# AI 开源趋势日报 2026-03-09 > 数据来源: GitHub Trending + GitHub Search API | 生成时间: 2026-03-09 00:07 UTC --- # AI 开源趋势日报 | 2026-03-09 --- ## 第一步:AI 相关性筛选 **Trending 榜单筛选结果(12 → 9 个 AI 相关)**: | 项目 | 判定 | 说明 | |:---|:---|:---| | GoogleCloudPlatform/generative-ai | ✅ 保留 | Google Cloud 生成式 AI 官方示例 | | 666ghj/MiroFish | ✅ 保留 | 群体智能预测引擎 | | shadcn-ui/ui | ❌ 排除 | 通用 UI 组件库,无 AI 特性 | | openclaw/openclaw | ✅ 保留 | 个人 AI 助手 | | toeverything/AFFiNE | ✅ 保留 | AI 知识库(含 AI 功能) | | Ed1s0nZ/CyberStrikeAI | ✅ 保留 | AI 原生安全测试平台 | | shareAI-lab/learn-claude-code | ✅ 保留 | Claude Code 类 Agent 实现 | | openai/skills | ✅ 保留 | OpenAI Codex 技能目录 | | virattt/ai-hedge-fund | ✅ 保留 | AI 对冲基金团队 | | is-a-dev/register | ❌ 排除 | 域名注册工具,无关 | | teng-lin/notebooklm-py | ✅ 保留 | Google NotebookLM 非官方 API | | pbakaus/impeccable | ✅ 保留 | AI 设计系统语言 | **主题搜索结果**:79 个已全部为 AI/ML 相关,全部保留。 --- ## 第二步:分类结果 | 项目 | 主要类别 | 次要类别 | |:---|:---|:---| | GoogleCloudPlatform/generative-ai | 🔧 AI 基础工具 | 🧠 大模型/训练 | | 666ghj/MiroFish | 🤖 AI 智能体/工作流 | 📦 AI 应用 | | openclaw/openclaw | 🤖 AI 智能体/工作流 | 📦 AI 应用 | | toeverything/AFFiNE | 📦 AI 应用 | 🔍 RAG/知识库 | | Ed1s0nZ/CyberStrikeAI | 📦 AI 应用 | 🤖 AI 智能体/工作流 | | shareAI-lab/learn-claude-code | 🔧 AI 基础工具 | 🤖 AI 智能体/工作流 | | openai/skills | 🔧 AI 基础工具 | 🧠 大模型/训练 | | virattt/ai-hedge-fund | 📦 AI 应用 | 🤖 AI 智能体/工作流 | | teng-lin/notebooklm-py | 🔧 AI 基础工具 | 📦 AI 应用 | | pbakaus/impeccable | 🔧 AI 基础工具 | 📦 AI 应用 | | ollama/ollama | 🔧 AI 基础工具 | 🧠 大模型/训练 | | huggingface/transformers | 🔧 AI 基础工具 | 🧠 大模型/训练 | | langgenius/dify | 🤖 AI 智能体/工作流 | 🔧 AI 基础工具 | | langchain-ai/langchain | 🤖 AI 智能体/工作流 | 🔧 AI 基础工具 | | open-webui/open-webui | 📦 AI 应用 | 🔧 AI 基础工具 | | firecrawl/firecrawl | 🔧 AI 基础工具 | 🔍 RAG/知识库 | | rasbt/LLMs-from-scratch | 🧠 大模型/训练 | 🔧 AI 基础工具 | | browser-use/browser-use | 🤖 AI 智能体/工作流 | 📦 AI 应用 | | infiniflow/ragflow | 🔍 RAG/知识库 | 🤖 AI 智能体/工作流 | | vllm-project/vllm | 🔧 AI 基础工具 | 🧠 大模型/训练 | | OpenHands/OpenHands | 🤖 AI 智能体/工作流 | 🔧 AI 基础工具 | | hiyouga/LlamaFactory | 🧠 大模型/训练 | 🔧 AI 基础工具 | | affaan-m/everything-claude-code | 🔧 AI 基础工具 | 🤖 AI 智能体/工作流 | | Mintplex-Labs/anything-llm | 📦 AI 应用 | 🔍 RAG/知识库 | | open-compass/opencompass | 🔧 AI 基础工具 | 🧠 大模型/训练 | | 0xPlaygrounds/rig | 🔧 AI 基础工具 | 🤖 AI 智能体/工作流 | | skyzh/tiny-llm | 🧠 大模型/训练 | 🔧 AI 基础工具 | | meilisearch/meilisearch | 🔍 RAG/知识库 | 🔧 AI 基础工具 | | run-llama/llama_index | 🔍 RAG/知识库 | 🤖 AI 智能体/工作流 | | milvus-io/milvus | 🔍 RAG/知识库 | 🔧 AI 基础工具 | | qdrant/qdrant | 🔍 RAG/知识库 | 🔧 AI 基础工具 | | chroma-core/chroma | 🔍 RAG/知识库 | 🔧 AI 基础工具 | | VectifyAI/PageIndex | 🔍 RAG/知识库 | 🔧 AI 基础工具 | | memvid/memvid | 🔍 RAG/知识库 | 🤖 AI 智能体/工作流 | | topoteretes/cognee | 🔍 RAG/知识库 | 🤖 AI 智能体/工作流 | | Shubhamsaboo/awesome-llm-apps | 📦 AI 应用 | 🤖 AI 智能体/工作流 | | PaddlePaddle/PaddleOCR | 🔧 AI 基础工具 | 🔍 RAG/知识库 | | FlowiseAI/Flowise | 🤖 AI 智能体/工作流 | 🔧 AI 基础工具 | | mem0ai/mem0 | 🔍 RAG/知识库 | 🤖 AI 智能体/工作流 | | thedotmack/claude-mem | 🔍 RAG/知识库 | 🤖 AI 智能体/工作流 | | khoj-ai/khoj | 📦 AI 应用 | 🔍 RAG/知识库 | | zhayujie/chatgpt-on-wechat | 📦 AI 应用 | 🤖 AI 智能体/工作流 | | CherryHQ/cherry-studio | 📦 AI 应用 | 🤖 AI 智能体/工作流 | | CopilotKit/CopilotKit | 🔧 AI 基础工具 | 🤖 AI 智能体/工作流 | | activepieces/activepieces | 🤖 AI 智能体/工作流 | 🔧 AI 基础工具 | | iOfficeAI/AionUi | 📦 AI 应用 | 🤖 AI 智能体/工作流 | | googleworkspace/cli | 🔧 AI 基础工具 | 🤖 AI 智能体/工作流 | | trycua/cua | 🔧 AI 基础工具 | 🤖 AI 智能体/工作流 | | e2b-dev/E2B | 🔧 AI 基础工具 | 🤖 AI 智能体/工作流 | | alibaba/OpenSandbox | 🔧 AI 基础工具 | 🤖 AI 智能体/工作流 | | crestalnetwork/intentkit | 🤖 AI 智能体/工作流 | 🔧 AI 基础工具 | --- ## 第三步:AI 开源趋势日报 --- ### 1. 今日速览 今日 AI 开源领域呈现**"Agent 工具链爆发"**与**"Claude Code 生态崛起"**两大主线。[openclaw/openclaw](https://github.com/openclaw/openclaw) 以 **+4,603 stars** 的惊人增速登顶,标志着个人 AI 助手进入"跨平台原生"新阶段;[shareAI-lab/learn-claude-code](https://github.com/shareAI-lab/learn-claude-code) 和 [affaan-m/everything-claude-code](https://github.com/affaan-m/everything-claude-code) 同时高热,反映开发者对 Claude Code 类 Agent 实现的强烈需求。与此同时,[666ghj/MiroFish](https://github.com/666ghj/MiroFish) 提出的"群体智能引擎"概念,以及 [Ed1s0nZ/CyberStrikeAI](https://github.com/Ed1s0nZ/CyberStrikeAI) 的 AI 安全测试平台,显示 AI 应用正向**垂直专业化**和**多智能体协作**方向深度演进。 --- ### 2. 各维度热门项目 #### 🔧 AI 基础工具(框架、SDK、推理引擎、开发工具、CLI) | 项目 | Stars | 一句话说明 | |:---|:---|:---| | [openclaw/openclaw](https://github.com/openclaw/openclaw) | 新榜 +4,603 | **今日增速冠军**——"龙虾方式"的跨平台个人 AI 助手,支持任意 OS/平台,定位对标 Claude Code 的开源替代 | | [shareAI-lab/learn-claude-code](https://github.com/shareAI-lab/learn-claude-code) | 新榜 +566 | 从零构建 Claude Code 类 Agent 的极简实现,"Bash is all you need"理念降低 Agent 开发门槛 | | [ollama/ollama](https://github.com/ollama/ollama) | 164,478 | 本地大模型运行的事实标准,今日新增支持 Kimi-K2.5、GLM-5、MiniMax 等国产模型 | | [vllm-project/vllm](https://github.com/vllm-project/vllm) | 72,445 | 高吞吐 LLM 推理引擎,生产级部署的首选基础设施 | | [firecrawl/firecrawl](https://github.com/firecrawl/firecrawl) | 89,544 | 将任意网站转为 LLM-ready 数据的 Web Data API,RAG pipeline 的关键组件 | | [affaan-m/everything-claude-code](https://github.com/affaan-m/everything-claude-code) | 67,405 | Claude Code 性能优化系统,涵盖技能、记忆、安全等 Agent harness 完整方案 | | [openai/skills](https://github.com/openai/skills) | 新榜 +612 | OpenAI 官方 Codex 技能目录,定义 Agent 能力标准化的最新基准 | | [googleworkspace/cli](https://github.com/googleworkspace/cli) | 16,276 | Google Workspace 官方 CLI 内置 AI Agent skills,企业工具链 Agent 化的标志性产品 | #### 🤖 AI 智能体/工作流(Agent 框架、自动化、多智能体) | 项目 | Stars | 一句话说明 | |:---|:---|:---| | [666ghj/MiroFish](https://github.com/666ghj/MiroFish) | 新榜 +1,104 | **"群体智能引擎"**——简洁通用的 Swarm Intelligence 框架,定位"预测万物",今日增速第二 | | [langgenius/dify](https://github.com/langgenius/dify) | 131,631 | 生产级 Agentic 工作流开发平台,"LLM 应用开发"到"Agent 工程"的范式升级代表 | | [OpenHands/OpenHands](https://github.com/OpenHands/OpenHands) | 68,767 | AI 驱动软件开发的开源标杆,代码生成 Agent 的社区首选 | | [browser-use/browser-use](https://github.com/browser-use/browser-use) | 79,928 | 让 AI Agent 操控浏览器自动化任务,Computer-Use Agent 的核心基础设施 | | [activepieces/activepieces](https://github.com/activepieces/activepieces) | 21,137 | 支持 ~400 个 MCP 服务器的 AI 工作流自动化平台,MCP 生态集成度领先 | | [crestalnetwork/intentkit](https://github.com/crestalnetwork/intentkit) | 6,491 | 自托管云 Agent 集群,多 Agent 协作管理的开源解决方案 | | [trycua/cua](https://github.com/trycua/cua) | 12,927 | 开源 Computer-Use Agent 基础设施,支持 macOS/Linux/Windows 全桌面控制 | | [e2b-dev/E2B](https://github.com/e2b-dev/E2B) | 11,178 | 企业级 Agent 安全沙箱环境,真实工具链集成的标准方案 | #### 📦 AI 应用(具体应用产品、垂直场景解决方案) | 项目 | Stars | 一句话说明 | |:---|:---|:---| | [virattt/ai-hedge-fund](https://github.com/virattt/ai-hedge-fund) | 新榜 +275 | AI 对冲基金团队——金融垂直场景的 Agent 多智能体应用 | | [Ed1s0nZ/CyberStrikeAI](https://github.com/Ed1s0nZ/CyberStrikeAI) | 新榜 +244 | **AI 原生安全测试平台**——集成 100+ 安全工具的智能编排引擎,网络安全+AI 的垂直突破 | | [toeverything/AFFiNE](https://github.com/toeverything/AFFiNE) | 新榜 +533 | Notion/Miro 替代方案,隐私优先的 AI 知识库,今日新增 533 stars | | [open-webui/open-webui](https://github.com/open-webui/open-webui) | 126,246 | 最友好的本地 LLM 界面,Ollama 生态的标配前端 | | [CherryHQ/cherry-studio](https://github.com/CherryHQ/cherry-studio) | 41,020 | 300+ 助手的 AI 生产力工作室,统一接入前沿大模型 | | [zhayujie/chatgpt-on-wechat](https://github.com/zhayujie/chatgpt-on-wechat) | 42,012 | 国产 CowAgent——支持多平台接入的超级 AI 助理,主动思考与任务规划 | | [saturndec/waoowaoo](https://github.com/saturndec/waoowaoo) | 8,745 | **工业级 AI 影视生产平台**——好莱坞标准工作流的 Agent 化,AI 视频生成的专业级应用 | | [GoogleCloudPlatform/generative-ai](https://github.com/GoogleCloudPlatform/generative-ai) | 新榜 +522 | Google Cloud 官方生成式 AI 示例库,Gemini on Vertex AI 的最佳实践 | #### 🧠 大模型/训练(模型权重、训练框架、微调工具) | 项目 | Stars | 一句话说明 | |:---|:---|:---| | [huggingface/transformers](https://github.com/huggingface/transformers) | 157,569 | 大模型定义框架的事实标准,支持文本/视觉/音频/多模态全场景 | | [hiyouga/LlamaFactory](https://github.com/hiyouga/LlamaFactory) | 68,050 | 100+ LLM/VLM 统一高效微调框架,ACL 2024 明星项目 | | [rasbt/LLMs-from-scratch](https://github.com/rasbt/LLMs-from-scratch) | 87,437 | 从零实现 ChatGPT 级 LLM 的 PyTorch 教程,系统工程师的入门圣经 | | [skyzh/tiny-llm](https://github.com/skyzh/tiny-llm) | 3,909 | Apple Silicon 上的 LLM 推理服务课程,vLLM+Qwen 的极简复现 | | [open-compass/opencompass](https://github.com/open-compass/opencompass) | 6,727 | 开源大模型评测平台,支持 100+ 数据集的标准化评估 | #### 🔍 RAG/知识库(向量数据库、检索增强、知识管理) | 项目 | Stars | 一句话说明 | |:---|:---|:---| | [infiniflow/ragflow](https://github.com/infiniflow/ragflow) | 74,429 | **RAG 引擎+Agent 能力融合**的领先开源方案,上下文层构建的首选 | | [run-llama/llama_index](https://github.com/run-llama/llama_index) | 47,484 | 文档 Agent 与 OCR 平台,RAG 到 Agent 的演进代表 | | [VectifyAI/PageIndex](https://github.com/VectifyAI/PageIndex) | 20,853 | **"无向量、基于推理的 RAG"**——97% 存储节省的革新性文档索引方案 | | [mem0ai/mem0](https://github.com/mem0ai/mem0) | 49,065 | AI Agent 的通用记忆层,长期记忆管理的标准组件 | | [topoteretes/cognee](https://github.com/topoteretes/cognee) | 13,055 | 6 行代码构建 AI Agent 记忆的知识引擎,记忆层开发的极简方案 | | [memvid/memvid](https://github.com/memvid/memvid) | 13,324 | 单文件 serverless 记忆层,替代复杂 RAG pipeline 的轻量方案 | | [thedotmack/claude-mem](https://github.com/thedotmack/claude-mem) | 33,494 | Claude Code 会话记忆插件,AI 压缩与上下文注入的专项优化 | | [milvus-io/milvus](https://github.com/milvus-io/milvus) / [qdrant/qdrant](https://github.com/qdrant/qdrant) / [chroma-core/chroma](https://github.com/chroma-core/chroma) | 43,200 / 29,375 / 26,513 | 云原生向量数据库三强,分别代表 Go/Rust/Rust 技术栈的高性能方案 | --- ### 3. 趋势信号分析 **Agent 基础设施进入"爆发期"**。今日 Trending 榜单中,[openclaw/openclaw](https://github.com/openclaw/openclaw) 以 +4,603 stars 的增速创下近期纪录,其"Any OS. Any Platform"的跨平台定位,直接挑战 Claude Code 的封闭生态。配合 [shareAI-lab/learn-claude-code](https://github.com/shareAI-lab/learn-claude-code)(+566)和 [affaan-m/everything-claude-code](https://github.com/affaan-m/everything-claude-code)(67K stars 基数),形成**"Claude Code 开源替代"**的完整工具链,反映开发者对 Agent 控制权和可定制性的强烈诉求。 **"群体智能"与"多智能体协作"成为新范式**。[666ghj/MiroFish](https://github.com/666ghj/MiroFish) 提出的 Swarm Intelligence Engine 以 +1,104 stars 快速崛起,与 [crestalnetwork/intentkit](https://github.com/crestalnetwork/intentkit) 的 Agent 集群管理、[virattt/ai-hedge-fund](https://github.com/virattt/ai-hedge-fund) 的多 Agent 金融团队形成呼应,显示单一 Agent 向**协作式多 Agent 系统**的演进趋势。 **垂直场景 Agent 专业化加速**。[Ed1s0nZ/CyberStrikeAI](https://github.com/Ed1s0nZ/CyberStrikeAI) 将 AI 与安全测试深度结合,[saturndec/waoowaoo](https://github.com/saturndec/waoowaoo) 瞄准好莱坞级影视生产,标志着 AI 应用从"通用聊天"向**高门槛专业领域**渗透。这与 OpenAI Codex、Claude 3.7 Sonnet 等模型的代码/推理能力发布直接相关,模型能力的提升正在解锁新的垂直场景。 **记忆层创新成为 RAG 下一站**。[VectifyAI/PageIndex](https://github.com/VectifyAI/PageIndex) 的"无向量推理 RAG"和 [memvid/memvid](https://github.com/memvid/memvid) 的"单文件 serverless 记忆层",正在挑战传统向量数据库的复杂性,**轻量化、推理增强、Agent 原生**成为记忆层设计的新标准。 --- ### 4. 社区关注热点 - **🔥 [openclaw/openclaw](https://github.com/openclaw/openclaw) — "龙虾方式"的跨平台 AI 助手** - 今日增速 +4,603 stars 的现象级项目,TypeScript 实现的个人 AI 助手,强调"Any OS. Any Platform"的开放性,直接对标 Claude Code 的封闭生态,可能成为开源 Agent 操作系统的重要候选。 - **🐟 [666ghj/MiroFish](https://github.com/666ghj/MiroFish) — 群体智能预测引擎** - 中文项目以"简洁通用、预测万物"为定位,Swarm Intelligence 的工程化实现,多智能体协作的新范式探索,值得跟踪其实际应用场景的落地。 - **🛡️ [Ed1s0nZ/CyberStrikeAI](https://github.com/Ed1s0nZ/CyberStrikeAI) — AI 原生安全测试平台** - 100+ 安全工具的智能编排,Go 语言实现的高性能安全 Agent,代表 AI 向网络安全这一高专业门槛领域的深度渗透,企业级安全市场的潜在变革者。 - **📚 [VectifyAI/PageIndex](https://github.com/VectifyAI/PageIndex) — 无向量推理 RAG** - 97% 存储节省的革新性方案,"Vectorless, Reasoning-based RAG"挑战传统向量检索范式,若效果验证通过,可能重塑 RAG 技术栈。 - **🎬 [saturndec/waoowaoo](https://github.com/saturndec/waoowaoo) — 工业级 AI 影视生产** - 从短视频到真人影视的完整 Agent 工作流,"好莱坞标准"的 AI 影视生产平台,AI 视频生成从玩具走向专业制作的关键标志。 --- *报告生成时间:2026-03-09 | 数据来源:GitHub Trending & Search API* --- *本日报由 [agents-radar](https://github.com/duanyytop/agents-radar) 自动生成。*