<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Projects on Peng Tan's AI Blog</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/</link><description>一个关注 AI 各领域的专题博客</description><atom:link href="https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/index.xml" rel="self" type="application/rss+xml"/><item><title>Gemini CLI 核心架构深度解析</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/gemini_cli_arch/</link><pubDate>Sun, 18 Jan 2026 18:50:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/gemini_cli_arch/</guid><description>Gemini CLI 核心架构深度解析</description></item><item><title>OpenCode CLI 核心架构深度解析</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/opencode_arch/</link><pubDate>Sun, 18 Jan 2026 18:50:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/opencode_arch/</guid><description>OpenCode CLI 核心架构深度解析</description></item><item><title>Happy: 手机app远程控制AI编程助手Claude Code/Codex</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/coder_related/remote_coder_from_phone/</link><pubDate>Thu, 15 Jan 2026 23:17:49 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/coder_related/remote_coder_from_phone/</guid><description>Happy: 手机app远程控制AI编程助手Claude Code/Codex</description></item><item><title>MemEvolve：智能体自我进化的元学习框架（OPPO）</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/memory/memevolve/</link><pubDate>Thu, 01 Jan 2026 22:00:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/memory/memevolve/</guid><description>MemEvolve是一个旨在实现大语言模型智能体自我进化的元学习框架。传统的智能体记忆系统往往采用固定架构，难以适应多变的任务需求，因此他们提出了双层优化机制：内环负责积累实践经验，外环则根据性能反馈动态重构记忆架构。</description></item><item><title>每周一个MCP：Mem0 记忆管家——把长期记忆能力接到你的 Agent 上</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/mem0_mcp/</link><pubDate>Wed, 31 Dec 2025 23:00:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/mem0_mcp/</guid><description>这个MCP服务器将 Mem0 的记忆 REST 能力封装成 MCP Tools，让对话式 Agent 获得可检索、可更新、可清理的长期记忆能力，解决“聊完就忘”“记忆不可控”的工程痛点。</description></item><item><title>Mem0: AI助手和数字人的记忆中间件</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/memory/mem0/</link><pubDate>Mon, 29 Dec 2025 23:10:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/memory/mem0/</guid><description>Mem0 是一个为 AI 助手和代理提供智能记忆层的开源项目，旨在实现个性化的 AI 交互。其核心特性包括多级记忆（用户、会话和代理状态的无缝记录）和智能记忆提取。</description></item><item><title>LLM Council</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/chatbot/llm_council/</link><pubDate>Tue, 09 Dec 2025 15:20:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/chatbot/llm_council/</guid><description>LLM Council是一个由 Andreas Karpathy 创建的开源项目，用于将多个大语言模型（LLM）组成“理事会”，通过本地 Web 应用统一处理用户查询。</description></item><item><title>TrendRadar - 多平台热点聚合和AI智能分析</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/info_collect/trendradar/</link><pubDate>Fri, 05 Dec 2025 00:00:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/info_collect/trendradar/</guid><description>TrendRadar 是一个用于多平台热点聚合和AI智能分析的开源项目，主要功能包括热点新闻监控、智能筛选及推送，同时支持基于MCP协议的深度分析。</description></item><item><title>OpenEnv: 智能体安全执行环境</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/agent/openenv/</link><pubDate>Thu, 04 Dec 2025 22:10:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/agent/openenv/</guid><description>Meta和Hugging Face正在合作推出**OpenEnv Hub**，这是一个共享和开放的社区中心，用于开发、分享和探索兼容OpenEnv的智能体环境，以支持下一波智能体开发浪潮。这些环境可以用于训练和部署，并作为可扩展智能体开发的基石。</description></item><item><title>每周一个MCP：Skill Seekers</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/skill_seeker/</link><pubDate>Sun, 02 Nov 2025 20:10:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/skill_seeker/</guid><description>Skill Seekers 是一个自动化工具，可以将文档网站、GitHub 仓库和 PDF 文件快速转换为 Claude Skills，通过自动化冲突检测和AI增强技术，大幅减少手工整理和理解文档所需的时间（20-40 分钟内生成生产就绪的Claude技能）。</description></item><item><title>CCManager：多AI编码助手会话管理工具</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/ccmanager/</link><pubDate>Wed, 29 Oct 2025 14:20:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/ccmanager/</guid><description>CCManager：多AI编码助手会话管理工具，无需借助tmux，完全独立运行</description></item><item><title>Claudable：基于Next.js的强大网页应用构建工具</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/claudable/</link><pubDate>Wed, 29 Oct 2025 14:20:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/claudable/</guid><description>Claudable：基于Next.js的强大网页应用构建工具</description></item><item><title>SuperClaude: 打造Claude Code成一个企业级的开发团队</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/superclaude/</link><pubDate>Sat, 25 Oct 2025 11:30:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/superclaude/</guid><description>SuperClaude: 打造Claude Code成一个全副武装的团队</description></item><item><title>每周一个MCP：Serena MCP</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/serena_mcp/</link><pubDate>Tue, 21 Oct 2025 15:03:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/serena_mcp/</guid><description>Serena MCP 是一个功能强大且开源的编码代理工具包，旨在将大型语言模型（LLMs）转化为具备完整功能的智能代码助手，专注于语义代码检索和编辑，模拟集成开发环境（IDE）的能力。</description></item><item><title>每周一个MCP： Windows-MCP</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/windows_mcp/</link><pubDate>Mon, 13 Oct 2025 15:00:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/windows_mcp/</guid><description>Windows-MCP Server 开源项目分析</description></item><item><title>每周一个MCP：基于Go的自动化运营小红书账号工具xiaohongshu-mcp</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/xiaohongshu-mcp/</link><pubDate>Fri, 10 Oct 2025 14:00:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/xiaohongshu-mcp/</guid><description>xiaohongshu-mcp 是一个开源项目，用于实现对小红书 (xiaohongshu.com) 的 Model Context Protocol (MCP) 支持，基于自动化运营小红书账号。</description></item><item><title>Claude Code Spec Workflow, 也支持MCP方式</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/spec-workflow-mcp/</link><pubDate>Sun, 14 Sep 2025 22:30:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/spec-workflow-mcp/</guid><description>Claude Code Spec Workflow, 也支持MCP方式</description></item><item><title>Spec Kit：基于规范驱动开发的工具包</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/spec-kit/</link><pubDate>Sun, 14 Sep 2025 21:10:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/spec-kit/</guid><description>Spec Kit：基于规范驱动开发的工具包</description></item><item><title>Specification-Driven Development (SDD) - 规范驱动开发</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/spec-driven-dev/</link><pubDate>Sat, 13 Sep 2025 23:03:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/spec-driven-dev/</guid><description>Specification-Driven Development (SDD) - 规范驱动开发</description></item><item><title>每周一个MCP： Magic UI MCP Server</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/magic_ui_mcp_server/</link><pubDate>Sat, 16 Aug 2025 23:03:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/magic_ui_mcp_server/</guid><description>Magic UI MCP Server 开源项目分析</description></item><item><title>每周一个MCP： Zen MCP Server</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/zen_mcp_server/</link><pubDate>Sat, 16 Aug 2025 23:03:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/zen_mcp_server/</guid><description>Zen MCP Server 开源项目分析</description></item><item><title>Claude-Code-Router：AI 时代的智能路由中枢</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/claude-code-router/</link><pubDate>Tue, 12 Aug 2025 22:10:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/claude-code-router/</guid><description>&lt;p>目录&lt;/p>
&lt;ul>
&lt;li>&lt;a href="https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/claude-code-router/#1-%e5%bc%95%e8%a8%80ai-%e6%9c%8d%e5%8a%a1%e6%99%ba%e8%83%bd%e8%b7%af%e7%94%b1%e7%9a%84%e6%96%b0%e8%8c%83%e5%bc%8f">1. 引言：AI 服务智能路由的新范式&lt;/a>&lt;/li>
&lt;li>&lt;a href="https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/claude-code-router/#2-claude-code-router-%e6%a0%b8%e5%bf%83%e6%9c%ba%e5%88%b6%e6%80%bb%e8%a7%88">2. Claude-Code-Router 核心机制总览&lt;/a>&lt;/li>
&lt;li>&lt;a href="https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/claude-code-router/#3-%e6%99%ba%e8%83%bd%e8%b7%af%e7%94%b1%e5%86%b3%e7%ad%96%e6%9c%ba%e5%88%b6%e8%af%a6%e8%a7%a3">3. 智能路由决策机制详解&lt;/a>&lt;/li>
&lt;li>&lt;a href="https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/claude-code-router/#4-%e8%af%b7%e6%b1%82%e8%bd%ac%e6%8d%a2%e4%b8%8e%e8%bd%ac%e5%8f%91%e6%9c%ba%e5%88%b6">4. 请求转换与转发机制&lt;/a>&lt;/li>
&lt;li>&lt;a href="https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/claude-code-router/#5-%e9%94%99%e8%af%af%e5%a4%84%e7%90%86%e4%b8%8e%e9%99%8d%e7%ba%a7%e7%ad%96%e7%95%a5">5. 错误处理与降级策略&lt;/a>&lt;/li>
&lt;li>&lt;a href="https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/claude-code-router/#6-%e6%8f%92%e4%bb%b6%e7%b3%bb%e7%bb%9f%e4%b8%8e%e6%89%a9%e5%b1%95%e6%80%a7">6. 插件系统与扩展性&lt;/a>&lt;/li>
&lt;li>&lt;a href="https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/claude-code-router/#7-%e6%80%a7%e8%83%bd%e4%bc%98%e5%8c%96%e4%b8%8e%e7%9b%91%e6%8e%a7">7. 性能优化与监控&lt;/a>&lt;/li>
&lt;li>&lt;a href="https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/code_assistant/claude-code-router/#8-%e6%9c%aa%e6%9d%a5%e5%b1%95%e6%9c%9b%e4%b8%8e%e6%8a%80%e6%9c%af%e6%8c%91%e6%88%98">8. 未来展望与技术挑战&lt;/a>&lt;/li>
&lt;/ul>
&lt;p>Claude-Code-Router (CCR) 是一款创新的AI模型智能路由工具，它通过拦截Claude Code 应用对Anthropic Claude模型的请求，进行多维度分析（如Token数量、用户指令、任务类型），然后依据动态路由规则和配置，将请求智能地导向最合适的AI模型（来自如Gemini、DeepSeek、本地Ollama模型等不同的模型服务提供商）。CCR的核心机制包括API格式的自动转换与适配、基于Express.js的中间件架构、异步请求处理，以及完善的错误检测、自动降级到兜底模型和潜在的重试策略，旨在提升AI服务调用的效率、灵活性和成本效益。&lt;/p></description></item><item><title>Taskmaster AI - 通过AI任务管理来提升开发效率</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/taskmaster/</link><pubDate>Wed, 09 Jul 2025 09:20:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/taskmaster/</guid><description>Taskmaster AI - 通过AI任务管理来提升开发效率，支持多种开发工具，如Cursor、Windsurf、VS Code、Claude Code CLI等。</description></item><item><title>每周一个MCP： 抖音视频无水印提取MCP服务器</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/douyin_extract_mcp/</link><pubDate>Mon, 07 Jul 2025 23:03:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/douyin_extract_mcp/</guid><description>douyin-mcp-server 开源项目分析</description></item><item><title>TradingAgents 开源项目分析</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/tradingagents/</link><pubDate>Thu, 03 Jul 2025 23:03:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/tradingagents/</guid><description>TradingAgents 开源项目分析</description></item><item><title>每周一个MCP：Context7</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/mcp_context7/</link><pubDate>Sun, 29 Jun 2025 22:30:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/mcp_context7/</guid><description>Context7 是一个用于 LLM 和 AI 编码编辑器的 MCP 服务器，可以提供最新的代码文档和代码示例，使得生成的代码更准确、版本相关且避免过时或虚假信息。非常适合配合AI编码助手使用更新版本的API。</description></item><item><title>SkyworkAI DeepResearchAgent</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/skyworkai_deepresearchagent/</link><pubDate>Mon, 16 Jun 2025 23:03:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/skyworkai_deepresearchagent/</guid><description>SkyworkAI DeepResearchAgent</description></item><item><title>DeerFlow - 字节跳动开源的Deep Research</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/deerflow/</link><pubDate>Wed, 21 May 2025 22:30:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/deerflow/</guid><description>DeerFlow - 字节跳动开源的Deep Research</description></item><item><title>OpenEvolve - 开源进化算法项目</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/openevolve/</link><pubDate>Tue, 20 May 2025 00:00:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/openevolve/</guid><description>OpenEvolve相关开源项目和资源链接</description></item><item><title>每周一个MCP：FastAPI-MCP</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/fastapi_mcp/</link><pubDate>Tue, 22 Apr 2025 15:03:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/fastapi_mcp/</guid><description>FastAPI-MCP 是一个扩展 FastAPI 的工具，用于将 FastAPI 的端点暴露为具有认证功能的 Model Context Protocol (MCP) 工具。</description></item><item><title>AI Agent Gateway</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/agentgateway/</link><pubDate>Thu, 10 Apr 2025 23:03:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/agentgateway/</guid><description>AI Agent Gateway</description></item><item><title>STORM - 通过检索和多视角提问来合成主题大纲和维基百科类文章</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/storm/</link><pubDate>Fri, 14 Jun 2024 14:30:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/storm/</guid><description>STORM - 通过检索和多视角提问来合成主题大纲和维基百科类文章</description></item><item><title>Nanobot 深度剖析：如何用 4000 行代码构建超轻量级个人 AI 助理</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/auto/nanobot/</link><pubDate>Fri, 15 Mar 2024 00:00:00 +0000</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/auto/nanobot/</guid><description>Nanobot 是一个超轻量级的个人 AI 助理框架，仅用 4000 行代码实现了 ReAct 循环、多渠道消息总线、工具/技能系统、长短期记忆、子智能体协同等功能。可以看成是 OpenClaw 的极简版，是一个非常好的学习项目。</description></item><item><title>claudia</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/claudia/</link><pubDate>Mon, 01 Jan 2024 00:00:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/claudia/</guid><description>claudia相关资源链接</description></item><item><title>openmanus</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/openmanus/</link><pubDate>Mon, 01 Jan 2024 00:00:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/openmanus/</guid><description>openmanus相关资源链接</description></item><item><title>UI-TARS Desktop - 字节跳动桌面应用</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/ui-tars-desktop/</link><pubDate>Mon, 01 Jan 2024 00:00:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/ui-tars-desktop/</guid><description>字节跳动UI-TARS桌面应用项目</description></item><item><title/><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/autogen/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/autogen/</guid><description>&lt;h2 id="autogen-04-版本">AutoGen 0.4 版本&lt;/h2>
&lt;p>根据我搜索到的信息，AutoGen 最新版本（特别是 0.4 版本及更高版本）引入了许多新特性和改进，主要集中在以下几个方面：&lt;/p>
&lt;p>核心架构和 API (Core &amp;amp; API)&lt;/p></description></item><item><title/><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/github/bettafish/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/github/bettafish/</guid><description/></item><item><title/><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/langchain/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/langchain/</guid><description>&lt;h2 id="langchain-技术架构与实现详解">LangChain 技术架构与实现详解&lt;/h2>
&lt;p>LangChain 是一个强大的框架，旨在简化和加速基于大型语言模型 (LLM) 的应用程序的开发。它的技术架构和实现围绕着&lt;strong>模块化、可组合性和标准化接口&lt;/strong>的核心思想。&lt;/p></description></item><item><title/><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/llm/nanochat/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/llm/nanochat/</guid><description>&lt;p>项目基本信息
名称: nanochat - &amp;ldquo;最便宜的$100 ChatGPT克隆&amp;rdquo;
用途: 完整的LLM训练和推理流程，从数据到Web UI
规模: ~8K行代码，45个文件，单个 8XH100 节点上 4 小时完成训练
成本: ~$100 (d20 模型，561M 参数)
🛠️ 技术栈
Python 3.10+ + PyTorch 2.8+
Rust tokenizer (rustbpe via PyO3)
FastAPI + Uvicorn 用于 Web 服务
HuggingFace datasets + wandb 实验追踪
uv 作为依赖管理器&lt;/p></description></item><item><title/><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/0_tbd_mcp/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/0_tbd_mcp/</guid><description>&lt;p>Zen MCP (Zread.ai)
Supabase MCP (&lt;a href="https://github.com/supabase-community/supabase-mcp">https://github.com/supabase-community/supabase-mcp&lt;/a>)
Browser MCP(It’s like having Google built into your editor. No more jumping between tabs.)
Claude Taskmaster
Exa MCP(&lt;a href="https://github.com/exa-labs/exa-mcp-server">https://github.com/exa-labs/exa-mcp-server&lt;/a>)&lt;/p>
&lt;p>Lastmile MCP: &lt;a href="https://github.com/lastmile-ai/mcp-agent">https://github.com/lastmile-ai/mcp-agent&lt;/a> (Lastmile MCP提供生产级框架，支持多代理协调与人类参与工作流，简化MCP服务器生命周期管理。)它对 Anthropic 的“构建有效代理”研究中每一种模式的实施，再加上 OpenAI 的 Swarm 多代理编排，树立了标准。该框架处理所有繁琐的 MCP 服务器生命周期管理工作，同时为您提供强大的工作流模式，例如评估器 - 优化器、编排器 - 工作者以及并行执行。
Claude Code MCP Server : &lt;a href="https://github.com/steipete/claude-code-mcp">https://github.com/steipete/claude-code-mcp&lt;/a> (其与众不同之处在于能够绕过权限中断，并一次性执行复杂的编码工作流。,对于那些想要利用 Claude 的系统级访问权限来完成其他集成开发环境无法处理的任务的开发者来说，这是很有价值的。)&lt;/p></description></item><item><title/><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/21st_dev_magicai_mcp/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/21st_dev_magicai_mcp/</guid><description>&lt;p>21st Dev Magic AI Agent making your app look pro,&lt;/p></description></item><item><title/><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/exa_mcp/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/exa_mcp/</guid><description>&lt;p>Exa MCP Server 是一个由 Exa Labs 开发的开源工具，主要用于为 AI 助手（如 Claude）提供高效的代码搜索与网络搜索服务。其核心功能包括：&lt;/p>
&lt;ol>
&lt;li>&lt;strong>代码上下文搜索&lt;/strong>：快速检索代码片段、API 用法和最佳实践，从 GitHub 库及其他资源中提供相关代码参考。&lt;/li>
&lt;li>&lt;strong>实时网络搜索与爬取&lt;/strong>：支持实时爬取网页内容（如 URL、文章、PDF 等），并优化搜索结果。&lt;/li>
&lt;li>&lt;strong>其他工具&lt;/strong>：包括公司信息爬取、LinkedIn 搜索、深度研究等功能。&lt;/li>
&lt;li>&lt;strong>多种集成方式&lt;/strong>：支持通过 Smithery、Claude Desktop、Codex、Claude Code 插件等平台进行配置与集成。&lt;/li>
&lt;li>&lt;strong>简单配置&lt;/strong>：通过 API 密钥及多种工具参数选择，用户可自定义启用功能模块。&lt;/li>
&lt;/ol>
&lt;p>项目支持远程或本地部署，核心搜索功能由 Exa 的搜索引擎驱动，适合开发者和研究者使用。技术栈以 TypeScript 为主，采用 MIT 开源协议，目前拥有 3.1k Star 和 231 Fork，可通过 NPM 或源码直接安装使用。&lt;/p></description></item><item><title/><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/graphiti/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/mcp/graphiti/</guid><description>&lt;p>Graphiti 是一个用于 AI 代理构建和查询动态图时序知识图的开源框架。相较传统的检索增强生成（RAG）方法，Graphiti 提供实时增量数据更新、双时间轴数据模型和高效的混合检索功能，可处理动态用户交互、企业数据和外部信息的整合。它支持精准历史查询、语义搜索、关键词搜索和图遍历，适合构建交互式、上下文感知的 AI 应用。&lt;/p></description></item><item><title/><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/openai_codex_cli/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/openai_codex_cli/</guid><description>&lt;p>&lt;a href="https://github.com/openai/codex">&lt;/a>&lt;/p></description></item><item><title/><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/presentationai/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/presentationai/</guid><description>&lt;p>ALLWEONE® Presentation AI 是一个开源、AI驱动的演示文稿生成工具，灵感来源于 Gamma.app，可快速生成美观且可定制的幻灯片。其主要功能包括：AI内容生成、多个内置主题、实时生成、实时编辑、自动保存、图像生成及演示模式等。支持使用本地模型（如 Ollama 和 LM Studio）进行生成。技术栈涵盖 Next.js、React、PostgreSQL 等。项目规划包括支持 PDF 导出、实时协作、多设备响应、插件系统等新功能。用户可通过贡献代码、报告问题或加入 Discord 社区提供支持。项目采用 MIT 许可证开源，主站为 &lt;a href="http://presentation.allweone.com">presentation.allweone.com&lt;/a>。&lt;/p></description></item><item><title/><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/quant/qlib_rdagent/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/quant/qlib_rdagent/</guid><description>&lt;h1 id="tldr">TLDR&lt;/h1>
&lt;p>RD-Agent是一个数据驱动的AI研发自动化框架，其核心理念是将研发过程分解为两个关键组件：&lt;/p>
&lt;ul>
&lt;li>R (Research)：主动探索，提出新想法&lt;/li>
&lt;li>D (Development)：实现想法，将概念转化为可执行的代码&lt;/li>
&lt;li>&lt;/li>
&lt;/ul>
&lt;p>qlib&lt;/p></description></item><item><title>LangGraph技术底座</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/langgraph/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/langgraph/</guid><description>&lt;h2 id="langgraph-技术架构与实现详解">LangGraph 技术架构与实现详解&lt;/h2>
&lt;p>本文档旨在全面剖析 LangGraph 的技术架构与底层实现。LangGraph 是一个用于构建有状态、可循环、多参与者（Multi-agent）应用的强大框架，它作为 LangChain 生态的关键扩展，为复杂的 AI 工作流提供了图计算的能力。&lt;/p></description></item><item><title>Open-AutoGLM</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/auto/open_autoglm/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/projects/auto/open_autoglm/</guid><description>&lt;p>Open-AutoGLM 是一个基于 AutoGLM 构建的手机端智能助手框架，旨在通过多模态技术理解手机屏幕内容，并结合自动化操作帮助用户完成各类任务。它通过 ADB（Android Debug Bridge）控制设备，使用视觉语言模型进行屏幕感知，并结合智能规划能力生成并执行操作。用户只需用自然语言描述需求，系统即可解析意图并完成任务，如“打开小红书搜索美食”等。&lt;/p></description></item></channel></rss>