[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-pocketpaw--pocketpaw":3,"tool-pocketpaw--pocketpaw":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",140436,2,"2026-04-05T23:32:43",[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":67,"owner_name":75,"owner_avatar_url":76,"owner_bio":77,"owner_company":78,"owner_location":78,"owner_email":79,"owner_twitter":78,"owner_website":78,"owner_url":80,"languages":81,"stars":121,"forks":122,"last_commit_at":123,"license":124,"difficulty_score":125,"env_os":126,"env_gpu":127,"env_ram":128,"env_deps":129,"category_tags":133,"github_topics":134,"view_count":23,"oss_zip_url":78,"oss_zip_packed_at":78,"status":16,"created_at":146,"updated_at":147,"faqs":148,"releases":181},4099,"pocketpaw\u002Fpocketpaw","pocketpaw","Your AI agent in 30 seconds. Not 30 hours. Self-hosted, open-source personal AI with desktop installer, multi-agent Command Center(Deep Work), and 7-layer security. Anthropic, OpenAI, or Ollama.","PocketPaw 是一款能在 30 秒内快速部署的本地化个人 AI 助手，旨在让用户完全掌控自己的智能代理。它解决了传统云端 AI 服务存在的数据隐私泄露风险、订阅费用高昂以及厂商锁定等痛点，确保所有数据仅存储并运行在用户自己的设备上，无需上传至第三方服务器。\n\n这款工具非常适合注重数据隐私的普通用户、希望低成本搭建私有化工作流的自由职业者，以及需要灵活集成多模型能力的开发者。PocketPaw 提供了直观的桌面客户端（支持 Windows、macOS 和 Linux），内置“深度工作”多智能体指挥中心，并能通过 Discord、Slack、WhatsApp 或 Telegram 等常用通讯软件与你互动。其核心技术亮点包括支持 Anthropic、OpenAI 及本地运行的 Ollama 等多种大模型后端，拥有七层安全防护机制，且完全开源免费。无论是想体验本地 AI 的初学者，还是寻求高度定制化方案的技术人员，PocketPaw 都能提供一个安全、高效且无订阅负担的解决方案。","\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpocketpaw_pocketpaw_readme_65a2514eaaed.png\" alt=\"PocketPaw\" width=\"100\">\n\u003C\u002Fp>\n\n\u003Ch1 align=\"center\">🐾 PocketPaw\u003C\u002Fh1>\n\n\u003Cp align=\"center\">\n  \u003Cstrong>An AI agent that runs on your machine, not someone else's.\u003C\u002Fstrong>\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fpypi.org\u002Fproject\u002Fpocketpaw\u002F\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fv\u002Fpocketpaw.svg\" alt=\"PyPI version\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fopensource.org\u002Flicenses\u002FMIT\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-yellow.svg\" alt=\"License: MIT\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fwww.python.org\u002Fdownloads\u002F\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fpython-3.11+-blue.svg\" alt=\"Python 3.11+\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fpypi.org\u002Fproject\u002Fpocketpaw\u002F\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fdm\u002Fpocketpaw.svg\" alt=\"Downloads\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Fstargazers\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fpocketpaw\u002Fpocketpaw?style=social\" alt=\"GitHub Stars\">\u003C\u002Fa>\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Freleases\u002Fdownload\u002Fclient-v0.1.3\u002FPocketPaw_0.1.3_x64-setup.exe\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FWindows-Download_.exe-0078D4?style=for-the-badge&logo=windows&logoColor=white\" alt=\"Download for Windows\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Freleases\u002Fdownload\u002Fclient-v0.1.3\u002FPocketPaw_0.1.3_aarch64.dmg\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FmacOS-Download_.dmg-000000?style=for-the-badge&logo=apple&logoColor=white\" alt=\"Download for macOS\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Freleases\u002Fdownload\u002Fclient-v0.1.3\u002FPocketPaw_0.1.3_amd64.AppImage\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLinux-Download_.AppImage-FCC624?style=for-the-badge&logo=linux&logoColor=black\" alt=\"Download for Linux\">\u003C\u002Fa>\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  Self-hosted AI agent with a native desktop app and web dashboard. Talks to you over \u003Cstrong>Discord\u003C\u002Fstrong>, \u003Cstrong>Slack\u003C\u002Fstrong>, \u003Cstrong>WhatsApp\u003C\u002Fstrong>, \u003Cstrong>Telegram\u003C\u002Fstrong>, or the browser.\u003Cbr>\n  No subscription. No cloud lock-in. Your data stays on your machine.\n\u003C\u002Fp>\n\n> ⚠️ **Beta:** This project is under active development. Expect breaking changes between versions.\n\n\u003Cp align=\"center\">\n  \u003Cvideo src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002Fa15bb8c7-6897-40d2-8111-aa905fe3fdfe\" width=\"700\" controls>\u003C\u002Fvideo>\n\u003C\u002Fp>\n\n---\n\n## Quick Start\n\n### Desktop App (Recommended)\n\nDownload the native desktop app. It bundles the backend installer and provides a full-featured UI with system tray, global shortcuts, side panel, and multi-window support.\n\n| Platform | Download |\n| --- | --- |\n| **Windows** | [PocketPaw_0.1.3_x64-setup.exe](https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Freleases\u002Fdownload\u002Fclient-v0.1.3\u002FPocketPaw_0.1.3_x64-setup.exe) |\n| **macOS (Apple Silicon)** | [PocketPaw_0.1.3_aarch64.dmg](https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Freleases\u002Fdownload\u002Fclient-v0.1.3\u002FPocketPaw_0.1.3_aarch64.dmg) |\n| **macOS (Intel)** | [PocketPaw_0.1.3_x64.dmg](https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Freleases\u002Fdownload\u002Fclient-v0.1.3\u002FPocketPaw_0.1.3_x64.dmg) |\n| **Linux (.deb)** | [PocketPaw_0.1.3_amd64.deb](https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Freleases\u002Fdownload\u002Fclient-v0.1.3\u002FPocketPaw_0.1.3_amd64.deb) |\n| **Linux (.AppImage)** | [PocketPaw_0.1.3_amd64.AppImage](https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Freleases\u002Fdownload\u002Fclient-v0.1.3\u002FPocketPaw_0.1.3_amd64.AppImage) |\n\n### Install via Terminal\n\n\u003Cdetails open>\n\u003Csummary>macOS \u002F Linux\u003C\u002Fsummary>\n\n**Prerequisites:**\n- Python 3.11 or higher ([download here](https:\u002F\u002Fwww.python.org\u002Fdownloads\u002F))\n- pip package manager (included with Python)\n\n**Quick install:**\n\n```bash\npip install pocketpaw && pocketpaw\n```\n\n**Recommended install (with virtual environment):**\n\n```bash\n# 1. Verify Python version (must be 3.11+)\npython3 --version\n\n# 2. Upgrade pip to latest version\npython3 -m pip install --upgrade pip\n\n# 3. Create and activate virtual environment (optional but recommended)\npython3 -m venv pocketpaw-env\nsource pocketpaw-env\u002Fbin\u002Factivate\n\n# 4. Install PocketPaw\npip install pocketpaw\n\n# 5. Run PocketPaw\npocketpaw\n```\n\n**Or use the automated install script:**\n\n```bash\ncurl -fsSL https:\u002F\u002Fpocketpaw.xyz\u002Finstall.sh | sh\n```\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>Windows (PowerShell)\u003C\u002Fsummary>\n\n### Windows Installation (Step-by-Step)\n\n**Prerequisites:**\n- **Python 3.11 or higher** — [Download Python](https:\u002F\u002Fwww.python.org\u002Fdownloads\u002F). During setup, check **\"Add Python to PATH\"**.\n- pip (included with Python)\n\n**Option A — Automated installer (recommended for beginners):**\n\n```powershell\npowershell -NoExit -Command \"iwr -useb https:\u002F\u002Fpocketpaw.xyz\u002Finstall.ps1 | iex\"\n```\n\n**Option B — Manual install:**\n\n1. Open PowerShell and verify Python version (must be 3.11+):\n\n   ```powershell\n   python --version\n   ```\n\n2. Upgrade pip:\n\n   ```powershell\n   python -m pip install --upgrade pip\n   ```\n\n3. (Optional but recommended) Create and activate a virtual environment:\n\n   ```powershell\n   python -m venv pocketpaw-env\n   .\\pocketpaw-env\\Scripts\\Activate.ps1\n   ```\n\n4. Install PocketPaw:\n\n   ```powershell\n   pip install pocketpaw\n   ```\n\n5. Verify the installation by running:\n\n   ```powershell\n   pocketpaw --help\n   ```\n\n   Or start the dashboard:\n\n   ```powershell\n   pocketpaw\n   ```\n\n### Windows Troubleshooting\n\nIf you installed PocketPaw with `pip install pocketpaw` and the `pocketpaw` command is not recognized:\n\n```text\n'pocketpaw' is not recognized as an internal or external command\n```\n\nThis usually means your Python Scripts directory is not on your PATH. By default it is at:\n\n```text\nC:\\Users\\\u003Cyour-username>\\AppData\\Local\\Python\\Python3.XX\\Scripts\n```\n\nFind your exact Scripts path:\n\n```powershell\npython -c \"import sysconfig; print(sysconfig.get_path('scripts'))\"\n```\n\n**How to fix:**\n\n1. Open Start → Search \"Environment Variables\"\n2. Click \"Edit the system environment variables\"\n3. Click \"Environment Variables\"\n4. Under User variables → Select `Path` → Click Edit\n5. Add the Scripts directory path\n6. Restart your terminal\n\nAlternatively, run PocketPaw without adding Scripts to PATH:\n\n```powershell\npython -m pocketpaw\n```\n\n> [!TIP]\n> **First Run:** After opening the dashboard, the system health may show **UNHEALTHY**. This is expected until at least one model provider is configured. The app itself is running correctly; only AI features are disabled. See [Setting up your first API key](#api-key-setup), or [use Ollama for free local inference](#features).\n\n> **Note:** Some features (browser automation, shell tools) work best under WSL2. Native Windows support covers the web dashboard and all LLM chat features.\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>Other methods\u003C\u002Fsummary>\n\n```bash\npipx install pocketpaw && pocketpaw    # Isolated install\nuvx pocketpaw                           # Run without installing\n\n# From source\ngit clone https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw.git\ncd pocketpaw && uv run pocketpaw\n```\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>Docker\u003C\u002Fsummary>\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw.git && cd pocketpaw\ncp .env.example .env\ndocker compose up -d\n```\n\nDashboard at `http:\u002F\u002Flocalhost:8888`. Get the access token:\n\n```bash\ndocker exec pocketpaw cat \u002Fhome\u002Fpocketpaw\u002F.pocketpaw\u002Faccess_token\n```\n\nAgent-created files appear in `.\u002Fworkspace\u002F` on the host. Optional profiles: `--profile ollama` (local LLMs), `--profile qdrant` (vector memory). Using Ollama on the host? Set `POCKETPAW_OLLAMA_HOST=http:\u002F\u002Fhost.docker.internal:11434` in `.env`.\n\n\u003C\u002Fdetails>\n\nThe web dashboard opens at `http:\u002F\u002Flocalhost:8888`. From there you can connect Discord, Slack, WhatsApp, or Telegram.\n\n\u003Ca id=\"api-key-setup\">\u003C\u002Fa>\n\n### 🔑 Setting up your first API key\n\nIf the dashboard shows **UNHEALTHY** on first run, PocketPaw is usually installed correctly. You just need to configure at least one model provider (API key or local provider).\n\n1. Open **Settings > API Keys** in the dashboard.\n2. Add a key for at least one provider:\n   - Anthropic: [console.anthropic.com\u002Fsettings\u002Fkeys](https:\u002F\u002Fconsole.anthropic.com\u002Fsettings\u002Fkeys)\n   - OpenAI: [platform.openai.com\u002Fapi-keys](https:\u002F\u002Fplatform.openai.com\u002Fapi-keys)\n   - Google Gemini: [aistudio.google.com\u002Fapikey](https:\u002F\u002Faistudio.google.com\u002Fapikey)\n3. Save settings and retry your chat.\n\n> [!TIP]\n> **Power users \u002F Docker \u002F headless:** You can also set API keys via environment variables instead of the dashboard. Use `POCKETPAW_ANTHROPIC_API_KEY`, `POCKETPAW_OPENAI_API_KEY`, or `POCKETPAW_GOOGLE_API_KEY`. See [Configuration](#configuration) for the full list.\n\nPrefer a free local option? Use Ollama instead (see [Features](#features) and [Configuration](#configuration)).\n\n---\n\n## Features\n\n| | |\n| --- | --- |\n| 📡 **9+ Channels** | Web Dashboard, Discord, Slack, WhatsApp, Telegram, Signal, Matrix, Teams, Google Chat |\n| 🧠 **6 Agent Backends** | Claude Agent SDK, OpenAI Agents, Google ADK, Codex CLI, OpenCode, Copilot SDK |\n| 🛠️ **50+ Tools** | Browser, web search, image gen, voice\u002FTTS\u002FSTT, OCR, research, delegation, skills |\n| 🔌 **Integrations** | Gmail, Calendar, Google Drive & Docs, Spotify, Reddit, MCP servers |\n| 💾 **Memory** | Long-term facts, session history, smart compaction, Mem0 semantic search |\n| 🔒 **Security** | Guardian AI, injection scanner, tool policy, plan mode, audit log, self-audit daemon |\n| 🏠 **Local-First** | Runs on your machine. Ollama for fully offline operation. macOS \u002F Windows \u002F Linux. |\n\n### Examples\n\n```\nYou:  \"Every Sunday evening, remind me which recycling bins to put out\"\nPaw:  Done. I'll check the recycling calendar and message you every Sunday at 6pm.\n\nYou:  \"Find that memory leak, the app crashes after 2 hours\"\nPaw:  Found it. The WebSocket handler never closes connections. Here's the fix.\n\nYou:  \"I need a competitor analysis report for our product launch\"\nPaw:  3 agents working on it. I'll ping you when it's ready.\n```\n\n---\n\n## Architecture\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpocketpaw_pocketpaw_readme_bdde85bd1977.webp\" alt=\"PocketPaw System Architecture\" width=\"800\">\n\u003C\u002Fp>\n\nEverything goes through an event-driven message bus. Channels publish messages, the `AgentLoop` picks them up and routes to whichever backend you've configured. All six backends implement the same `AgentBackend` protocol, so swapping one for another doesn't touch the rest of the system.\n\nThe **desktop client** (`client\u002F`) is a Tauri 2.0 + SvelteKit app that connects to the Python backend over REST and WebSocket. It provides system tray integration, global hotkeys, multi-window support (side panel, quick ask), and an onboarding wizard that handles backend installation.\n\n### Agent Backends\n\n| Backend | Key | Providers | MCP |\n| --- | --- | --- | :---: |\n| **Claude Agent SDK** (Default) | `claude_agent_sdk` | Anthropic, Ollama | Yes |\n| **OpenAI Agents SDK** | `openai_agents` | OpenAI, Ollama | No |\n| **Google ADK** | `google_adk` | Google (Gemini) | Yes |\n| **Codex CLI** | `codex_cli` | OpenAI | Yes |\n| **OpenCode** | `opencode` | External server | No |\n| **Copilot SDK** | `copilot_sdk` | Copilot, OpenAI, Azure, Anthropic | No |\n\n### Security\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpocketpaw_pocketpaw_readme_89a516612bea.webp\" alt=\"PocketPaw 7-Layer Security Stack\" width=\"500\">\n\u003C\u002Fp>\n\nA secondary LLM (Guardian AI) reviews every tool call before it runs. On top of that: injection scanning, configurable tool policies, plan mode for human approval, `--security-audit` CLI, a self-audit daemon, and an append-only audit log. [Details in the docs](https:\u002F\u002Fpocketpaw.xyz\u002Fsecurity).\n\n\u003Cdetails>\n\u003Csummary>Detailed security architecture\u003C\u002Fsummary>\n\u003Cbr>\n\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpocketpaw_pocketpaw_readme_d120b0907ffb.webp\" alt=\"PocketPaw Security Architecture (Defense-in-Depth)\" width=\"800\">\n\u003C\u002Fp>\n\u003C\u002Fdetails>\n\n---\n\n## Configuration\n\nSettings live in `~\u002F.pocketpaw\u002Fconfig.json`. You can also use `POCKETPAW_`-prefixed env vars or the dashboard Settings panel. API keys are encrypted at rest.\n\n### Environment Variables (.env file)\n\nPocketPaw ships with a `.env.example` file that lists every supported environment variable with inline comments. Before running PocketPaw (especially when cloning from source or using Docker), copy it to `.env` and fill in the values you need:\n\n```bash\n# macOS \u002F Linux\ncp .env.example .env\n\n# Windows (PowerShell)\nCopy-Item .env.example .env\n```\n\nThen open `.env` in your editor and uncomment \u002F set the variables relevant to your setup. For example:\n\n```dotenv\n# LLM provider\nPOCKETPAW_ANTHROPIC_API_KEY=sk-ant-...\nPOCKETPAW_AGENT_BACKEND=claude_agent_sdk   # or openai_agents, google_adk, etc.\n\n# Optional: Telegram bot\nPOCKETPAW_TELEGRAM_BOT_TOKEN=...\nPOCKETPAW_ALLOWED_USER_ID=...\n```\n\n> **Tip:** Variables set in `.env` are loaded automatically at startup. You can also export them directly in your shell or set them via the dashboard **Settings** panel — all three methods work interchangeably.\n\nAlternatively, export variables directly in your shell:\n\n```bash\nexport POCKETPAW_ANTHROPIC_API_KEY=\"sk-ant-...\"   # Required for Claude SDK backend\nexport POCKETPAW_AGENT_BACKEND=\"claude_agent_sdk\"  # or openai_agents, google_adk, etc.\n```\n\n> **Note:** An Anthropic API key from [console.anthropic.com](https:\u002F\u002Fconsole.anthropic.com\u002Fsettings\u002Fkeys) is required for the Claude SDK backend. OAuth tokens from Claude Free\u002FPro\u002FMax plans are [not permitted](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Flegal-and-compliance#authentication-and-credential-use) for third-party use. For free local inference, use Ollama instead.\n\nSee the [full configuration reference](https:\u002F\u002Fpocketpaw.xyz\u002Fgetting-started\u002Fconfiguration) for all settings.\n\n---\n\n## Development\n\n### Backend (Python)\n\n**Prerequisites:**\n- Python 3.11 or higher ([download here](https:\u002F\u002Fwww.python.org\u002Fdownloads\u002F))\n- [uv](https:\u002F\u002Fdocs.astral.sh\u002Fuv\u002F) package manager\n\n**Install uv:**\n\n> [!WARNING]\n> **Windows users:** You must **open a new terminal window** after running the install script below before `uv` will be recognized. The installer updates your PATH, but this change does not apply to your current terminal session — running `uv` immediately after install will give a `CommandNotFoundException` error.\n```bash\n# macOS\u002FLinux\ncurl -LsSf https:\u002F\u002Fastral.sh\u002Fuv\u002Finstall.sh | sh\n\n# Windows (PowerShell)\npowershell -ExecutionPolicy ByPass -c \"irm https:\u002F\u002Fastral.sh\u002Fuv\u002Finstall.ps1 | iex\"\n\n# Or via pip\npip install uv\n```\n\n> [!TIP]\n> **Windows:** To use `uv` immediately without opening a new terminal, run:\n> ```powershell\n> $env:Path = \"$env:USERPROFILE\\.local\\bin;$env:Path\"\n> ```\n> Then verify the installation: `uv --version`\n\n**Setup and run:**\n\n```bash\n# 1. Verify Python version\npython3 --version\n\n# 2. Clone and enter the repository\ngit clone https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw.git && cd pocketpaw\n\n# 3. Install with dev dependencies\nuv sync --dev\n\n# 4. Run PocketPaw in development mode (auto-reload)\nuv run pocketpaw --dev\n\n# 5. Run tests\nuv run pytest --ignore=tests\u002Fe2e    # Run tests (2900+)\n\n# 6. Lint & format\nuv run ruff check . && uv run ruff format .\n```\n\n\u003Cdetails>\n\u003Csummary>Optional extras\u003C\u002Fsummary>\n\n```bash\npip install pocketpaw[openai-agents]       # OpenAI Agents backend\npip install pocketpaw[google-adk]          # Google ADK backend\npip install pocketpaw[discord]             # Discord\npip install pocketpaw[slack]               # Slack\npip install pocketpaw[memory]              # Mem0 semantic memory\npip install pocketpaw[all]                 # Everything\n```\n\n\u003C\u002Fdetails>\n\n### Desktop Client (Tauri + SvelteKit)\n\nThe native desktop app lives in `client\u002F`. It connects to the Python backend via REST\u002FWebSocket.\n\n**Prerequisites:**\n- [Bun](https:\u002F\u002Fbun.sh\u002F) (package manager)\n- [Rust](https:\u002F\u002Frustup.rs\u002F) (for Tauri)\n- Python backend running on `localhost:8888`\n\n```bash\ncd client\nbun install                    # Install dependencies\nbun run dev                    # Vite dev server (http:\u002F\u002Flocalhost:1420)\nbun run tauri dev              # Full desktop app (frontend + Tauri shell)\nbun run check                  # Type check\nbun run tauri build            # Production build\n```\n\n**Tech stack:** SvelteKit 2 + Svelte 5, Tailwind CSS 4, shadcn-svelte, Tauri 2.0 (Rust). See `client\u002FCLAUDE.md` for full architecture details.\n\n---\n\n## Documentation\n\n**[pocketpaw.xyz](https:\u002F\u002Fpocketpaw.xyz)** covers getting started, backends, channels, tools, integrations, security, memory, and the full API reference.\n\n---\n\n## Star History\n\n\u003Ca href=\"https:\u002F\u002Fstar-history.com\u002F#pocketpaw\u002Fpocketpaw&Date\">\n \u003Cpicture>\n   \u003Csource media=\"(prefers-color-scheme: dark)\" srcset=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpocketpaw_pocketpaw_readme_462c031c3175.png&theme=dark\" \u002F>\n   \u003Csource media=\"(prefers-color-scheme: light)\" srcset=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpocketpaw_pocketpaw_readme_462c031c3175.png\" \u002F>\n   \u003Cimg alt=\"Star History Chart\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpocketpaw_pocketpaw_readme_462c031c3175.png\" \u002F>\n \u003C\u002Fpicture>\n\u003C\u002Fa>\n\n## Contributors\n\n\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Fgraphs\u002Fcontributors\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpocketpaw_pocketpaw_readme_4578b7d495e7.png\" alt=\"Contributors\" \u002F>\n\u003C\u002Fa>\n\n---\n\n## Join the Pack\n\n- Twitter: [@prakashd88](https:\u002F\u002Ftwitter.com\u002Fprakashd88)\n- Discord: [discord.gg\u002FasRrtm95Zc](https:\u002F\u002Fdiscord.gg\u002FasRrtm95Zc)\n- Email: pocketpawai@gmail.com\n\nPRs welcome. Come build with us.\n\n## License\n\nMIT © PocketPaw Team\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpocketpaw_pocketpaw_readme_65a2514eaaed.png\" alt=\"PocketPaw\" width=\"40\">\n  \u003Cbr>\n  \u003Cstrong>Built for people who'd rather own their AI than rent it\u003C\u002Fstrong>\n\u003C\u002Fp>\n","\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpocketpaw_pocketpaw_readme_65a2514eaaed.png\" alt=\"PocketPaw\" width=\"100\">\n\u003C\u002Fp>\n\n\u003Ch1 align=\"center\">🐾 PocketPaw\u003C\u002Fh1>\n\n\u003Cp align=\"center\">\n  \u003Cstrong>一款在您本地运行的 AI 助手，而非依赖他人的服务器。\u003C\u002Fstrong>\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fpypi.org\u002Fproject\u002Fpocketpaw\u002F\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fv\u002Fpocketpaw.svg\" alt=\"PyPI 版本\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fopensource.org\u002Flicenses\u002FMIT\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-yellow.svg\" alt=\"许可证：MIT\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fwww.python.org\u002Fdownloads\u002F\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fpython-3.11+-blue.svg\" alt=\"Python 3.11+\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fpypi.org\u002Fproject\u002Fpocketpaw\u002F\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fdm\u002Fpocketpaw.svg\" alt=\"下载量\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Fstargazers\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fpocketpaw\u002Fpocketpaw?style=social\" alt=\"GitHub 星标\">\u003C\u002Fa>\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Freleases\u002Fdownload\u002Fclient-v0.1.3\u002FPocketPaw_0.1.3_x64-setup.exe\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FWindows-Download_.exe-0078D4?style=for-the-badge&logo=windows&logoColor=white\" alt=\"Windows 下载\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Freleases\u002Fdownload\u002Fclient-v0.1.3\u002FPocketPaw_0.1.3_aarch64.dmg\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FmacOS-Download_.dmg-000000?style=for-the-badge&logo=apple&logoColor=white\" alt=\"macOS 下载\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Freleases\u002Fdownload\u002Fclient-v0.1.3\u002FPocketPaw_0.1.3_amd64.AppImage\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLinux-Download_.AppImage-FCC624?style=for-the-badge&logo=linux&logoColor=black\" alt=\"Linux 下载\">\u003C\u002Fa>\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  自托管的 AI 助手，配备原生桌面应用和 Web 控制面板。可通过 \u003Cstrong>Discord\u003C\u002Fstrong>、\u003Cstrong>Slack\u003C\u002Fstrong>、\u003Cstrong>WhatsApp\u003C\u002Fstrong>、\u003Cstrong>Telegram\u003C\u002Fstrong> 或浏览器与您交流。\u003Cbr>\n  无需订阅，无云端锁定，您的数据始终保留在本地设备上。\n\u003C\u002Fp>\n\n> ⚠️ **测试版：** 本项目仍在积极开发中，版本间可能存在不兼容的变更。\n\n\u003Cp align=\"center\">\n  \u003Cvideo src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002Fa15bb8c7-6897-40d2-8111-aa905fe3fdfe\" width=\"700\" controls>\u003C\u002Fvideo>\n\u003C\u002Fp>\n\n---\n\n## 快速入门\n\n### 桌面应用（推荐）\n\n下载原生桌面应用，它集成了后端安装程序，并提供功能齐全的用户界面，包括系统托盘、全局快捷键、侧边栏以及多窗口支持。\n\n| 平台 | 下载 |\n| --- | --- |\n| **Windows** | [PocketPaw_0.1.3_x64-setup.exe](https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Freleases\u002Fdownload\u002Fclient-v0.1.3\u002FPocketPaw_0.1.3_x64-setup.exe) |\n| **macOS（Apple Silicon）** | [PocketPaw_0.1.3_aarch64.dmg](https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Freleases\u002Fdownload\u002Fclient-v0.1.3\u002FPocketPaw_0.1.3_aarch64.dmg) |\n| **macOS（Intel）** | [PocketPaw_0.1.3_x64.dmg](https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Freleases\u002Fdownload\u002Fclient-v0.1.3\u002FPocketPaw_0.1.3_x64.dmg) |\n| **Linux (.deb)** | [PocketPaw_0.1.3_amd64.deb](https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Freleases\u002Fdownload\u002Fclient-v0.1.3\u002FPocketPaw_0.1.3_amd64.deb) |\n| **Linux (.AppImage)** | [PocketPaw_0.1.3_amd64.AppImage](https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Freleases\u002Fdownload\u002Fclient-v0.1.3\u002FPocketPaw_0.1.3_amd64.AppImage) |\n\n### 通过终端安装\n\n\u003Cdetails open>\n\u003Csummary>macOS \u002F Linux\u003C\u002Fsummary>\n\n**先决条件：**\n- Python 3.11 或更高版本（[点击此处下载](https:\u002F\u002Fwww.python.org\u002Fdownloads\u002F)）\n- pip 包管理器（随 Python 一起提供）\n\n**快速安装：**\n\n```bash\npip install pocketpaw && pocketpaw\n```\n\n**推荐安装（使用虚拟环境）：**\n\n```bash\n# 1. 验证 Python 版本（必须是 3.11+）\npython3 --version\n\n# 2. 将 pip 升级到最新版本\npython3 -m pip install --upgrade pip\n\n# 3. 创建并激活虚拟环境（可选但建议）\npython3 -m venv pocketpaw-env\nsource pocketpaw-env\u002Fbin\u002Factivate\n\n# 4. 安装 PocketPaw\npip install pocketpaw\n\n# 5. 运行 PocketPaw\npocketpaw\n```\n\n**或者使用自动化安装脚本：**\n\n```bash\ncurl -fsSL https:\u002F\u002Fpocketpaw.xyz\u002Finstall.sh | sh\n```\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>Windows（PowerShell）\u003C\u002Fsummary>\n\n### Windows 安装步骤\n\n**先决条件：**\n- **Python 3.11 或更高版本** — [下载 Python](https:\u002F\u002Fwww.python.org\u002Fdownloads\u002F)。安装时请勾选“将 Python 添加到 PATH”选项。\n- pip（随 Python 一起提供）\n\n**方案 A — 自动化安装程序（推荐给初学者）：**\n\n```powershell\npowershell -NoExit -Command \"iwr -useb https:\u002F\u002Fpocketpaw.xyz\u002Finstall.ps1 | iex\"\n```\n\n**方案 B — 手动安装：**\n\n1. 打开 PowerShell 并验证 Python 版本（必须是 3.11+）：\n\n   ```powershell\n   python --version\n   ```\n\n2. 升级 pip：\n\n   ```powershell\n   python -m pip install --upgrade pip\n   ```\n\n3. （可选但建议）创建并激活虚拟环境：\n\n   ```powershell\n   python -m venv pocketpaw-env\n   .\\pocketpaw-env\\Scripts\\Activate.ps1\n   ```\n\n4. 安装 PocketPaw：\n\n   ```powershell\n   pip install pocketpaw\n   ```\n\n5. 验证安装是否成功：\n\n   ```powershell\n   pocketpaw --help\n   ```\n\n   或者直接启动控制面板：\n\n   ```powershell\n   pocketpaw\n   ```\n\n### Windows 故障排除\n\n如果您使用 `pip install pocketpaw` 安装了 PocketPaw，但系统提示无法识别 `pocketpaw` 命令：\n\n```text\n'pocketpaw' 不是内部或外部命令\n```\n\n这通常意味着您的 Python Scripts 目录未添加到系统 PATH 中。默认路径为：\n\n```text\nC:\\Users\\\u003Cyour-username>\\AppData\\Local\\Python\\Python3.XX\\Scripts\n```\n\n您可以使用以下命令找到确切的 Scripts 路径：\n\n```powershell\npython -c \"import sysconfig; print(sysconfig.get_path('scripts'))\"\n```\n\n**解决方法：**\n\n1. 打开“开始”菜单 → 搜索“环境变量”\n2. 点击“编辑系统环境变量”\n3. 点击“环境变量”\n4. 在用户变量中选择 `Path` → 点击“编辑”\n5. 添加 Scripts 目录路径\n6. 重启终端\n\n或者，您也可以在不将 Scripts 添加到 PATH 的情况下运行 PocketPaw：\n\n```powershell\npython -m pocketpaw\n```\n\n> [!TIP]\n> **首次运行：** 打开控制面板后，系统健康状态可能会显示为 **UNHEALTHY**。这是正常现象，直到您配置至少一个模型提供商为止。应用程序本身已正常运行，只是 AI 功能暂时被禁用。请参阅【设置第一个 API 密钥】，或【使用 Ollama 进行免费的本地推理】。\n\n> **注意：** 某些功能（浏览器自动化、Shell 工具）在 WSL2 下效果最佳。而原生 Windows 支持则覆盖了 Web 控制面板及所有 LLM 对话功能。\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>其他方法\u003C\u002Fsummary>\n\n```bash\npipx install pocketpaw && pocketpaw    # 隔离式安装\nuvx pocketpaw                           # 无需安装即可运行\n\n# 从源代码安装\ngit clone https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw.git\ncd pocketpaw && uv run pocketpaw\n```\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>使用 Docker\u003C\u002Fsummary>\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw.git && cd pocketpaw\ncp .env.example .env\ndocker compose up -d\n```\n\n仪表板地址为 `http:\u002F\u002Flocalhost:8888`。获取访问令牌：\n\n```bash\ndocker exec pocketpaw cat \u002Fhome\u002Fpocketpaw\u002F.pocketpaw\u002Faccess_token\n```\n\n代理创建的文件会出现在主机上的 `.\u002Fworkspace\u002F` 目录中。可选配置：`--profile ollama`（本地大模型）、`--profile qdrant`（向量存储）。如果在主机上使用 Ollama，请在 `.env` 文件中设置 `POCKETPAW_OLLAMA_HOST=http:\u002F\u002Fhost.docker.internal:11434`。\n\n\u003C\u002Fdetails>\n\nWeb 仪表板将在 `http:\u002F\u002Flocalhost:8888` 打开。您可以通过它连接 Discord、Slack、WhatsApp 或 Telegram。\n\n\u003Ca id=\"api-key-setup\">\u003C\u002Fa>\n\n### 🔑 设置您的第一个 API 密钥\n\n如果首次运行时仪表板显示 **UNHEALTHY**，通常说明 PocketPaw 已正确安装，您只需配置至少一个模型提供商（API 密钥或本地模型）即可。\n\n1. 在仪表板中打开 **Settings > API Keys**。\n2. 为至少一个提供商添加密钥：\n   - Anthropic：[console.anthropic.com\u002Fsettings\u002Fkeys](https:\u002F\u002Fconsole.anthropic.com\u002Fsettings\u002Fkeys)\n   - OpenAI：[platform.openai.com\u002Fapi-keys](https:\u002F\u002Fplatform.openai.com\u002Fapi-keys)\n   - Google Gemini：[aistudio.google.com\u002Fapikey](https:\u002F\u002Faistudio.google.com\u002Fapikey)\n3. 保存设置并重新尝试聊天。\n\n> [!TIP]\n> **高级用户 \u002F Docker \u002F 无界面模式：** 您也可以通过环境变量而非仪表板来设置 API 密钥。使用 `POCKETPAW_ANTHROPIC_API_KEY`、`POCKETPAW_OPENAI_API_KEY` 或 `POCKETPAW_GOOGLE_API_KEY`。完整列表请参阅 [Configuration](#configuration)。\n\n如果您更倾向于免费的本地选项，可以使用 Ollama（详见 [Features](#features) 和 [Configuration](#configuration)）。\n\n---\n\n## 功能特性\n\n| | |\n| --- | --- |\n| 📡 **9+ 通道** | Web 仪表板、Discord、Slack、WhatsApp、Telegram、Signal、Matrix、Teams、Google Chat |\n| 🧠 **6 种代理后端** | Claude Agent SDK、OpenAI Agents、Google ADK、Codex CLI、OpenCode、Copilot SDK |\n| 🛠️ **50+ 工具** | 浏览器、网络搜索、图像生成、语音\u002FTTS\u002FSTT、OCR、研究、委托任务、技能等 |\n| 🔌 **集成** | Gmail、日历、Google Drive 和 Docs、Spotify、Reddit、MCP 服务器 |\n| 💾 **记忆** | 长期事实、会话历史、智能压缩、Mem0 语义搜索 |\n| 🔒 **安全** | Guardian AI、注入扫描器、工具策略、计划模式、审计日志、自审计守护进程 |\n| 🏠 **本地优先** | 在您的设备上运行。使用 Ollama 可实现完全离线操作。支持 macOS \u002F Windows \u002F Linux。 |\n\n### 示例\n\n```\n你：  “每个周日晚上提醒我该把哪些回收箱拿出来”\nPaw： 好的。我会查看回收日历，并在每周日晚上 6 点给你发消息。\n\n你：  “找到那个内存泄漏问题，应用运行两小时就会崩溃”\nPaw： 找到了。WebSocket 处理程序没有关闭连接。这是修复方案。\n\n你：  “我需要一份关于我们产品发布的竞争对手分析报告”\nPaw： 有 3 个代理正在处理。准备好后我会通知你。\n```\n\n---\n\n## 架构设计\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpocketpaw_pocketpaw_readme_bdde85bd1977.webp\" alt=\"PocketPaw 系统架构\" width=\"800\">\n\u003C\u002Fp>\n\n所有流程都通过事件驱动的消息总线进行。各个渠道发布消息，`AgentLoop` 接收并路由到您配置的后端。六种后端都实现了相同的 `AgentBackend` 协议，因此更换其中一种不会影响系统的其他部分。\n\n**桌面客户端**（`client\u002F`）是一个基于 Tauri 2.0 和 SvelteKit 的应用，通过 REST 和 WebSocket 与 Python 后端通信。它提供系统托盘集成、全局快捷键、多窗口支持（侧边栏、快速提问）以及用于处理后端安装的引导向导。\n\n### 代理后端\n\n| 后端 | 关键 | 提供商 | MCP |\n| --- | --- | --- | :---: |\n| **Claude Agent SDK**（默认） | `claude_agent_sdk` | Anthropic、Ollama | 是 |\n| **OpenAI Agents SDK** | `openai_agents` | OpenAI、Ollama | 否 |\n| **Google ADK** | `google_adk` | Google（Gemini） | 是 |\n| **Codex CLI** | `codex_cli` | OpenAI | 是 |\n| **OpenCode** | `opencode` | 外部服务器 | 否 |\n| **Copilot SDK** | `copilot_sdk` | Copilot、OpenAI、Azure、Anthropic | 否 |\n\n### 安全性\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpocketpaw_pocketpaw_readme_89a516612bea.webp\" alt=\"PocketPaw 7 层安全栈\" width=\"500\">\n\u003C\u002Fp>\n\n二级 LLM（Guardian AI）会在每次工具调用前对其进行审查。此外，还有注入扫描、可配置的工具策略、需人工批准的计划模式、`--security-audit` 命令行选项、自审计守护进程以及追加式的审计日志。更多详情请参阅文档：[pocketpaw.xyz\u002Fsecurity](https:\u002F\u002Fpocketpaw.xyz\u002Fsecurity)。\n\n\u003Cdetails>\n\u003Csummary>详细的安全架构\u003C\u002Fsummary>\n\u003Cbr>\n\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpocketpaw_pocketpaw_readme_d120b0907ffb.webp\" alt=\"PocketPaw 安全架构（纵深防御）\" width=\"800\">\n\u003C\u002Fp>\n\u003C\u002Fdetails>\n\n---\n\n## 配置\n\n设置信息存储在 `~\u002F.pocketpaw\u002Fconfig.json` 中。您也可以使用以 `POCKETPAW_` 开头的环境变量，或者通过仪表板的设置面板进行配置。API 密钥在静止状态下会被加密。\n\n### 环境变量（.env 文件）\n\nPocketPaw 自带一个 `.env.example` 文件，其中列出了所有支持的环境变量，并附有内联注释。在运行 PocketPaw 之前（尤其是从源代码克隆或使用 Docker 时），请将其复制到 `.env` 并填写所需的值：\n\n```bash\n# macOS \u002F Linux\ncp .env.example .env\n\n# Windows（PowerShell）\nCopy-Item .env.example .env\n```\n\n然后在编辑器中打开 `.env` 文件，取消注释并设置与您的配置相关的变量。例如：\n\n```dotenv\n# 大模型提供商\nPOCKETPAW_ANTHROPIC_API_KEY=sk-ant-...\nPOCKETPAW_AGENT_BACKEND=claude_agent_sdk   # 或 openai_agents、google_adk 等\n\n# 可选：Telegram 机器人\nPOCKETPAW_TELEGRAM_BOT_TOKEN=...\nPOCKETPAW_ALLOWED_USER_ID=...\n```\n\n> **提示：** 在 `.env` 中设置的变量会在启动时自动加载。您也可以直接在 shell 中导出这些变量，或者通过仪表板的 **Settings** 面板进行设置——这三种方法可以互换使用。\n\n另外，您也可以直接在 shell 中导出变量：\n\n```bash\nexport POCKETPAW_ANTHROPIC_API_KEY=\"sk-ant-...\"   # 使用 Claude SDK 后端时必需\nexport POCKETPAW_AGENT_BACKEND=\"claude_agent_sdk\"  # 或 openai_agents、google_adk 等\n```\n\n> **注意：** 使用 Claude SDK 后端时，必须提供来自 [console.anthropic.com](https:\u002F\u002Fconsole.anthropic.com\u002Fsettings\u002Fkeys) 的 Anthropic API 密钥。Claude Free\u002FPro\u002FMax 计划中的 OAuth 令牌根据条款规定不允许第三方使用。如需免费的本地推理，建议使用 Ollama。\n\n有关所有设置的详细参考，请参阅 [配置指南](https:\u002F\u002Fpocketpaw.xyz\u002Fgetting-started\u002Fconfiguration)。\n\n---\n\n## 开发\n\n### 后端（Python）\n\n**先决条件：**\n- Python 3.11 或更高版本（[在此下载](https:\u002F\u002Fwww.python.org\u002Fdownloads\u002F)）\n- [uv](https:\u002F\u002Fdocs.astral.sh\u002Fuv\u002F) 包管理器\n\n**安装 uv：**\n\n> [!WARNING]\n> **Windows 用户：** 在运行下方安装脚本后，必须 **打开一个新的终端窗口**，`uv` 才会被识别。安装程序会更新您的 PATH 环境变量，但此更改不会应用于当前的终端会话——如果在安装后立即运行 `uv`，将会出现 `CommandNotFoundException` 错误。\n```bash\n# macOS\u002FLinux\ncurl -LsSf https:\u002F\u002Fastral.sh\u002Fuv\u002Finstall.sh | sh\n\n# Windows (PowerShell)\npowershell -ExecutionPolicy ByPass -c \"irm https:\u002F\u002Fastral.sh\u002Fuv\u002Finstall.ps1 | iex\"\n\n# 或通过 pip 安装\npip install uv\n```\n\n> [!TIP]\n> **Windows：** 若要在不打开新终端的情况下立即使用 `uv`，请运行：\n> ```powershell\n> $env:Path = \"$env:USERPROFILE\\.local\\bin;$env:Path\"\n> ```\n> 然后验证安装：`uv --version`\n\n**设置与运行：**\n\n```bash\n# 1. 验证 Python 版本\npython3 --version\n\n# 2. 克隆并进入仓库\ngit clone https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw.git && cd pocketpaw\n\n# 3. 安装开发依赖\nuv sync --dev\n\n# 4. 以开发模式运行 PocketPaw（自动重载）\nuv run pocketpaw --dev\n\n# 5. 运行测试\nuv run pytest --ignore=tests\u002Fe2e    # 运行测试（2900+）\n\n# 6. 代码检查与格式化\nuv run ruff check . && uv run ruff format .\n```\n\n\u003Cdetails>\n\u003Csummary>可选扩展\u003C\u002Fsummary>\n\n```bash\npip install pocketpaw[openai-agents]       # OpenAI Agents 后端\npip install pocketpaw[google-adk]          # Google ADK 后端\npip install pocketpaw[discord]             # Discord\npip install pocketpaw[slack]               # Slack\npip install pocketpaw[memory]              # Mem0 语义记忆\npip install pocketpaw[all]                 # 全部功能\n```\n\n\u003C\u002Fdetails>\n\n### 桌面客户端（Tauri + SvelteKit）\n\n原生桌面应用位于 `client\u002F` 目录下。它通过 REST\u002FWebSocket 与 Python 后端连接。\n\n**先决条件：**\n- [Bun](https:\u002F\u002Fbun.sh\u002F)（包管理器）\n- [Rust](https:\u002F\u002Frustup.rs\u002F)（用于 Tauri）\n- Python 后端正在 `localhost:8888` 上运行\n\n```bash\ncd client\nbun install                    # 安装依赖\nbun run dev                    # Vite 开发服务器（http:\u002F\u002Flocalhost:1420）\nbun run tauri dev              # 完整桌面应用（前端 + Tauri 框架）\nbun run check                  # 类型检查\nbun run tauri build            # 生产构建\n```\n\n**技术栈：** SvelteKit 2 + Svelte 5、Tailwind CSS 4、shadcn-svelte、Tauri 2.0（Rust）。完整架构详情请参阅 `client\u002FCLAUDE.md`。\n\n---\n\n## 文档\n\n**[pocketpaw.xyz](https:\u002F\u002Fpocketpaw.xyz)** 涵盖入门指南、后端、渠道、工具、集成、安全、记忆以及完整的 API 参考。\n\n---\n\n## 星标历史\n\n\u003Ca href=\"https:\u002F\u002Fstar-history.com\u002F#pocketpaw\u002Fpocketpaw&Date\">\n \u003Cpicture>\n   \u003Csource media=\"(prefers-color-scheme: dark)\" srcset=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpocketpaw_pocketpaw_readme_462c031c3175.png&theme=dark\" \u002F>\n   \u003Csource media=\"(prefers-color-scheme: light)\" srcset=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpocketpaw_pocketpaw_readme_462c031c3175.png\" \u002F>\n   \u003Cimg alt=\"星标历史图表\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpocketpaw_pocketpaw_readme_462c031c3175.png\" \u002F>\n \u003C\u002Fpicture>\n\u003C\u002Fa>\n\n## 贡献者\n\n\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Fgraphs\u002Fcontributors\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpocketpaw_pocketpaw_readme_4578b7d495e7.png\" alt=\"贡献者\" \u002F>\n\u003C\u002Fa>\n\n---\n\n## 加入团队\n\n- Twitter：[@prakashd88](https:\u002F\u002Ftwitter.com\u002Fprakashd88)\n- Discord：[discord.gg\u002FasRrtm95Zc](https:\u002F\u002Fdiscord.gg\u002FasRrtm95Zc)\n- 邮箱：pocketpawai@gmail.com\n\n欢迎提交 PR。一起共建吧。\n\n## 许可证\n\nMIT © PocketPaw 团队\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpocketpaw_pocketpaw_readme_65a2514eaaed.png\" alt=\"PocketPaw\" width=\"40\">\n  \u003Cbr>\n  \u003Cstrong>专为那些更愿意拥有自己的 AI 而非租用它的人打造\u003C\u002Fstrong>\n\u003C\u002Fp>","# PocketPaw 快速上手指南\n\nPocketPaw 是一款运行在您本地机器上的自托管 AI 智能体。它拥有原生桌面应用和 Web 仪表盘，支持通过 Discord、Slack、WhatsApp、Telegram 或浏览器与您交互。无订阅费，无云端锁定，数据完全保留在本地。\n\n> ⚠️ **注意**：本项目目前处于 Beta 阶段，正在积极开发中，版本间可能存在不兼容的变更。\n\n## 环境准备\n\n在开始之前，请确保您的系统满足以下要求：\n\n*   **操作系统**：Windows (x64), macOS (Intel\u002FApple Silicon), 或 Linux。\n*   **Python 版本**：Python 3.11 或更高版本（终端安装必需）。\n    *   下载地址：[python.org](https:\u002F\u002Fwww.python.org\u002Fdownloads\u002F)\n    *   **Windows 用户特别注意**：安装 Python 时务必勾选 **\"Add Python to PATH\"**。\n*   **包管理器**：`pip`（随 Python 自动安装）。\n\n## 安装步骤\n\n您可以选择下载原生桌面应用（推荐）或通过终端命令安装。\n\n### 方式一：下载原生桌面应用（推荐）\n\n桌面应用集成了后端安装程序，并提供系统托盘、全局快捷键、侧边栏等完整功能。\n\n| 平台 | 下载链接 |\n| :--- | :--- |\n| **Windows** | [下载 .exe 安装包](https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Freleases\u002Fdownload\u002Fclient-v0.1.3\u002FPocketPaw_0.1.3_x64-setup.exe) |\n| **macOS (Apple Silicon)** | [下载 .dmg (M1\u002FM2\u002FM3)](https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Freleases\u002Fdownload\u002Fclient-v0.1.3\u002FPocketPaw_0.1.3_aarch64.dmg) |\n| **macOS (Intel)** | [下载 .dmg (Intel)](https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Freleases\u002Fdownload\u002Fclient-v0.1.3\u002FPocketPaw_0.1.3_x64.dmg) |\n| **Linux (.deb)** | [下载 .deb](https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Freleases\u002Fdownload\u002Fclient-v0.1.3\u002FPocketPaw_0.1.3_amd64.deb) |\n| **Linux (.AppImage)** | [下载 .AppImage](https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Freleases\u002Fdownload\u002Fclient-v0.1.3\u002FPocketPaw_0.1.3_amd64.AppImage) |\n\n### 方式二：通过终端安装 (macOS \u002F Linux)\n\n**1. 快速安装（单行命令）：**\n```bash\npip install pocketpaw && pocketpaw\n```\n\n**2. 推荐安装（使用虚拟环境）：**\n```bash\n# 1. 验证 Python 版本 (必须 3.11+)\npython3 --version\n\n# 2. 升级 pip\npython3 -m pip install --upgrade pip\n\n# 3. 创建并激活虚拟环境\npython3 -m venv pocketpaw-env\nsource pocketpaw-env\u002Fbin\u002Factivate\n\n# 4. 安装 PocketPaw\npip install pocketpaw\n\n# 5. 运行\npocketpaw\n```\n\n**3. 自动化脚本安装：**\n```bash\ncurl -fsSL https:\u002F\u002Fpocketpaw.xyz\u002Finstall.sh | sh\n```\n\n### 方式三：通过终端安装 (Windows PowerShell)\n\n**选项 A：自动安装脚本（推荐）**\n```powershell\npowershell -NoExit -Command \"iwr -useb https:\u002F\u002Fpocketpaw.xyz\u002Finstall.ps1 | iex\"\n```\n\n**选项 B：手动安装**\n```powershell\n# 1. 验证 Python 版本\npython --version\n\n# 2. 升级 pip\npython -m pip install --upgrade pip\n\n# 3. (可选) 创建虚拟环境\npython -m venv pocketpaw-env\n.\\pocketpaw-env\\Scripts\\Activate.ps1\n\n# 4. 安装 PocketPaw\npip install pocketpaw\n\n# 5. 启动仪表盘\npocketpaw\n```\n*如果运行 `pocketpaw` 提示命令未找到，请尝试使用 `python -m pocketpaw` 或将 Python Scripts 目录添加到系统环境变量 PATH 中。*\n\n### 方式四：Docker 部署\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw.git && cd pocketpaw\ncp .env.example .env\ndocker compose up -d\n```\n* 仪表盘地址：`http:\u002F\u002Flocalhost:8888`\n* 获取访问令牌：`docker exec pocketpaw cat \u002Fhome\u002Fpocketpaw\u002F.pocketpaw\u002Faccess_token`\n\n## 基本使用\n\n安装完成后，按照以下步骤配置并开始使用：\n\n### 1. 访问仪表盘\n在浏览器中打开 `http:\u002F\u002Flocalhost:8888`。\n> **提示**：首次启动时，系统健康状态可能显示为 **UNHEALTHY**。这是正常现象，因为尚未配置任何 AI 模型提供商。\n\n### 2. 配置 API Key\n要让 AI 功能正常工作，您需要配置至少一个模型提供商：\n\n1. 在仪表盘中进入 **Settings > API Keys**。\n2. 添加以下任一服务的密钥：\n   *   **Anthropic**: [console.anthropic.com\u002Fsettings\u002Fkeys](https:\u002F\u002Fconsole.anthropic.com\u002Fsettings\u002Fkeys)\n   *   **OpenAI**: [platform.openai.com\u002Fapi-keys](https:\u002F\u002Fplatform.openai.com\u002Fapi-keys)\n   *   **Google Gemini**: [aistudio.google.com\u002Fapikey](https:\u002F\u002Faistudio.google.com\u002Fapikey)\n3. 保存设置。\n\n> **本地免费方案**：如果您希望完全离线运行，可以安装 [Ollama](https:\u002F\u002Follama.com)，PocketPaw 会自动检测本地的 Ollama 服务。\n\n### 3. 开始对话\n配置完成后，您可以在仪表盘直接对话，或连接外部渠道（Discord, Slack, Telegram 等）。\n\n**使用示例：**\n\n*   **定时提醒**：\n    > 用户：\"Every Sunday evening, remind me which recycling bins to put out\"\n    > PocketPaw: \"Done. I'll check the recycling calendar and message you every Sunday at 6pm.\"\n\n*   **代码调试**：\n    > 用户：\"Find that memory leak, the app crashes after 2 hours\"\n    > PocketPaw: \"Found it. The WebSocket handler never closes connections. Here's the fix.\"\n\n*   **任务协作**：\n    > 用户：\"I need a competitor analysis report for our product launch\"\n    > PocketPaw: \"3 agents working on it. I'll ping you when it's ready.\"\n\n现在，您已经成功启动了属于您自己的本地 AI 智能体！","自由职业开发者李明需要在处理敏感客户数据的同时，利用 AI 辅助编写代码和管理日常任务，但他极度担忧云端服务的隐私泄露风险。\n\n### 没有 pocketpaw 时\n- **隐私焦虑严重**：每次向云端 AI 发送代码片段或客户数据前，都必须手动脱敏，生怕核心逻辑泄露给第三方服务商。\n- **协作割裂低效**：为了获取 AI 建议，不得不频繁切换浏览器标签页或在不同应用间复制粘贴，打断深度工作流。\n- **订阅成本高昂**：同时购买多个 AI 服务的会员费用不菲，且无法根据本地算力灵活调整模型，造成资源浪费。\n- **数据孤岛问题**：AI 对话记录分散在各个网页端，无法在本地统一归档和检索，难以形成个人知识库。\n\n### 使用 pocketpaw 后\n- **数据完全自主**：pocketpaw 部署在李明自己的电脑上，所有代码和客户数据均在本地处理，彻底消除了上传云端的隐私顾虑。\n- **多平台无缝集成**：通过 Discord 或 Slack 直接唤起 pocketpaw，在聊天窗口中即可完成代码审查和任务规划，无需离开当前工作环境。\n- **模型灵活切换**：李明可以根据任务难度，在本地 Ollama 模型和付费 API 之间自由切换，既节省了日常开销，又保证了复杂任务的智能水平。\n- **统一指挥中心**：pocketpaw 的多代理“指挥中心”将历史对话、任务状态和本地文件索引整合在一起，让知识管理变得井井有条。\n\npocketpaw 让开发者在享受顶级 AI 助力的同时，真正实现了数据主权回归与工作流程的极致融合。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fpocketpaw_pocketpaw_5774caf8.png","Pocket Paw","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fpocketpaw_198473ec.png","Your Pocket Paw",null,"pocketpawai@gmail.com","https:\u002F\u002Fgithub.com\u002Fpocketpaw",[82,86,90,94,98,102,106,110,113,117],{"name":83,"color":84,"percentage":85},"Python","#3572A5",58.7,{"name":87,"color":88,"percentage":89},"JavaScript","#f1e05a",18.7,{"name":91,"color":92,"percentage":93},"Svelte","#ff3e00",10.3,{"name":95,"color":96,"percentage":97},"HTML","#e34c26",7,{"name":99,"color":100,"percentage":101},"TypeScript","#3178c6",2.9,{"name":103,"color":104,"percentage":105},"Rust","#dea584",1.6,{"name":107,"color":108,"percentage":109},"CSS","#663399",0.3,{"name":111,"color":112,"percentage":109},"Shell","#89e051",{"name":114,"color":115,"percentage":116},"PowerShell","#012456",0.2,{"name":118,"color":119,"percentage":120},"Dockerfile","#384d54",0.1,756,290,"2026-04-05T15:41:46","MIT",1,"Linux, macOS, Windows","未说明（支持完全离线运行，可通过 Ollama 使用本地模型，具体显卡需求取决于所选本地模型）","未说明",{"notes":130,"python":131,"dependencies":132},"提供原生桌面应用（基于 Tauri 2.0 + SvelteKit）和 Docker 部署方式。支持通过 Ollama 实现完全本地化离线运行。部分功能（如浏览器自动化、Shell 工具）在 WSL2 环境下表现更佳。首次运行若未配置 API Key 或本地模型，系统健康状态会显示为'UNHEALTHY'，属正常现象。","3.11+",[67],[53,15,26],[135,136,137,138,139,140,141,142,143,144,145],"ai-agents","cli","jarvis-assistant","multi-agent-systems","ollama","open-source","personal-assistant","python","security","self-hosted","telegram-bot-ai-assistant","2026-03-27T02:49:30.150509","2026-04-06T09:24:14.625944",[149,154,159,164,169,173,177],{"id":150,"question_zh":151,"answer_zh":152,"source_url":153},18677,"如何在自托管环境中监控主机健康状态（如 CPU、内存、磁盘）？","PocketPaw v0.4.6 版本已内置 SystemInfoTool。该工具允许代理报告 CPU 使用率和负载平均值、内存（RAM）可用性、主分区磁盘空间状态以及基本系统信息（操作系统版本、运行时间）。它支持跨平台（Windows、macOS、Linux），并将 psutil 作为可选依赖；如果未安装 psutil，工具会使用 Python 标准库（如读取 \u002Fproc\u002Fuptime 或使用 shutil.disk_usage）提供基础的 OS 和磁盘信息。","https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Fissues\u002F135",{"id":155,"question_zh":156,"answer_zh":157,"source_url":158},18678,"在 Windows (NTFS) 上运行时，为什么文件权限检查会失败？","这是由于 Windows NTFS 文件系统不完全尊重 POSIX 风格的权限位掩码（如 os.chmod(0o600)），导致文件权限显示为 0o666 从而引发测试失败。该问题已在 v0.4.4_2026.02.25 版本中修复：代码现在将 os.chmod 包裹在 try\u002Fexcept 块中，并在 Windows\u002FNTFS 系统上跳过仅针对 Unix 的权限检查，改用平台感知的验证逻辑。","https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Fissues\u002F279",{"id":160,"question_zh":161,"answer_zh":162,"source_url":163},18679,"相对时间解析（例如\"in 5 minutes\"）在提醒功能中不起作用怎么办？","这是一个已知问题，已在后续更新中通过增加对\"week\u002Fweeks\"等相对时间单位的支持来修复调度器解析逻辑。用户反馈表明，在修复后的版本（如 v0.4.4）中，使用\"in 2 minutes\"创建提醒可以成功触发并在聊天视图中显示，端到端流程正常工作。如果遇到此问题，请确保升级到包含调度器修复的最新版本。","https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Fissues\u002F115",{"id":165,"question_zh":166,"answer_zh":167,"source_url":168},18680,"在 Windows 上使用 .exe 或命令安装 PocketPaw 失败如何解决？","早期版本的安装程序存在问题，但安装程序已被完全重构，原始报告的问题不再适用。如果当前仍遇到 Windows 安装问题，请参考追踪特定安装错误的 Issue（如 #209, #217, #207, #205）以获取最新解决方案，因为旧的安装错误报告已被标记为过时并关闭。","https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Fissues\u002F54",{"id":170,"question_zh":171,"answer_zh":172,"source_url":163},18681,"哪些通信渠道（Channel Adapters）和代理后端目前处于稳定化测试阶段？","根据 v0.4.5 稳定化计划，以下功能需要用户反馈以确认端到端可靠性：\n1. 渠道适配器：Discord（斜杠命令、DM）、Slack（Socket Mode）、WhatsApp（商业 API\u002F个人二维码）、Signal、Matrix（已知有接收消息问题）、Teams、Google Chat。\n2. 代理后端：Claude Agent SDK（默认）、OpenAI Agents、Google ADK、Codex CLI、OpenCode、Copilot SDK。\n建议用户在测试这些功能时报告具体问题，特别是 Matrix 的消息接收回调问题和 WhatsApp 的取消按钮 UI 问题。",{"id":174,"question_zh":175,"answer_zh":176,"source_url":163},18682,"如果 reminders.json 文件损坏会导致数据丢失吗？","在较新的修复中（见 PR #662），调度器现已改进：当检测到 reminders.json 文件损坏时，系统会记录警告并自动备份损坏的文件，而不是静默地丢失数据。这确保了即使配置文件出错，用户的历史提醒数据也能得到保留以便恢复。",{"id":178,"question_zh":179,"answer_zh":180,"source_url":163},18683,"如何处理空 API 响应导致的守护程序（Guardian）错误？","针对空 API 响应可能导致的问题，最新的修复（PR #644）引入了“故障关闭”（fail-closed）默认行为。这意味着当 Guardian 收到空的 API 响应时，它会采取安全的默认处理方式，防止因响应缺失而导致的意外行为或崩溃。",[182,187,192,197,202,207,212,217,222,227,232,237,242,247,252,257,262,267,272,277],{"id":183,"version":184,"summary_zh":185,"released_at":186},109167,"client-v0.1.4","此次更新引入了 v0.4.15 版本中的所有后端改动——A2A 协议、Deep Agents、Soul Protocol v0.2.4、七层安全加固、10 个新的 CLI 命令，以及自 v0.1.3 以来的所有修复。\n\n桌面壳本身未做任何更改。二进制文件正在构建中，很快就会在此处发布。\n\n**下载**\n准备好后，请根据您的平台下载相应文件：\n- **macOS (Apple Silicon):** `PocketPaw_0.1.4_aarch64.dmg`\n- **macOS (Intel):** `PocketPaw_0.1.4_x64.dmg`\n- **Windows:** `PocketPaw_0.1.4_x64-setup.exe` 或 `.msi`\n- **Linux (Ubuntu\u002FDebian):** `PocketPaw_0.1.4_amd64.deb`\n- **Linux (Fedora\u002FRHEL):** `PocketPaw-0.1.4-1.x86_64.rpm`\n- **Linux (AppImage):** `PocketPaw_0.1.4_amd64.AppImage`\n\n**注意：** 安装程序首次运行时，会从 PyPI 拉取最新的 `pocketpaw` Python 包。","2026-03-22T18:03:23",{"id":188,"version":189,"summary_zh":190,"released_at":191},109168,"v0.4.15","## PocketPaw v0.4.15\n\n自 v0.4.10 以来，共提交了 101 次更改。主要进行了安全加固，新增两个子系统（A2A 和 Deep Agents），并整合了大量社区贡献。\n\n### 功能特性\n\n- **A2A 协议（代理间通信）** — 完整实现 A2A v0.2.5，用于代理间的通信。包括 JSON-RPC 2.0 调度器、阻塞式及 SSE 流式任务提交、轮询\u002F取消\u002F重新订阅功能、状态机强制执行、带有代理卡片缓存的异步客户端，以及带有 SSRF 防护的 `delegate_to_a2a_agent` 工具 (#667, #510, #450)\n- **Deep Agents 后端（LangChain\u002FLangGraph）** — 新增代理后端，支持持久化执行、内置规划与子代理工具、多提供商 LLM 支持（Anthropic、OpenAI、Google、Ollama、OpenRouter、LiteLLM），并通过 langchain-mcp-adapters 集成 MCP 工具 (#684)\n- **Soul 协议 v0.2.4** — 支持重新加载、评估、生物节律跟踪、自动保存生命周期、损坏文件恢复、外部 `.soul` 文件同步。仪表板新增运行时启用\u002F禁用切换开关 (#685)\n- **Discord 服务器全局对话模式** — 机器人在所有频道回复，可按频道排除特定频道。流式分块缓冲抑制了 `[NO_RESPONSE]` 标记。支持 Docker\u002FCoolify 部署中的 Claude Code OAuth 令牌 (#683)\n- **10 个新 CLI 命令** — `doctor`、`health`、`channels`、`skills`、`sessions`、`memory`、`config`、`errors`、`logs`、`update`。全部支持 `--json` 格式，便于脚本化使用 (#675)\n- **Telegram 上的语音转语音功能** — 改进了语音笔记的用户体验，支持配置 STT 模型和提供商（OpenAI、ElevenLabs、Sarvam）(#575)\n- **VectorDB 层** — 在 `vector` 可选依赖项后添加了 ChromaDB 适配器 (#623)\n- **后端回退机制** — 当主后端出现故障时，会按顺序触发回退链 (#537)\n\n### 安全性\n\n- **七层安全加固**，新增 164 项测试 (#699)：\n  - 第一层：在注入扫描器中进行 NFKC Unicode 规范化和去除零宽字符，以抵御同形异义字符和编码攻击。\n  - 第五层：新增 33 种危险命令模式（从之前的 20 种增加），涵盖 base64 混淆、反向 Shell、crontab 注入、SSH 密钥操作、权限提升和数据外泄等。在 claude_sdk 中实现了编译后的正则表达式匹配。\n  - 第六层：在错误和异常路径上启用 Guardian AI 审计日志记录。\n  - 第七层：保证审计日志为追加模式（通过归档和轮换取代 DELETE 端点）。登录、登出、令牌再生、OAuth 刷新\u002F撤销以及 API 密钥轮换等认证事件均会被记录。\n  - 所有后端工具封装层（OpenAI、ADK、LangChain）现在都会对工具输出进行注入扫描。\n- 认证中间件重写为显式白名单机制，而非负向匹配。\n- A2A 委托中增加了 DNS 重绑定防护（httpx 传输层固定解析后的 IP 地址）。\n- A2A 任务 ID 格式验证（仅允许字母数字字符，最大长度 128 个字符）。\n- 基于 TTL 的 A2A 任务清理机制（终端任务有效期为 1 小时）。\n- `\u002Ffiles\u002Fwrite` 端点新增 `require_scope(\"files:write\")` 权限要求。\n- 对 fetch 工具路径进行 Pydantic 验证。\n- 明文密钥在升级时会自动迁移到加密存储中 (#690)。\n- 全面的敏感信息脱敏处理（覆盖 24+ 种模式）。\n\n### Soul\u002FM","2026-03-22T12:41:36",{"id":193,"version":194,"summary_zh":195,"released_at":196},109169,"client-v0.1.3","## PocketPaw 桌面版 v0.1.3（Alpha）\n\nPocketPaw 的原生桌面客户端，基于 Tauri 2.0 + SvelteKit 构建。\n\n### 新增内容\n\n- **首次安装的可靠性** — 重写了启动流程，解决了新安装时出现“无法启动 PocketPaw”的问题 (#583)\n- **生产环境 CSP** — 内容安全策略现允许 `ws:\u002F\u002Flocalhost:*` 和 `http:\u002F\u002Flocalhost:*`，确保 WebSocket 连接在各平台上稳定运行\n- **客户端日志记录** — 新增结构化日志器，用于调试连接和会话问题\n- **WebSocket 连接修复** — 后端连接重试机制更加可靠\n- **会话存储修复** — 解决了会话列表加载中的边缘情况\n- **引导流程优化** — 后端设置向导改进\n- **Rust 后端** — 扩展了 IPC 命令，以更好地检测安装状态并处理错误（commands.rs 中新增 580 行代码）\n\n### 下载\n\n| 平台 | 文件 |\n| --- | --- |\n| **macOS（Apple Silicon）** | `PocketPaw_0.1.3_aarch64.dmg` |\n| **macOS（Intel）** | `PocketPaw_0.1.3_x64.dmg` |\n| **Windows** | `PocketPaw_0.1.3_x64-setup.exe` 或 `.msi` |\n| **Linux** | `.AppImage`、`.deb` 或 `.rpm` |\n\n### macOS 注意事项\n该应用目前尚未经过公证。若要打开，请执行以下命令：\n```\nxattr -cr \u002FApplications\u002FPocketPaw.app\n```\n\n### 已知限制\n- 需要在本地主机的 8888 端口上运行 PocketPaw 后端服务\n- 尚未实现自动更新\n- 未进行代码签名（macOS\u002FWindows 可能会显示安全警告）\n\n### 贡献者\n\n- **@DevRohit06**（Rohit Kushwaha）— 首次安装流程、CSP 修复、IPC 命令\n- **@prakashUXtech**（Prakash Dalai）— 发布协调\n\n本版本为 Alpha 测试版。如遇问题，请前往 https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Fissues 报告。\n\n**完整变更日志：** https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Fcompare\u002Fclient-v0.1.2...client-v0.1.3","2026-03-16T18:46:27",{"id":198,"version":199,"summary_zh":200,"released_at":201},109170,"v0.4.10","## PocketPaw v0.4.10\n\n自 v0.4.9 以来共提交了 35 次更改。新增功能、安全加固以及大量 bug 修复。\n\n### 功能特性\n\n- **Soul 协议集成** — 基于 Soul 协议的持久化 AI 身份、记忆与情感。智能体可在不同会话间保持记忆，随时间发展个性，并在不同渠道间传递上下文信息 (#622)\n- **智能体原语工具** — `run_python`、`install_package`、`edit_file`、`deliver_artifact` 等工具，适用于非 Claude 后端。完整工作流：安装依赖、编写脚本、执行并返回输出文件给用户 (#582)\n- **Discord 重构** — 使用 discli 适配器替代 discord.py，以提升可靠性并增强服务器管理功能。同时包含 Docker 部署配置 (#604、#597、#602)\n- **提供商适配器模式 + LiteLLM** — 在所有后端中实现统一的提供商抽象层。可将任意后端指向 LiteLLM 代理，从而访问 100 多种模型 (#595)\n- **智能体状态 API** — 提供实时 SSE 流和 CLI 命令，用于监控活跃会话、工具使用情况及 Token 消耗量 (#586)\n- **文件浏览器升级** — 在仪表板中支持下载文件、语法高亮预览、内联编辑以及目录压缩包下载 (#557)\n\n### 安全性\n\n- 对 `\u002Fauth\u002Fsession`、`\u002Fauth\u002Flogin` 和 `\u002Fqr` 接口实施速率限制 (#634)\n- 认证异常现已被记录，不再被静默吞掉 (#630)\n- 数值型配置字段强制应用 `gt=0`\u002F`ge=0` 约束 (#635)\n- 为 `whatsapp_mode`、`tts_provider` 和 `stt_provider` 引入 `Literal` 类型注解 (#646)\n- Soul 导入接口被沙盒隔离至 `~\u002F.pocketpaw\u002Fsoul\u002F` 目录下，杜绝路径遍历攻击\n\n### 修复内容\n\n- 桌面客户端首次安装流程及生产环境 CSP 配置 (#583)\n- 桌面客户端缺失的 `\u002Fapi\u002Fv1\u002Fws` 和 `\u002Fv1\u002Fws` WebSocket 路由 (#612)\n- 提升 Codex CLI 缓冲区限制，以处理大型 MCP 输出 (#640)\n- 修复 IP 检测中的套接字资源泄漏问题，现已采用上下文管理器 (#645)\n- 使用量统计器将 `cached_input_tokens` 纳入总量计算 (#589)\n- 为 Groq 等严格提供商优化工具 Schema 清理 (#577)\n- 展平复合可选依赖项，解决递归自我引用问题 (#564)\n- Discord 适配器：回复线程、输入指示器以及 `\u002Fkill` 边界场景的修复 (#602)\n- 最近文件 Bash 路径启发式算法 (#603)\n- 频道模态窗口安装覆盖层位置调整 (#621)\n\n### CI 与测试\n\n- 测试矩阵扩展至 Python 3.11、3.12 和 3.13 版本\n- 新增烟雾测试脚本及无头集成测试 (#571)\n- 对所有测试文件进行代码风格清理 (#572)\n\n---\n\n### 贡献者\n\n感谢本次发布中贡献代码的所有人：\n\n- **@DevRohit06**（Rohit Kushwaha）— Soul 协议集成、Discord 重构、LiteLLM 提供商重构、智能体状态 API、Codex CLI 修复、部署配置\n- **@ragini-pandey**（Ragini Pandey）— 认证速率限制、异常日志记录、配置验证、Literal 类型引入、套接字泄漏修复、依赖项展平\n- **@prakashUXtech**（Prakash Dalai）— 智能体原语工具、CI 测试矩阵、烟雾测试、代码风格修复、版本协调\n- **@DakshParnami21** — 频道模态窗口覆盖层修复\n- **@RahulChoudharyGitREPO**（Rahul Choudh","2026-03-16T18:41:50",{"id":203,"version":204,"summary_zh":205,"released_at":206},109171,"v0.4.9","## 热修复：工具在消息渠道上卡住\n\n修复了一个 bug：所有基于 Bash 的工具（内存、网络搜索、Gmail、日历等）在 Telegram、Discord、Slack 和 WhatsApp 上都会静默卡住。代理会说“我无法访问内存工具”，而不是真正执行保存操作。\n\n### 问题描述\n\nClaude SDK 后端的权限绕过功能，不小心被一个默认关闭的设置所控制。如果没有启用该设置，工具调用就会等待交互式终端的确认——而在消息渠道上，这种确认永远不会出现。这个问题自 v0.3.0 版本以来就存在，但直到添加了内存工具后才被发现。\n\n### 解决方案\n\n由于 PocketPaw 始终以无头模式运行，因此始终绕过交互式权限检查。不过，PreToolUse 安全钩子仍然会阻止危险命令的执行。\n\n### 升级方法\n\n```bash\npip install --upgrade pocketpaw\n# 或者\nuv pip install --upgrade pocketpaw\n```\n\n**完整变更日志：** https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Fcompare\u002Fv0.4.8...v0.4.9","2026-03-11T05:32:25",{"id":208,"version":209,"summary_zh":210,"released_at":211},109172,"client-v0.1.2","## PocketPaw 桌面版 v0.1.2（Alpha）\n\nPocketPaw 的原生桌面客户端——基于 Tauri 2.0 和 SvelteKit 构建。\n\n### 包含内容\n- 带后端安装程序的引导向导\n- 连接到本地 PocketPaw 后端的聊天界面\n- 会话管理、活动日志、设置\n- 带快速提问窗口的系统托盘\n- macOS 原生控制栏按钮\n- 多窗口支持（侧边栏、快速提问）\n\n### 下载链接\n\n| 平台 | 文件 |\n| --- | --- |\n| **macOS（Apple Silicon）** | `PocketPaw_0.1.2_aarch64.dmg` |\n| **macOS（Intel）** | `PocketPaw_0.1.2_x64.dmg` |\n| **Windows** | `PocketPaw_0.1.2_x64-setup.exe` 或 `.msi` |\n| **Linux** | `.AppImage`、`.deb` 或 `.rpm` |\n\n### macOS 注意事项\n该应用目前尚未经过公证。若要打开，请执行以下命令：\n```\nxattr -cr \u002FApplications\u002FPocketPaw.app\n```\n\n### 本次更新的修复\n- macOS 控制栏按钮（关闭\u002F最小化\u002F最大化）现已正确显示\n- 已解决跨平台编译错误\n- Unix 安装程序可在无交互模式下运行（无需 TTY）\n- 桌面启动器构建不再与 Python 发行版捆绑\n\n### 已知限制\n- 需要在本地主机的 8888 端口上运行 PocketPaw 后端\n- 尚未实现自动更新\n- 未进行代码签名（macOS 和 Windows 可能会显示安全警告）\n\n本版本为 Alpha 测试版——如遇问题，请前往 https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Fissues 报告。","2026-03-10T19:40:44",{"id":213,"version":214,"summary_zh":215,"released_at":216},109173,"v0.4.8","## PocketPaw v0.4.8\n\n### 安全性\n- **PII 检测与脱敏** — 扫描电子邮件、电话号码、社会安全号码、信用卡号，并可为每种类型配置不同的处理动作 (#552)\n- **基于缓冲区的流式脱敏** — 在流式传输过程中捕获跨越数据块边界的敏感信息 (#433)\n- **身份漂移预防** — 每 20 轮强化一次，以保持智能体人格的一致性 (#554)\n- **Discord 注入扫描** — 对话历史在传递给大语言模型之前会先经过注入扫描器\n- **Tauri CSP 加固** — 定义了内容安全策略，资产范围仅限于应用目录\n- **OAuth 令牌权限** — 在 Unix 系统上，令牌文件以 0o600 权限写入\n- **文件沙盒前缀绕过修复** — 正确阻止了文件沙盒中的路径遍历攻击 (#542)\n- **WebSocket 会话恢复中的路径遍历修复** (#516)\n- **工具包、API 和上下文构建器** 针对路径遍历和提示注入进行了加固\n\n### 桌面客户端（alpha 版）\n- 完整的 Tauri 2.0 + SvelteKit 桌面应用，支持 macOS、Windows 和 Linux (#502, #538)\n- OAuth 认证并自动刷新\n- 含后端设置的引导向导\n- 侧边栏资源管理器、会话管理、指标仪表板\n- 系统托盘及快速提问窗口\n- 跨平台构建修复 (#523, #524, #525, #526)\n\n### Discord\n- 对话模式 — 机器人无需提及即可自然加入群聊 (#545)\n- 状态控制 — 可通过斜杠命令设置状态和活动\n- 9 个管理员用斜杠命令（setname、setstatus、restrict、info、ping 等）\n- 频道级访问控制\n\n### 性能\n- **倒排索引** 在 FileMemoryStore 中 — O(k) 搜索取代了 O(n) 的全表扫描 (#513)\n- 会话锁垃圾回收防止字典无限增长 (#441)\n\n### 智能体后端\n- **OpenRouter** 成为 Claude SDK 和 OpenAI Agents 的一等公民提供商 (#543)\n- 非 Claude 后端也可完全访问工具 — shell 和文件系统工具不再被排除 (#543)\n- Codex CLI：通过标准输入传递提示，以避免 Windows 命令行长度限制 (#544)\n\n### 深度工作 v2\n- 目标和任务支持重试、超时和取消 (#375)\n- 输出链式调用和 PawKit 架构\n\n### 其他\n- AGENTS.md 支持项目级智能体指令 (#551)\n- 提醒消息现在会持久化到会话历史中 (#549)\n- 减少了 7 个模块中的宽泛异常处理 (#555)\n- 诊断、调度器、隧道、更新检查器等模块共修复了 5 处 minor bug (#506)\n- 文档全面升级：桌面客户端文档、SEO 指南、FAQ 结构化数据、交叉链接\n- 修复了 Anthropic API 密钥生成 URL (#553)\n- Windows 安装程序的 UTF-8 和管道兼容性修复\n\n---\n\n### 贡献者\n\n感谢所有为本次发布做出贡献的人：\n\n- **@DevRohit06**（Rohit Kushwaha）— Discord 对话模式、PII 检测、身份漂移预防、OpenRouter 支持、Codex 标准输入修复、桌面客户端功能\n- **@ragini-pandey**（Ragini Pandey）— AGENTS.md 支持、异常处理重构、提醒消息持久化\n- **@anish1301**（Anish Kumar）— 倒排内存索引、流式脱敏\n- *","2026-03-10T17:55:59",{"id":218,"version":219,"summary_zh":220,"released_at":221},109174,"v0.4.7","## 功能特性\n\n- **\u002Fkill 命令** — 用于紧急停止正在运行的代理任务。可在任何渠道（Telegram、Discord、Slack、仪表板）中使用。仅取消当前会话，不会影响所有用户的任务 (#473, #504)\n- **Telegram 键入指示器** — 机器人在处理消息时会显示“正在输入…”提示，并采用智能消息缓冲机制，将快速连续的消息分块批量合并为一次编辑 (#470)\n- **紧急按钮生效** — 仪表板上的紧急按钮现在确实会终止正在运行的代理任务，而不再是空操作 (#460)\n\n## 性能优化\n\n- **身份文件缓存** — `DefaultBootstrapProvider` 现在使用 mtime 失效机制缓存身份文件，从而减少每次消息处理时的冗余磁盘读取操作 (#472)\n- **非阻塞版本检查** — PyPI 版本检查现在线程池守护线程中异步执行，因此不会延迟启动过程 (#492)\n\n## 安全性改进\n\n- **工具级超时机制** — `ToolRegistry.execute()` 在每次调用工具时都会包裹 `asyncio.wait_for()`，默认超时时间为 60 秒。这样可以防止单个工具无限期挂起整个代理 (#495)\n- **机器人令牌脱敏** — `\u002Fsetup` 的错误响应以及 Slack 验证失败的错误信息中不再泄漏 API 令牌到 HTTP 请求体中 (#486)\n- **渠道令牌验证** — 仪表板会在保存配置前验证 Discord、Slack、Telegram 和 WhatsApp 的令牌，提前发现拼写错误，避免因配置错误导致启动失败\n- **会话所有权检查** — `switch_session` WebSocket 操作会验证会话文件是否存在，并阻止路径遍历攻击尝试\n- **Router.stop() 作用域限定于会话** — `\u002Fkill` 命令不再销毁全局路由器，而是仅取消发起请求的会话任务\n- **移除虚构模型名称** — 前端不再为未配置的后端提供占位符模型名称\n\n## 修复内容\n\n- **切换会话时 WebSocket 卡死问题** — 切换到已删除或不存在的会话时，不会再静默丢弃请求而导致前端永久卡死\n- **AgentEvent 协议** — 执行器和 paw CLI 现在使用 `.type`\u002F`.content` 属性代替 `.get()` 字典访问方式，解决了 Deep Work 任务执行过程中出现的 `AttributeError` 崩溃问题 (#482, #499)\n- **仪表板主机\u002F端口持久化** — 在设置中更改主机或端口后，新配置会在重启时正确保存并生效 (#481)\n- **Slack Socket Mode** — 在尝试连接之前会先验证 `xapp-` 和 `xoxb-` 令牌，并给出清晰的错误提示 (#465)\n- **非 UTF-8 编码的身份文件** — 现在能够优雅地处理 CRLF 换行符及非 UTF-8 字节，而不会导致程序崩溃 (#503)\n- **Deep Work 后端** — 计划器现在会使用用户配置的后端，而不是硬编码 `claude_agent_sdk` (#499)\n- **仪表板重启循环限制** — 服务器重启次数被限制为最多 5 次，以防止无限循环\n- **Settings 模型复制** — 执行器改用 `model_copy(update={...})` 方法，而非手动复制 8 个字段\n- **废弃的 asyncio API 替换** — 在 Telegram、Slack 和 Discord 适配器中，已将 `get_event_loop().time()` 替换为 `get_running_loop().time()`。\n\n## 测试覆盖\n\n- 为 Telegram 适配器新增 346 个测试用例（包括键入指示器、消息缓冲、速率限制及错误恢复等功能）\n- 为 Matrix 适配器新增 315 个测试用例（涵盖重新连接、同步等场景）","2026-03-07T19:21:34",{"id":223,"version":224,"summary_zh":225,"released_at":226},109175,"v0.4.6","## 功能\n\n- **DirectoryTreeTool** — 递归目录树列表，支持文件隔离、深度限制和隐藏文件过滤 (#282)\n- **SystemInfoTool** — 通过 `psutil` 报告 CPU、内存、磁盘、网络 I\u002FO 和 top 进程信息 (#282)\n- **Google Workspace MCP** — 新增 `gws` CLI 预设，用于集成 Gmail、日历、云端硬盘和文档，并附有设置文档\n\n## 修复\n\n- **缺少 API 密钥 → 降级** — 健康引擎现在在未设置 API 密钥时报告 `warning` 而不是 `critical`，并提供清晰的入门引导 (#359)\n- **Deep Work 模态框** — 修复“创建任务”模态框中优先级和经办人选择器不可见的问题 (#395)\n- **提醒加载动画** — 通过专用的 WebSocket 事件停止提醒错误时的无限加载动画\n- **提醒解析** — 使时间解析中的 `in` 前缀可选，这样“5 分钟”与“in 5 分钟”效果相同 (#352)\n- **Matrix 适配器** — 添加初始同步保护、重连逻辑和回调错误处理\n- **格式错误的 JSON** — `PUT \u002Fapi\u002Fidentity` 现在对无效的 JSON 请求体返回 400 错误，而非 500 错误 (#387)\n- **后端可用性** — 在检查后端可用性时，防止非 ImportError 异常\n- **MCP 预设** — 修正 `gws` 预设的包名及启动检查 (#431)\n- **Windows 安装程序** — 在 PowerShell 中强制使用 UTF-8 编码 (#413)，并在 `Get-Command` 中添加 `ErrorAction` 参数 (#425)\n- **Neonize UX** — 改进安装流程，增加重启提示 (#391)\n- **代码规范清理** — 移除未使用的导入，修复 UP038 `isinstance` 调用，解决 82 个文件中的行长度违规问题\n\n## CI 与工具链\n\n- **Pre-commit 钩子** — 提交时运行 ruff 代码检查与格式化，推送时运行 pytest\n- **CODEOWNERS** — 添加以实现自动化评审分配\n- **PR 质量门控** — 分支强制执行、大小标签、外观问题检测、敏感文件告警、依赖变更告警、安全扫描以及密钥泄露检测\n- **安全扫描工作流** — 扫描更改的 Python 文件是否存在危险模式（如 `eval`、`exec`、`os.system`、`pickle.loads` 等），并检查差异中是否泄漏了密钥\n\n## 文档\n\n- **Google Workspace MCP 集成指南** — 设置、配置和使用说明\n- **Windows 安装文档** — 针对初学者进行了优化 (#430)\n\n## 测试\n\n- 为 DirectoryTreeTool 新增 208 个测试用例（路径遍历、符号链接处理、文件隔离、深度限制）\n- 为 SystemInfoTool 新增 103 个测试用例（psutil 可用\u002F不可用、进程列表）\n- 为自然语言时间解析新增 238 个测试用例\n- 为 Matrix 适配器新增 66 个测试用例（重连、同步保护、错误处理）\n- 为 gws 集成新增 10 个 MCP 预设测试用例\n- 使用 `pytest-playwright` 依赖更新了端到端测试基础设施\n- 总计：2723 个测试通过\n\n## 贡献者\n\n- @sonwr — DirectoryTreeTool 和 SystemInfoTool (#282)\n- @AMRITESH240304 — 修复提醒加载动画、更新端到端测试\n- @Vaibhavee89 — 修复提醒解析问题 (#352)、改进 Neonize UX (#391)\n- @NidhiPednekar — 修复 `PUT \u002Fapi\u002Fidentity` 格式错误问题 (#387)\n- @YamunaKalindi — 后端可用性保护\n- @abhishekDeshmukh74 — 修复 Windows PowerShell 的 UTF-8 编码问题 (#413)\n- @abhishekabck — De","2026-03-05T17:00:32",{"id":228,"version":229,"summary_zh":230,"released_at":231},109176,"v0.4.5.1","## 热修复\n\n修复了 v0.4.5 中引入的硬性安装失败问题，即 `pip install pocketpaw[paw]` 和 `pip install pocketpaw[soul]` 会报错：\n\n```\nERROR: No matching distribution found for soul-protocol>=0.2.0\n```\n\n`soul-protocol` 尚未发布到 PyPI。已从可选依赖中移除该依赖项——如果用户需要 soul 功能，可以直接从源代码安装：\n\n```bash\npip install git+https:\u002F\u002Fgithub.com\u002Fqbtrix\u002Fsoul-protocol.git\n```\n\n基础的 `pip install pocketpaw` 并不受此问题影响。\n\n## 变更内容\n\n- `pyproject.toml`：从 `[soul]` 可选依赖中移除了 `soul-protocol>=0.2.0`","2026-03-03T14:29:48",{"id":233,"version":234,"summary_zh":235,"released_at":236},109177,"v0.4.5","## Fixes\n\n- **Auth enforcement** — enforce authentication on ALL request paths, not just \\`\u002Fapi*\\` and \\`\u002Fws*\\`; any path not on the public exemption list now requires a valid token (#325)\n- **API key validation** — validate Anthropic, OpenAI, and Telegram token formats before saving; returns warnings instead of hard-blocking (#351, #369)\n- **Session encoding** — enforce UTF-8 in all session file read\u002Fwrite operations, fixes \\`UnicodeEncodeError\\` on Windows with non-ASCII content (#381, #388)\n- **Malformed JSON** — return HTTP 400 instead of 500 for invalid JSON body in \\`\u002Fapi\u002Fidentity\\` (#374)\n- **UnboundLocalError on startup** — initialize \\`router = None\\` before the try block in the agent loop; prevents crash when memory backend fails before router is assigned (#336)\n- **Fallback backend info** — \\`get_backend_info()\\` now calls \\`self._backend.info()\\` directly; previously used a registry re-lookup that returned \\`None\\` when the configured backend had fallen back to a different one\n- **OpenAI health check** — add missing \\`else\\` branch in \\`check_llm_reachable\\` for non-200\u002F401 responses (429, 500, etc.), preventing misleading \"ok\" status on network errors (#322)\n- **Bootstrap User-Agent** — add required \\`User-Agent\\` header to GitHub API calls; anonymous requests without it get lower rate limits and silently fall back to a pinned uv version (#323)\n- **Reminder countdown** — replace static countdown display with live countdown timer (#383)\n- **Onboarding banner** — improve missing API key UX; banner logic now aligns with health check results and gives clearer setup guidance (#370)\n- **Settings session secret** — removed \\`session_secret\\` from \\`\u002Fsetup\\` response; was leaking a pairing secret to the client with no reason to\n\n## Changes\n\n- **CI pipeline** — pytest + ruff lint now run on every PR; was completely absent before (#353)\n- **PR quality gates** — auto-close PRs targeting main directly, size labels (XS\u002FS\u002FM\u002FL\u002FXL), stale bot at 21 days for PRs \u002F 44 days for issues, updated CONTRIBUTING.md (#353)\n- **Installer backends** — update backend and LLM provider lists in the interactive installer to match the current registry; removes deprecated \\`pocketpaw_native\\` and \\`open_interpreter\\` (#280)\n- **Discord server** — Discord link added across docs and project metadata\n- **Windows PATH docs** — add uv PATH troubleshooting for Windows developers; restart or manual PATH refresh required after install (#382)\n- **E2E test docs** — document the one-time Playwright browser installation step; was undocumented, causing 25 E2E tests to fail with \"fixture 'page' not found\" (#337)\n\n## Tests\n\n- 28 new edge-case tests for the memory subsystem: concurrent access, compaction thresholds, Unicode\u002Femoji, auto-learn triggers, session lifecycle (#339)\n- 22 new tests for \\`validate_api_key()\\`: valid keys, invalid prefixes, empty values, edge cases\n- New tests covering fallback backend info, API key format health check, malformed JSON in \\`\u002Fapi\u002Fidentity\\`, session encoding\n\n## Contributors\n\n- @Prajwalabhang1 — auth bypass fix, OpenAI health check, bootstrap User-Agent (#323, #322, #325)\n- @Vaibhavee89 — API key validation, UnboundLocalError fix, memory edge-case tests, Windows docs (#351, #336, #339, #382)\n- @AryaSadigale — installer backend and provider list update (#280)\n- @GuntashKaur — malformed JSON 400 response in \\`\u002Fapi\u002Fidentity\\` (#374)\n- @ritigya04 — UTF-8 encoding fixes for session file read\u002Fwrite (#381)\n- @AMRITESH240304 — live countdown timer (#383)\n- @nijil71 — onboarding UX for missing API key (#370)\n- @myukta0110 — API key validation warnings via WebSocket settings (#369)\n- @riddhivinayak — Windows UTF-8 encoding in \\`read_text()\\` calls (#388)\n- @DevRohit06 — review, merges, post-review fixes, CI setup (#353, #394)\n\n## Related\n\n**PRs:** #280, #322, #323, #325, #336, #337, #339, #351, #353, #369, #370, #374, #381, #382, #383, #388\n\n**Closes:** #33, #36, #322, #326, #333, #361, #379, #380","2026-03-02T03:22:27",{"id":238,"version":239,"summary_zh":240,"released_at":241},109178,"v0.4.4_2026.02.25","## Fixes\n\n- **Windows headless detection** — `_is_headless()` returns `False` on win32, dashboard binds correctly\n- **Windows UTF-8 console** — force UTF-8 encoding before imports, no more `UnicodeEncodeError` with emoji\n- **Windows Ctrl+C cleanup** — use `asyncio.new_event_loop()` in finally block, narrow exception scope to `(RuntimeError, OSError)`\n- **Windows NTFS chmod** — wrap `os.chmod` in try\u002Fexcept, skip Unix-only permission checks\n- **Windows uv not found** — add `LOCALAPPDATA\\uv\\bin` to PATH refresh in install.ps1 and `find_uv()`\n- **Windows pipe-to-execution** — replace `Invoke-Expression` with safe download-then-execute\n- **Windows PowerShell quoting** — quote `$PipCmd=\"uv pip\"` so installer.py gets a single argument\n- **Windows screen blinking** — add `CREATE_NO_WINDOW` flag to subprocess calls\n- **Windows install errors** — include log path and Windows-specific troubleshooting hints\n- **SSE envelope shape** — `events.py` now uses `\"event\"` key matching `chat.py` format\n- **SSE session filter** — block events without `session_key` from leaking to all SSE clients\n- **Tool param validation** — reject `None` values for required parameters\n- **WhatsApp QR import** — use canonical `dashboard_state` instead of `dashboard_channels`\n- **install_extras error** — return JSON error body instead of HTTP 500 (dashboard compat)\n- **OpenAI Agents tool name** — extract from `raw_item` instead of nonexistent `.name`\n- **AgentRouter fallback** — return correct backend info for resolved fallback\n- **Hatchling wheel build** — fix src-layout packaging with proper source remapping\n- **OAuth dynamic ports** — accept dynamic localhost ports for native desktop clients (RFC 8252)\n- **.env.example** — fix invalid model name `gpt-5.2` → `gpt-4o`\n\n## Changes\n\n- **Health heartbeat** — fire first health check on startup instead of waiting 5 minutes\n- **XSS prevention** — replace `innerHTML` with `textContent` in `showToast()`\n- **Docs URL** — updated from docs.pocketpaw.xyz to pocketpaw.xyz\n\n## Tests\n\n- 27 new tests covering SSE filter, envelope shape, OAuth redirect matching, param validation, tool name extraction, shutdown exception scope\n\n## Contributors\n\n- @DevRohit06 — Windows bug fixes, API fixes, security hardening, docs (#291, #316, #317, #318, #320)\n\n## Related\n\n**PRs:** #291, #316, #317, #318, #320, #340\n\n**Closes:** #69, #124, #148, #151, #192, #205, #207, #209, #217, #221, #240, #269, #279, #283, #284, #286, #287, #288, #293, #296, #298, #300, #303, #304, #305, #307, #308, #310, #312, #314","2026-02-25T15:37:15",{"id":243,"version":244,"summary_zh":245,"released_at":246},109179,"v0.4.4","## What's New in v0.4.4\n\n### Versioned REST API (`\u002Fapi\u002Fv1\u002F`)\nA complete external client API with 21 domain routers, Pydantic v2 request\u002Fresponse schemas, and OpenAPI docs at `\u002Fapi\u002Fv1\u002Fdocs`.\n\n- **API Key Authentication** — Create, list, revoke, and rotate API keys (`pp_*` prefix, SHA256 hash-only storage)\n- **OAuth2\u002FPKCE** — Full authorization code flow with PKCE for desktop clients (`ppat_*\u002Fpprt_*` tokens)\n- **Chat REST + SSE Streaming** — `\u002Fapi\u002Fv1\u002Fchat\u002Fstream` for real-time agent responses\n- **Events SSE** — Live system event stream for tool calls, thinking, errors\n- **Per-Key Rate Limiting** — Configurable rate limits per API key\n- **`pocketpaw serve`** — New command for API-only deployments (no dashboard, no WebSocket)\n\n### Dashboard Refactor\nBroke up the oversized `dashboard.py` (3,810 lines) into focused modules:\n- `dashboard_auth.py` — Auth middleware, token management\n- `dashboard_channels.py` — Channel adapter lifecycle and REST API\n- `dashboard_ws.py` — WebSocket handler and file helpers\n- `dashboard_lifecycle.py` — Startup\u002Fshutdown and broadcast helpers\n- `dashboard_state.py` — Shared mutable state\n\nResult: `dashboard.py` down to 1,652 lines (-57%), `__main__.py` down to 254 lines (-66%).\n\n### Docker Improvements\n- Production-ready image with workspace volume for agent-created files\n- Node.js + Claude Code CLI + Codex CLI installed (multi-stage build from official `node:22-slim`)\n- `host.docker.internal` for host-side Ollama access\n- Proper networking and environment variable support\n\n### Bug Fixes\n- **Session search** — Moved synchronous file I\u002FO off the asyncio event loop (`asyncio.to_thread`)\n- **Chat SSE** — Corrected MessageBus API calls (was using nonexistent methods)\n- **Auth middleware** — Removed ~220 lines of duplicate code left behind during extraction\n- **Health checks** — Fixed wrong package name (`claude-code-sdk` → `claude-agent-sdk`)\n- **LAN access** — Dashboard now exposes `web_host`\u002F`web_port` settings for local network access\n\n### Security\n- Output-level `REDACT_PATTERNS` for 15+ secret formats (API keys, Bearer tokens, JWT, private keys)\n- API keys stored as SHA256 hashes only — plaintext never persisted\n- OAuth tokens encrypted with Fernet, file permissions 0o600\n- Per-key and per-endpoint rate limiting\n\n### Closes\n- #259 — Versioned REST API layer\n- #260 — Break up oversized dashboard.py\n- #265 — Fresh install dependency failure\n- #108 — Session search blocking event loop\n- #239 — LAN access for dashboard\n- #238 — Docker workspace file access\n- #100 — Error handling in dashboard API\n- #116 — Live agent event streaming\n\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Fcompare\u002Fv0.4.3...v0.4.4","2026-02-21T15:01:24",{"id":248,"version":249,"summary_zh":250,"released_at":251},109180,"v0.4.3","## What's new\n\n### Multi-SDK backend architecture\n\nPocketPaw no longer ships with a single hardwired agent backend. Six backends are now available through a pluggable registry system, and you can switch between them at runtime with `\u002Fbackend \u003Cname>`.\n\n- **Claude SDK** (default) — Anthropic's official Agent SDK with built-in tools\n- **OpenAI Agents** — OpenAI Agents SDK with GPT models and Ollama support\n- **Google ADK** — Google's Agent Development Kit with native Gemini and MCP\n- **Codex CLI** — OpenAI Codex CLI subprocess wrapper\n- **OpenCode** — Go-based OpenCode server via REST API\n- **Copilot SDK** — GitHub Copilot with multi-provider BYOK support (OpenAI, Azure, Anthropic)\n\nAll backends implement a shared `AgentBackend` protocol with a `BackendInfo` dataclass and `Capability` flags (streaming, tools, MCP, multi-turn, custom system prompt). The router uses a lazy-import registry — no backend code loads until you actually use it. Legacy backend names (`pocketpaw_native`, `open_interpreter`, `claude_code`, `gemini_cli`) gracefully map to their replacements.\n\nNew slash commands: `\u002Fbackend`, `\u002Fbackends`, `\u002Fmodel`, `\u002Ftools`. Work across Telegram, Discord, Slack, and the web dashboard.\n\nA new `tool_bridge.py` wraps PocketPaw's tools for non-Anthropic backends via function signature introspection, so the same tool definitions work everywhere.\n\n### Deep Work goal parser\n\nThe \"Start Project\" flow now has a structured analysis step before planning begins. Instead of going straight from a text box to a full planning cycle, PocketPaw first runs a goal parser that extracts domain, complexity, AI vs. human capability split, and clarification questions.\n\nThe frontend shows a two-step flow: describe your goal, review the analysis, then start planning. A \"Skip\" button is there for people who know exactly what they want. Supports 7 domains (code, business, creative, education, events, home, hybrid) and maps complexity to S\u002FM\u002FL\u002FXL with estimated phase counts. Research depth defaults to \"auto\" and uses the parser's suggestion.\n\n### `--doctor` CLI\n\n`pocketpaw --doctor` runs all health checks and prints a categorized diagnostic report grouped by Config, Storage, Updates, and Connectivity. Version checking is now a health engine check — the sidebar dot turns yellow when you're behind. Terminal users get a styled ANSI box on stderr at startup (suppressed in CI and non-TTY environments).\n\n### Stop button\n\nCancel in-flight agent responses from the web dashboard. The send button swaps to a stop button during streaming, and the WebSocket `stop` action triggers `AgentLoop` cancellation.\n\n### Contribution quality gates\n\nNew issue templates (bug report, feature request) with structured forms and required fields. PR template with linked-issue requirement and testing evidence. Automated quality gate workflow that labels incomplete PRs as `needs-work`. Blank issues are disabled — all issues go through templates now.\n\n## Fixes\n\n- **Identity UI desync** (#173): Dashboard identity fields weren't loading saved backend state on page load\n- **Playwright browser install** (#233): Fixed `uv tool` isolated environment conflicts with browser binaries\n- **Installer streaming** (#188): Dependency installation output now streams to prevent the installer from appearing hung on slow connections\n- **Windows install docs** (#225): Added beginner-friendly PowerShell installation guide\n- **Broken curl install path** (#183): Removed non-functional `curl | sh` install reference, added Windows PowerShell path\n- **Architecture diagrams**: Updated security and system diagrams to correct file locations\n\n## Testing\n\n2,200 tests passing. 32 test files changed — new suites for all 6 backends, the tool bridge, backend protocol, backend registry, goal parser (60 tests), channel autostart, and updated health checks. Legacy test files for removed backends (`pocketpaw_native`, `open_interpreter`, MCP native) cleaned up.\n\n## Contributors\n\nThanks to @DevRohit06, @anjaliPatel06, @Vaibhavee89, @mishraayan, @Siddharth-singh2111, and @Vedant-Baldwa.\n\n## Install \u002F upgrade\n\n```bash\npip install --upgrade pocketpaw\n# or\nuv pip install --upgrade pocketpaw\n```","2026-02-20T03:38:27",{"id":253,"version":254,"summary_zh":255,"released_at":256},109181,"v0.4.2","## What's new\n\n### Health engine\n\nPocketPaw can tell when something is broken now. The health engine runs 11 pure Python checks against config, storage, and LLM connectivity on every startup. When things go wrong, errors land in a persistent log (JSONL with rotation) instead of vanishing as fleeting toast notifications. The agent gets three new diagnostic tools, `health_check`, `error_log`, and `config_doctor`, so when you ask \"what went wrong?\" it can actually find the answer and suggest a fix.\n\nThe dashboard shows a pulsing dot in the sidebar (green\u002Fyellow\u002Fred) and a diagnostics modal with a \"Fix Issues\" button that sends a diagnostic prompt straight to the agent. A heartbeat runs every 5 minutes and only broadcasts on status transitions. Stays quiet when things are healthy.\n\nIf the LLM itself is down, the health engine still works. All checks are pure Python, no LLM dependency.\n\n### MCP OAuth and expanded presets\n\nMCP servers that need authentication now go through a proper OAuth flow: browser popup, token persistence, callback coordination. The preset catalog went from ~35 to 55 servers across 8 new categories (finance, design, communication, analytics, CMS, marketing, monitoring, security). The old Registry tab is gone, replaced by the curated catalog.\n\nError handling improved too. `ExceptionGroup` unwrapping now shows root-cause messages instead of unhelpful anyio cancel-scope noise.\n\n### Deep Work output files\n\nNew Output Files panel in the project sidebar, between PRD and Timestamps. Shows file list with sizes and an \"Open in File Browser\" action. Auto-loads when you select a project.\n\n### Developer experience\n\n- **`--dev` flag**: `pocketpaw --dev` auto-reloads on file changes (*.py, *.html, *.js, *.css). Browser auto-open is suppressed in dev mode.\n- **Startup version check**: Checks PyPI on launch (cached 24h). Dashboard sidebar shows version + amber update banner when a newer release exists.\n- **SDK skill auto-discovery**: The Claude SDK backend now finds SKILL.md files from `~\u002F.claude\u002Fskills\u002F` and `.claude\u002Fskills\u002F`. Skill tool is properly mapped in the policy system.\n- **Claude SDK settings**: New `claude_sdk_model` and `claude_sdk_max_turns` settings, exposed in the dashboard when the SDK backend is selected.\n- **Smart routing off by default**: Was overriding Claude Code CLI's model selection. Now opt-in.\n\n## Fixes\n\n- **macOS installer crash** (#184): `curl | sh` installs on macOS threw `OSError(22, \"Invalid argument\")` because prompt_toolkit couldn't configure the terminal with stdin redirected from \u002Fdev\u002Ftty. Falls back to plain text prompts now. Also catches `EOFError` for disconnected terminals.\n- **Telegram won't start**: `ChatAction` moved from `telegram` to `telegram.constants` in python-telegram-bot v20+. The old import was swallowed by a blanket `except ImportError`, producing a misleading \"not installed\" error. Fixed, and added auto-install infrastructure with a dep-check modal in the dashboard.\n- **Windows installer**: Source parity with Unix installer, safer uv installation flow, expanded PATH handling.\n- **Auto-install cleanup**: Removed hardcoded package versions from `_EXTRA_PACKAGES`. Uses `pocketpaw[{extra}]` now so pyproject.toml is the single source of truth.\n- **Dashboard port**: Fixed outdated port references in getting-started docs.\n\n## Refactoring\n\nThe 1,698-line `mission-control.js` monolith is now 4 focused modules: `mc-agents.js` (agent CRUD), `mc-tasks.js` (task CRUD + execution), `deep-work.js` (project orchestration), and `mc-events.js` (WebSocket events). New `EventBus` for cross-module pub\u002Fsub and a deep-merge loader for shared state composition.\n\n## Documentation\n\n- Health engine concept page and 4 API endpoint docs\n- MCP OAuth flow, CIMD setup, and transport type documentation\n- Claude SDK settings and smart routing docs\n- Skill paths updated for SDK auto-discovery\n- `--dev` flag usage guide\n- Docs deploy now triggers on installer script changes\n\n## Contributors\n\nThanks to @DevRohit06, @DS-Kushagra, and @shivakarnati2004.\n\n## Install \u002F upgrade\n\n```bash\npip install --upgrade pocketpaw\n# or\nuv pip install --upgrade pocketpaw\n```","2026-02-17T19:00:52",{"id":258,"version":259,"summary_zh":260,"released_at":261},109182,"v0.4.1","## What's new\n\n- **Gemini LLM provider**: First-class support. Set your Google API key, pick a model, and go. Works with the PocketPaw Native backend (Claude SDK backend will tell you to switch if you try).\n\n- **OpenAI-compatible endpoints**: Point PocketPaw at any OpenAI-API-compatible server. LiteLLM, OpenRouter, vLLM, NVIDIA NIM all work. Three new config fields (`openai_compatible_base_url`, `openai_compatible_api_key`, `openai_compatible_model`), plus a `--check-openai-compatible` CLI flag to test connectivity.\n\n- **Sarvam AI integration**: Translation (23 Indian languages), text-to-speech (Bulbul v3), speech-to-text (Saaras v3), and OCR (Vision) via Sarvam's APIs. New `translate` tool, updated `voice`, `stt`, and `ocr` tools with Sarvam as an alternative provider.\n\n- **Performance fast-path**: Simple messages (greetings, short questions) now route through a direct API call instead of spawning a CLI subprocess. Persistent HTTP client reused across messages. Parallel I\u002FO for system prompt + memory loading. Result: ~5s down to ~1s on simple queries.\n\n- **Token usage tracking**: Token counts now show in the Activity panel after each response.\n\n## Fixes\n\n- **Deep Work planner errors**: Previously, if the LLM API failed (missing key, timeout), the planner returned \"no tasks\" with no explanation. Now surfaces the actual error message.\n- **Claude SDK OAuth**: If you have the Claude Code CLI installed and authenticated via OAuth but no explicit API key, PocketPaw now finds and uses those credentials instead of falling back to Ollama.\n- **Neonize crash prevention**: WhatsApp adapter (neonize\u002Fwhatsmeow) would crash the entire Python process with a Go panic when the network was down. Added a TCP preflight check so it raises a catchable error instead.\n- **Sarvam API key encryption**: Added to the encrypted credential store so it doesn't sit in plaintext config.\n- **Gemini error type fix**: The Claude SDK backend's Gemini incompatibility guard now yields the correct event type instead of crashing.\n- **MCP registry search**: Fixed empty results caused by payload shape mismatch.\n- **Windows `pwd` compatibility**: Shell tool returns `os.getcwd()` directly instead of running `pwd` subprocess.\n- **Guardian rails consolidation**: Dangerous-command patterns extracted into shared `security\u002Frails.py` module, used by both Claude SDK and Native backends.\n\n## Docs\n\n- **LLM providers guide**: New `docs\u002Fconcepts\u002Fllm-providers.mdx` covering all providers, backend compatibility, smart routing, and troubleshooting.\n- **Security troubleshooting**: New `docs\u002Fsecurity\u002Ftroubleshooting.mdx`.\n- **CONTRIBUTING.md overhaul**: Project structure, new tool\u002Fadapter guides, security considerations, code review expectations.\n- **Windows PowerShell install instructions** added.\n- **Beginner quick-start tip** added.\n\n## Contributors\n\nThanks to @DevRohit06, @sonijitendra, @SandeepGKP, @Gayathri-9390, @Sameer6305, @ssatviksingh, and @ranjit485.\n\n## Install \u002F upgrade\n\n```bash\npip install --upgrade pocketpaw\n# or\nuv pip install --upgrade pocketpaw\n```","2026-02-16T14:54:14",{"id":263,"version":264,"summary_zh":265,"released_at":266},109183,"v0.4.0","## Breaking changes\n\n- **Package renamed**: `pocketclaw` → `pocketpaw` everywhere. Python imports, env vars (`POCKETCLAW_*` → `POCKETPAW_*`), config directory (`~\u002F.pocketclaw\u002F` → `~\u002F.pocketpaw\u002F`). On startup, a warning prints if old config or env vars are detected, with migration instructions.\n\n- **Dashboard deps moved to core**: `fastapi`, `uvicorn`, `jinja2` are core dependencies now. `pip install pocketpaw` just works without `[dashboard]`. The extra still exists (empty) so existing scripts don't break.\n\n## What's new\n\n- **Docker support**: Multi-stage Dockerfile, `docker-compose.yml` with profiles for Ollama and Qdrant. Cookie auth (HttpOnly) for container networking, since localhost bypass doesn't apply inside Docker.\n\n- **Ollama backend**: New `LLMClient` abstraction (`src\u002Fpocketpaw\u002Fllm\u002Fclient.py`) that works across all three agent backends. Run local models without an API key.\n\n- **Session export**: Export chat sessions as JSON or Markdown from the dashboard header.\n\n- **Guardian fail-closed**: Without an API key, Guardian falls back to local regex pattern matching instead of silently allowing everything through.\n\n## Fixes\n\n- Windows installer: fixed venv targeting wrong Python version\n- Installer now uses `uv --python` to bootstrap with the correct interpreter\n- Dependency conflicts resolved (FastAPI >= 0.115.0 for starlette compat, duplicate `rich` entry removed)\n- Export button alignment fix in dashboard\n\n## Contributors\n\nThanks to @DevRohit06, @abhijit-23blaze, @Sameer6305, @aman-2482, and @goncaloalves.\n\n## Install \u002F upgrade\n\n```bash\npip install --upgrade pocketpaw\n# or\nuv pip install --upgrade pocketpaw\n```\n\nIf upgrading from v0.3.0, copy your config:\n```bash\ncp -r ~\u002F.pocketclaw\u002F* ~\u002F.pocketpaw\u002F\n```\nAnd rename any `POCKETCLAW_*` env vars to `POCKETPAW_*`.","2026-02-14T19:13:31",{"id":268,"version":269,"summary_zh":270,"released_at":271},109184,"v0.3.0","## What's New\r\n\r\n### Deep Work — Project-Based Task Management\r\n- **Planner Agent**: Natural-language project input → AI-driven research → PRD → task breakdown → team assignment\r\n- **Dependency Scheduler**: Topological task ordering, auto-dispatch on blocker resolution, concurrent execution (up to 5 tasks)\r\n- **Session Orchestrator**: Full project lifecycle — draft → planning → approval → executing → completed\r\n- **Configurable Research Depth**: None \u002F Quick \u002F Standard \u002F Deep research phases\r\n- **Manual Task Creation**: Add tasks to projects from the detail sheet\r\n- **Inline File Tree**: Expand projects in the sidebar to browse files (2-level lazy-loaded tree)\r\n- **Hash-Based URL Routing**: `#\u002Fchat`, `#\u002Fcrew`, `#\u002Fproject\u002F{id}` — survives refresh, supports browser back\u002Fforward\r\n\r\n### Session tools & Welcome Hints for channels\r\n-  Contextual welcome hints\r\n- Cross-channel session tools for Telegram, Discord, Slack, WhatsApp\r\n\r\n### Mission Control Enhancements\r\n- Project CRUD with on-disk directories (`~\u002F.pocketclaw\u002Fprojects\u002F`)\r\n- Task status: added `SKIPPED` status for bypassing tasks\r\n- Execution levels with dependency grouping for parallel dispatch\r\n- Enriched project API responses with `folder_path` and `file_count`\r\n- Task executor: project context (PRD summary, upstream deliverables) in agent prompts\r\n\r\n### Dashboard & UI\r\n- Dashboard UI overhaul: modals, identity\u002Fskills APIs, MCP management\r\n- Beta labels on experimental features (Open Interpreter, MCP)\r\n- Feature Module Loader for auto-discovery of JS modules\r\n\r\n### Channel Adapters\r\n- Signal, Matrix, Microsoft Teams, Google Chat adapters\r\n- Cross-channel command support\r\n\r\n### Integrations\r\n- MCP preset catalog + Claude SDK & Native integration\r\n- Mem0 semantic vector memory\r\n- Media: STT (Whisper), Google Drive\u002FDocs, Spotify, OCR, Reddit\r\n- OAuth framework: Google, Spotify auth flows\r\n\r\n### Infrastructure\r\n- Sender-scoped memory isolation\r\n- Session-aware chat with REST API\r\n- Autonomous channel notifier\r\n- Security: injection scanner, model router, plan mode approval\r\n- Desktop launcher: cross-platform with tray icon, auto-update\r\n- Installer: branch refs updated to `main` as primary\r\n\r\n## Bug Fixes\r\n- **Fixed 409 Conflict on task execution**: `execute_task_background()` no longer self-defeats via stale `_running_tasks` check\r\n- **Fixed stream crash**: Task execution now always uses `bypass_permissions=True` (headless mode)\r\n- **Hardened hook callback**: `PreToolUse` hook wrapped in try\u002Fexcept to prevent CLI stream teardown\r\n- Fixed app freeze when using Claude Agent SDK backend\r\n- Fixed atomic file writes (`Path.replace()` instead of `Path.rename()`)\r\n- Fixed hidden files in project browser\r\n- Fixed task status update via JSON body\r\n\r\n## Full Changelog\r\nhttps:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Fcompare\u002Fv0.2.5...v0.3.0","2026-02-12T13:40:59",{"id":273,"version":274,"summary_zh":275,"released_at":276},109185,"v0.2.5","## What's New\n\n### Multi-Channel Support\n- **Discord** — Slash command `\u002Fpaw`, DM & mention support, stream buffering with edit-in-place\n- **Slack** — Socket Mode (no public URL needed), thread support, app_mention + DM events\n- **WhatsApp Business** — Cloud API via httpx, webhook integration\n- **WhatsApp Personal** — QR-code scan pairing via neonize (experimental)\n\n### Web Dashboard Overhaul\n- Web-first UX — dashboard is now the default mode\n- Channel management UI — configure, start, and stop channels from the sidebar\n- Auto-starts all configured adapters on startup\n- REST API: `GET \u002Fapi\u002Fchannels\u002Fstatus`, `POST \u002Fapi\u002Fchannels\u002Fsave`, `POST \u002Fapi\u002Fchannels\u002Ftoggle`\n\n### Modular Dependencies\n- Optional pip extras: `pocketpaw[discord]`, `pocketpaw[slack]`, `pocketpaw[full]`\n- Genuine dependency isolation — only install what you use\n\n### Other\n- Subsystem refactor for channel adapter integration\n- Updated architecture docs with Mermaid diagrams\n- Improved README and mascot image\n\n## Install \u002F Upgrade\n\n```bash\npip install --upgrade pocketpaw        # core\npip install --upgrade pocketpaw[full]  # everything\n```\n\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Fcompare\u002Fv0.2.1...v0.2.5","2026-02-07T19:30:03",{"id":278,"version":279,"summary_zh":280,"released_at":281},109186,"v0.2.1","## What's New\n\n### Redesigned README\n- Added PocketPaw mascot branding\n- Full architecture diagram showing the data flow (Channel Adapters → Message Bus → Agent Loop \u002F Mission Control → Tools → LLM → Memory)\n- Practical use case examples across 4 categories: Daily Life, Research & Browsing, Coding & DevOps, Multi-Agent Workflows\n- Structured feature tables (Core Platform, Intelligence, Security)\n- Annotated project structure\n- Roadmap with concrete next steps\n- Collapsible install options (pip, pipx, uvx, source)\n\n### Feature Audit vs OpenClaw\n- Detailed 19-feature comparison against OpenClaw\u002FClawBot\n- Code-level audit with file references\n- Gap analysis with effort estimates\n- 4-phase priority roadmap\n\n### Includes changes from recent PRs\n- PR #10: Mission Control multi-agent orchestration\n- PR #11: Tool policy system with profiles and allow\u002Fdeny lists\n- PR #12: Dev branch merge\n\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fpocketpaw\u002Fpocketpaw\u002Fcompare\u002Fv0.2.0...v0.2.1","2026-02-06T06:53:14"]