key`: SearXNG API 密钥（可选）\n- `--serper-api-key`: Serper API 密钥（可选，未提供时将使用环境变量）\n- `--openai-base-url`: OpenAI API 基础 URL（可选，未提供时将使用 `OPENAI_BASE_URL` 环境变量）\n\n### 与 SmolAgents 和 LiteLLM 集成 🤖⚙️\n\n####\n\n```python\nfrom opendeepsearch import OpenDeepSearchTool\nfrom smolagents import CodeAgent, LiteLLMModel\nimport os\n\n# 设置 API 密钥环境变量\nos.environ[\"SERPER_API_KEY\"] = \"your-serper-api-key-here\"  # 如果使用 Serper\n# 或对于 SearXNG\n# os.environ[\"SEARXNG_INSTANCE_URL\"] = \"https:\u002F\u002Fyour-searxng-instance.com\"\n# os.environ[\"SEARXNG_API_KEY\"] = \"your-api-key-here\"  # 可选\n\nos.environ[\"OPENROUTER_API_KEY\"] = \"your-openrouter-api-key-here\"\nos.environ[\"JINA_API_KEY\"] = \"your-jina-api-key-here\"\n\n# 使用 Serper（默认）\nsearch_agent = OpenDeepSearchTool(\n    model_name=\"openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001\",\n    reranker=\"jina\"\n)\n\n# 或者使用 SearXNG\n# search_agent = OpenDeepSearchTool(\n#     model_name=\"openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001\",\n#     reranker=\"jina\",\n#     search_provider=\"searxng\",\n#     searxng_instance_url=\"https:\u002F\u002Fyour-searxng-instance.com\",\n#     searxng_api_key=\"your-api-key-here\"  # 可选\n# )\n\nmodel = LiteLLMModel(\n    \"openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001\",\n    temperature=0.2\n)\n\ncode_agent = CodeAgent(tools=[search_agent], model=model)\nquery = \"猎豹以全速奔跑，需要多长时间才能跑完亚历山大三世桥的长度？\"\nresult = code_agent.run(query)\n\nprint(result)\n```\n### 带有数学和搜索工具的 ReAct 代理 🤖⚙️\n\n####\n```python\nfrom opendeepsearch import OpenDeepSearchTool\nfrom opendeepsearch.wolfram_tool import WolframAlphaTool\nfrom opendeepsearch.prompts import REACT_PROMPT\nfrom smolagents import LiteLLMModel, ToolCallingAgent, Tool\nimport os\n\n# 设置 API 密钥环境变量\nos.environ[\"SERPER_API_KEY\"] = \"your-serper-api-key-here\"\nos.environ[\"JINA_API_KEY\"] = \"your-jina-api-key-here\"\nos.environ[\"WOLFRAM_ALPHA_APP_ID\"] = \"your-wolfram-alpha-app-id-here\"\nos.environ[\"FIREWORKS_API_KEY\"] = \"your-fireworks-api-key-here\"\n\nmodel = LiteLLMModel(\n    \"fireworks_ai\u002Fllama-v3p1-70b-instruct\",  # 您的 Fireworks Deepseek 模型\n    temperature=0.7\n)\nsearch_agent = OpenDeepSearchTool(model_name=\"fireworks_ai\u002Fllama-v3p1-70b-instruct\", reranker=\"jina\") # 将重排序器设置为“jina”或“infinity”\n\n# 初始化 Wolfram Alpha 工具\nwolfram_tool = WolframAlphaTool(app_id=os.environ[\"WOLFRAM_ALPHA_APP_ID\"])\n\n# 使用搜索和 Wolfram 工具初始化 React 代理\nreact_agent = ToolCallingAgent(\n    tools=[search_agent, wolfram_tool],\n    model=model,\n    prompt_templates=REACT_PROMPT # 使用 REACT_PROMPT 作为系统提示\n)\n\n# React 代理的示例查询\nquery = \"罗马斗兽场与威尼斯里亚托桥之间的距离是多少米？\"\nresult = react_agent.run(query)\n\nprint(result)\n```\n\n## 搜索模式 🔄\n\nOpenDeepSearch 提供两种不同的搜索模式，以在速度和深度之间取得平衡：\n\n### 默认模式 ⚡\n- 使用基于 SERP 的交互快速获取结果\n- 处理开销最小\n- 适用于单跳、简单的查询\n- 响应时间快\n- 非常适合基础信息检索\n\n### 专业模式 🔍\n- 包含全面的网页抓取\n- 实现结果的语义重排序\n- 包括数据的高级后处理\n- 处理时间稍长\n- 在以下方面表现出色：\n  - 多跳查询\n  - 复杂的搜索需求\n  - 详细的信息收集\n  - 需要交叉验证的问题\n\n## 致谢 💡\n\nOpenDeepSearch 建立在众多优秀的开源项目之上：\n\n- **[SmolAgents](https:\u002F\u002Fhuggingface.co\u002Fdocs\u002Fsmolagents\u002Findex)** 🤗 – 提供代理框架和推理能力。\n- **[Crawl4AI](https:\u002F\u002Fgithub.com\u002Funclecode\u002Fcrawl4ai)** 🕷️ – 提供数据爬取支持。\n- **[Infinity Embedding API](https:\u002F\u002Fgithub.com\u002Fmichaelfeil\u002Finfinity)** 🌍 – 提供语义搜索功能。\n- **[LiteLLM](https:\u002F\u002Fwww.litellm.ai\u002F)** 🔥 – 用于高效的 AI 模型集成。\n- **各种开源库** 📚 – 增强搜索和检索功能。\n\n## 引用\n\n如果您在工作中使用了 `OpenDeepSearch`，请使用以下 BibTex 条目进行引用：\n\n```\n@misc{alzubi2025opendeepsearchdemocratizing,\n      title={Open Deep Search: 以开源推理代理民主化搜索},\n      author={Salaheddin Alzubi 和 Creston Brooks 和 Purva Chiniya 和 Edoardo Contente 和 Chiara von Gerlach 和 Lucas Irwin 和 Yihan Jiang 和 Arda Kaz 和 Windsor Nguyen 和 Sewoong Oh 和 Himanshu Tyagi 和 Pramod Viswanath},\n      year={2025},\n      eprint={2503.20201},\n      archivePrefix={arXiv},\n      primaryClass={cs.LG},\n      url={https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.20201},\n}\n```\n\n\n## 联系方式 📩\n\n如有任何问题或合作意向，请提交 issue 或直接联系维护人员。","# OpenDeepSearch 快速上手指南\n\nOpenDeepSearch 是一款轻量级但功能强大的开源搜索工具，专为与 AI Agent（特别是 Hugging Face 的 **SmolAgents** 生态）无缝集成而设计。它支持深度网络搜索和检索，在单跳和多跳查询任务中表现优异。\n\n## 环境准备\n\n在开始之前，请确保您的开发环境满足以下要求：\n\n*   **操作系统**: Linux, macOS 或 Windows (WSL 推荐)\n*   **Python 版本**: Python 3.8+\n*   **核心依赖**: 必须预先安装 `torch`。\n*   **包管理器**: 推荐使用 `uv` 以获得更快的安装体验，或使用标准的 `pip` \u002F `pdm`。\n\n## 安装步骤\n\n### 1. 安装依赖\n\n您可以选择以下任意一种方式进行安装：\n\n**方式 A：使用 uv (推荐，速度更快)**\n```bash\nuv pip install -e .\nuv pip install -r requirements.txt\n```\n\n**方式 B：使用 pip**\n```bash\npip install -e .\npip install -r requirements.txt\n```\n\n**方式 C：使用 PDM (现代包管理器)**\n```bash\n# 安装 PDM\ncurl -sSL https:\u002F\u002Fraw.githubusercontent.com\u002Fpdm-project\u002Fpdm\u002Fmain\u002Finstall-pdm.py | python3 -\n\n# 初始化项目并安装依赖\npdm init\npdm install\n\n# 激活虚拟环境\neval \"$(pdm venv activate)\"\n```\n\n### 2. 配置 API 密钥与环境变量\n\n在使用前，您需要配置搜索提供商、重排序模型以及 LLM 提供商。\n\n#### A. 配置搜索提供商 (二选一)\n\n*   **选项 1: Serper.dev (推荐新手)**\n    访问 [serper.dev](https:\u002F\u002Fserper.dev) 注册获取免费额度，然后设置环境变量：\n    ```bash\n    export SERPER_API_KEY='your-api-key-here'\n    ```\n\n*   **选项 2: SearXNG (自托管\u002F公共实例)**\n    如果您有自托管或公共的 SearXNG 实例：\n    ```bash\n    export SEARXNG_INSTANCE_URL='https:\u002F\u002Fyour-searxng-instance.com'\n    export SEARXNG_API_KEY='your-api-key-here'  # 可选，视实例要求而定\n    ```\n\n#### B. 配置重排序模型 (Reranker)\n\n*   **快速开始 (Jina AI)**: 访问 [Jina AI](https:\u002F\u002Fjina.ai\u002F) 获取 API Key 并导出。\n*   **自托管方案**: 本地部署 [Infinity Embeddings](https:\u002F\u002Fgithub.com\u002Fmichaelfeil\u002Finfinity) 并使用开源模型（如 `Qwen2-7B-instruct`）。\n\n#### C. 配置 LLM 提供商 (LiteLLM)\n\nOpenDeepSearch 通过 LiteLLM 支持多种模型提供商（OpenAI, Anthropic, Google, OpenRouter, HuggingFace 等）。\n\n设置您的提供商 API Key：\n```bash\nexport OPENAI_API_KEY='your-api-key-here'\n# 或者\nexport ANTHROPIC_API_KEY='your-api-key-here'\n```\n\n*(可选)* 指定默认模型或针对特定任务（搜索、编排、评估）的模型：\n```bash\n# 通用默认模型\nexport LITELLM_MODEL_ID='openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001'\n\n# 针对搜索任务的模型\nexport LITELLM_SEARCH_MODEL_ID='openrouter\u002Fgoogle\u002Fgemini-2.0-flash-001'\n```\n\n## 基本使用\n\n### 独立使用模式\n\n以下是最简单的 Python 调用示例，展示如何初始化并执行搜索：\n\n```python\nfrom opendeepsearch import OpenDeepSearchTool\nimport os\n\n# 确保环境变量已设置 (如果在终端未设置，也可在此处代码中设置)\n# os.environ[\"SERPER_API_KEY\"] = \"your-api-key\"\n# os.environ[\"OPENAI_API_KEY\"] = \"your-api-key\"\n\n# 初始化工具\n# model_name 格式为 \"provider\u002Fmodel-name\"，例如 \"openai\u002Fgpt-4o\" 或 \"anthropic\u002Fclaude-3-opus-20240229\"\nsearch_agent = OpenDeepSearchTool(model_name=\"openai\u002Fgpt-4o-mini\")\n\n# 执行搜索查询\nquery = \"2024 年诺贝尔物理学奖得主是谁？\"\nresults = search_agent.run(query)\n\nprint(results)\n```\n\n### 运行 Gradio 演示 (可选)\n\n如果您想快速体验图形界面，可以在项目目录下运行：\n\n```bash\npython app.py\n```\n*(注：具体启动脚本文件名请以项目实际文件为准，通常为 `app.py` 或 `demo.py`)*\n\n### 模式选择\n\n在初始化或调用时，您可以选择不同的搜索模式：\n\n*   **Default Mode (默认模式 ⚡)**: 速度快，延迟低，适合大多数常规查询。\n*   **Pro Mode (专业模式 🔍)**: 进行更深度的搜索和推理，结果更准确，但耗时较长，适合复杂的多跳查询。","某金融科技公司的量化分析师需要在极短时间内，从全球数千份最新的英文财报、监管文件及行业研报中，挖掘出影响半导体供应链的关键风险信号。\n\n### 没有 OpenDeepSearch 时\n- **信息检索浅层化**：传统搜索引擎仅能匹配关键词，无法理解“产能受限”与“设备交付延迟”之间的深层逻辑关联，导致大量高价值隐性信息被遗漏。\n- **人工验证成本高**：分析师需手动打开数十个链接逐一阅读长文档，耗时数小时才能确认一个假设，且容易因疲劳产生误判。\n- **推理链条断裂**：面对跨文档的复杂线索（如 A 厂停产导致 B 材料短缺），缺乏自动推理能力，难以将碎片化信息拼凑成完整的因果图谱。\n- **时效性滞后**：等待人工整理报告往往需要隔天输出，错失市场波动的最佳决策窗口。\n\n### 使用 OpenDeepSearch 后\n- **深度语义洞察**：OpenDeepSearch 利用推理模型主动分析文档上下文，精准识别出未直接提及但逻辑相关的供应链中断风险，召回率显著提升。\n- **自动化证据链生成**：工具自动遍历多个数据源，提取关键段落并生成带有引用来源的推理报告，将数小时的工作压缩至分钟级。\n- **多步逻辑推演**：内置的智能体能够自主规划搜索路径，串联起分散在不同文件中的事件，完整呈现“原材料涨价 - 厂商减产 - 终端缺货”的传导路径。\n- **实时决策支持**：基于最新发布的文档即时完成深度分析，帮助团队在盘前会议中迅速调整投资策略。\n\nOpenDeepSearch 通过将搜索从简单的“关键词匹配”升级为具备逻辑推理能力的“深度探究”，彻底释放了专业人员在复杂信息处理上的潜能。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fsentient-agi_OpenDeepSearch_33474a3b.png","sentient-agi","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fsentient-agi_5e672b47.png","",null,"https:\u002F\u002Fgithub.com\u002Fsentient-agi",[78],{"name":79,"color":80,"percentage":81},"Python","#3572A5",100,3801,338,"2026-04-10T19:30:05","Apache-2.0","未说明","非必需。若使用自托管重排序模型（如 Qwen2-7B-instruct），建议配备支持运行 7B 参数模型的 GPU；若使用 Jina AI 等云端 API 则无需本地 GPU。","未说明（运行本地 7B 模型建议 16GB+）",{"notes":90,"python":86,"dependencies":91},"1. 必须安装 torch 库。\n2. 推荐使用 uv 或 PDM 进行依赖管理以简化安装。\n3. 需配置搜索提供商 API（Serper.dev 或 SearXNG）。\n4. 需配置重排序服务（Jina AI API 或本地部署 Infinity Embeddings）。\n5. 需配置 LiteLLM 支持的 LLM 提供商 API Key（如 OpenAI, Anthropic, Google 等）。",[92,93,94,95,96,97],"torch","Crawl4AI","SmolAgents","LiteLLM","transformers (用于本地重排序模型)","PDM (可选包管理器)",[35,13],"2026-03-27T02:49:30.150509","2026-04-11T10:01:31.175527",[],[]]