[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-VRSEN--agency-swarm":3,"tool-VRSEN--agency-swarm":61},[4,18,26,36,44,53],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":17},4358,"openclaw","openclaw\u002Fopenclaw","OpenClaw 是一款专为个人打造的本地化 AI 助手，旨在让你在自己的设备上拥有完全可控的智能伙伴。它打破了传统 AI 助手局限于特定网页或应用的束缚，能够直接接入你日常使用的各类通讯渠道，包括微信、WhatsApp、Telegram、Discord、iMessage 等数十种平台。无论你在哪个聊天软件中发送消息，OpenClaw 都能即时响应，甚至支持在 macOS、iOS 和 Android 设备上进行语音交互，并提供实时的画布渲染功能供你操控。\n\n这款工具主要解决了用户对数据隐私、响应速度以及“始终在线”体验的需求。通过将 AI 部署在本地，用户无需依赖云端服务即可享受快速、私密的智能辅助，真正实现了“你的数据，你做主”。其独特的技术亮点在于强大的网关架构，将控制平面与核心助手分离，确保跨平台通信的流畅性与扩展性。\n\nOpenClaw 非常适合希望构建个性化工作流的技术爱好者、开发者，以及注重隐私保护且不愿被单一生态绑定的普通用户。只要具备基础的终端操作能力（支持 macOS、Linux 及 Windows WSL2），即可通过简单的命令行引导完成部署。如果你渴望拥有一个懂你",349277,3,"2026-04-06T06:32:30",[13,14,15,16],"Agent","开发框架","图像","数据工具","ready",{"id":19,"name":20,"github_repo":21,"description_zh":22,"stars":23,"difficulty_score":10,"last_commit_at":24,"category_tags":25,"status":17},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,"2026-04-05T11:01:52",[14,15,13],{"id":27,"name":28,"github_repo":29,"description_zh":30,"stars":31,"difficulty_score":32,"last_commit_at":33,"category_tags":34,"status":17},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",157379,2,"2026-04-15T23:32:42",[14,13,35],"语言模型",{"id":37,"name":38,"github_repo":39,"description_zh":40,"stars":41,"difficulty_score":32,"last_commit_at":42,"category_tags":43,"status":17},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",108322,"2026-04-10T11:39:34",[14,15,13],{"id":45,"name":46,"github_repo":47,"description_zh":48,"stars":49,"difficulty_score":32,"last_commit_at":50,"category_tags":51,"status":17},6121,"gemini-cli","google-gemini\u002Fgemini-cli","gemini-cli 是一款由谷歌推出的开源 AI 命令行工具，它将强大的 Gemini 大模型能力直接集成到用户的终端环境中。对于习惯在命令行工作的开发者而言，它提供了一条从输入提示词到获取模型响应的最短路径，无需切换窗口即可享受智能辅助。\n\n这款工具主要解决了开发过程中频繁上下文切换的痛点，让用户能在熟悉的终端界面内直接完成代码理解、生成、调试以及自动化运维任务。无论是查询大型代码库、根据草图生成应用，还是执行复杂的 Git 操作，gemini-cli 都能通过自然语言指令高效处理。\n\n它特别适合广大软件工程师、DevOps 人员及技术研究人员使用。其核心亮点包括支持高达 100 万 token 的超长上下文窗口，具备出色的逻辑推理能力；内置 Google 搜索、文件操作及 Shell 命令执行等实用工具；更独特的是，它支持 MCP（模型上下文协议），允许用户灵活扩展自定义集成，连接如图像生成等外部能力。此外，个人谷歌账号即可享受免费的额度支持，且项目基于 Apache 2.0 协议完全开源，是提升终端工作效率的理想助手。",100752,"2026-04-10T01:20:03",[52,13,15,14],"插件",{"id":54,"name":55,"github_repo":56,"description_zh":57,"stars":58,"difficulty_score":32,"last_commit_at":59,"category_tags":60,"status":17},4721,"markitdown","microsoft\u002Fmarkitdown","MarkItDown 是一款由微软 AutoGen 团队打造的轻量级 Python 工具，专为将各类文件高效转换为 Markdown 格式而设计。它支持 PDF、Word、Excel、PPT、图片（含 OCR）、音频（含语音转录）、HTML 乃至 YouTube 链接等多种格式的解析，能够精准提取文档中的标题、列表、表格和链接等关键结构信息。\n\n在人工智能应用日益普及的今天，大语言模型（LLM）虽擅长处理文本，却难以直接读取复杂的二进制办公文档。MarkItDown 恰好解决了这一痛点，它将非结构化或半结构化的文件转化为模型“原生理解”且 Token 效率极高的 Markdown 格式，成为连接本地文件与 AI 分析 pipeline 的理想桥梁。此外，它还提供了 MCP（模型上下文协议）服务器，可无缝集成到 Claude Desktop 等 LLM 应用中。\n\n这款工具特别适合开发者、数据科学家及 AI 研究人员使用，尤其是那些需要构建文档检索增强生成（RAG）系统、进行批量文本分析或希望让 AI 助手直接“阅读”本地文件的用户。虽然生成的内容也具备一定可读性，但其核心优势在于为机器",93400,"2026-04-06T19:52:38",[52,14],{"id":62,"github_repo":63,"name":64,"description_en":65,"description_zh":66,"ai_summary_zh":67,"readme_en":68,"readme_zh":69,"quickstart_zh":70,"use_case_zh":71,"hero_image_url":72,"owner_login":73,"owner_name":74,"owner_avatar_url":75,"owner_bio":76,"owner_company":76,"owner_location":76,"owner_email":76,"owner_twitter":77,"owner_website":78,"owner_url":79,"languages":80,"stars":109,"forks":110,"last_commit_at":111,"license":112,"difficulty_score":32,"env_os":113,"env_gpu":114,"env_ram":114,"env_deps":115,"category_tags":122,"github_topics":76,"view_count":32,"oss_zip_url":76,"oss_zip_packed_at":76,"status":17,"created_at":123,"updated_at":124,"faqs":125,"releases":155},7954,"VRSEN\u002Fagency-swarm","agency-swarm","Reliable Multi-Agent Orchestration Framework","Agency Swarm 是一个专为构建多智能体协作应用而设计的可靠编排框架。它基于 OpenAI Agents SDK 进行了深度扩展，旨在将现实世界中的组织架构理念引入 AI 自动化领域，让开发者能够像组建真实公司一样，定义 CEO、助理或开发者等具有特定职责的智能体角色，并让它们高效协同工作。\n\n该工具主要解决了多智能体系统中沟通混乱、状态难以管理以及行为不可控的痛点。通过明确的定向通信流机制，Agency Swarm 确保了智能体之间的交互井然有序；同时，它支持灵活的状态持久化方案，方便用户在不同会话间保存和加载对话历史，非常适合需要长期记忆的生产级应用。\n\nAgency Swarm 特别适合希望打造复杂自动化流程的 AI 开发者和技术研究人员。其独特的技术亮点包括：利用 Pydantic 模型实现类型安全的工具开发，自动验证参数以减少错误；提供对提示词（Prompts）的完全控制权，确保智能体行为精准符合预期；以及兼容多种主流大模型后端（如 Claude、Gemini 等），不仅限于 OpenAI 系列。无论是快速原型验证还是部署企业级服务，Agency Swarm 都能提供","Agency Swarm 是一个专为构建多智能体协作应用而设计的可靠编排框架。它基于 OpenAI Agents SDK 进行了深度扩展，旨在将现实世界中的组织架构理念引入 AI 自动化领域，让开发者能够像组建真实公司一样，定义 CEO、助理或开发者等具有特定职责的智能体角色，并让它们高效协同工作。\n\n该工具主要解决了多智能体系统中沟通混乱、状态难以管理以及行为不可控的痛点。通过明确的定向通信流机制，Agency Swarm 确保了智能体之间的交互井然有序；同时，它支持灵活的状态持久化方案，方便用户在不同会话间保存和加载对话历史，非常适合需要长期记忆的生产级应用。\n\nAgency Swarm 特别适合希望打造复杂自动化流程的 AI 开发者和技术研究人员。其独特的技术亮点包括：利用 Pydantic 模型实现类型安全的工具开发，自动验证参数以减少错误；提供对提示词（Prompts）的完全控制权，确保智能体行为精准符合预期；以及兼容多种主流大模型后端（如 Claude、Gemini 等），不仅限于 OpenAI 系列。无论是快速原型验证还是部署企业级服务，Agency Swarm 都能提供直观且强大的基础设施支持。","# 🐝 Agency Swarm\n\n![Framework](https:\u002F\u002Ffirebasestorage.googleapis.com\u002Fv0\u002Fb\u002Fvrsen-ai\u002Fo\u002Fpublic%2Fgithub%2FLOGO_BG_large_bold_shadow%20(1).jpg?alt=media&token=8c681331-2a7a-4a69-b21b-3ab1f9bf1a23)\n\n## Overview\n\nThe **Agency Swarm** is a framework for building multi-agent applications. It leverages and extends the [OpenAI Agents SDK](https:\u002F\u002Fgithub.com\u002Fopenai\u002Fopenai-agents-python), providing specialized features for creating, orchestrating, and managing collaborative swarms of AI agents.\n\nThis framework continues the original vision of Arsenii Shatokhin (aka VRSEN) to simplify the creation of AI agencies by thinking about automation in terms of real-world organizational structures, making it intuitive for both agents and users.\n\n**Migrating from v0.x?** Please see our [Migration Guide](https:\u002F\u002Fagency-swarm.ai\u002Fmigration\u002Fguide) for details on adapting your project to this new SDK-based version.\n\n[![Docs](https:\u002F\u002Fimg.shields.io\u002Fwebsite?label=Docs&up_message=available&url=https:\u002F\u002Fagency-swarm.ai\u002F)](https:\u002F\u002Fagency-swarm.ai)\n[![Coverage](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fcoverage-92%25-brightgreen)](https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Factions?query=branch%3Amain+event%3Apush)\n[![Subscribe on YouTube](https:\u002F\u002Fimg.shields.io\u002Fyoutube\u002Fchannel\u002Fsubscribers\u002FUCSv4qL8vmoSH7GaPjuqRiCQ)](https:\u002F\u002Fyoutube.com\u002F@vrsen\u002F)\n[![Follow on Twitter](https:\u002F\u002Fimg.shields.io\u002Ftwitter\u002Ffollow\u002F__vrsen__.svg?style=social&label=Follow%20%40__vrsen__)](https:\u002F\u002Ftwitter.com\u002F__vrsen__)\n[![Join our Discord!](https:\u002F\u002Fimg.shields.io\u002Fdiscord\u002F1200037936352202802?label=Discord)](https:\u002F\u002Fdiscord.gg\u002Fcw2xBaWfFM)\n[![Agents-as-a-Service](https:\u002F\u002Fimg.shields.io\u002Fwebsite?label=Agents-as-a-Service&up_message=For%20Business&url=https%3A%2F%2Fvrsen.ai)](https:\u002F\u002Fagents.vrsen.ai)\n\n### Key Features\n\n- **Customizable Agent Roles**: Define distinct agent roles (e.g., CEO, Virtual Assistant, Developer) with tailored instructions, tools, and capabilities within the Agency Swarm framework, leveraging the underlying OpenAI Agents SDK.\n- **Full Control Over Prompts\u002FInstructions**: Maintain complete control over each agent’s guiding prompts (instructions) for precise behavior customization.\n- **Type-Safe Tools**: Develop tools using Pydantic models for automatic argument validation, compatible with the OpenAI Agents SDK’s `FunctionTool` format.\n- **Orchestrated Agent Communication**: Agents communicate via a dedicated `send_message` tool, with interactions governed by explicit, directional `communication_flows` defined on the `Agency`.\n- **Flexible State Persistence**: Manage conversation history by providing `load_threads_callback` and `save_threads_callback` to the `Agency`, enabling persistence across sessions (e.g., DB\u002Ffile storage).\n- **Multi-Agent Orchestration**: Build agent workflows on the OpenAI Agents SDK foundation, enhanced by Agency Swarm’s structured orchestration layer.\n- **Production-Ready Focus**: Built for reliability and designed for easy deployment in real-world environments.\n\n## Installation\n\n```bash\npip install -U agency-swarm\n```\n\n> **v1.x note:** The framework targets the OpenAI Agents SDK + Responses API.\n> Migrating from v0.x? See the [Migration Guide](https:\u002F\u002Fagency-swarm.ai\u002Fmigration\u002Fguide).\n\n### Compatibility\n- **Python**: 3.12+\n- **Model backends:**\n  - **OpenAI (native):** GPT-5 family, GPT-4o, etc.\n  - **Via LiteLLM (router):** Anthropic (Claude), Google (Gemini), Grok (xAI), Azure OpenAI, **OpenRouter (gateway)**, etc.\n- **OS**: macOS, Linux, Windows\n\nIf you hit environment issues, see the [Installation guide](https:\u002F\u002Fagency-swarm.ai\u002Fwelcome\u002Finstallation).\n\n## Getting Started\n\n> **Recommended**: Start with the [Agency Starter Template](https:\u002F\u002Fgithub.com\u002Fagency-ai-solutions\u002Fagency-starter-template) before you customize anything.\n\n1. **Set Your OpenAI Key**:\n    - Create a `.env` file with `OPENAI_API_KEY=your_key` (auto-loaded), or export it in your shell:\n    ```bash\n    export OPENAI_API_KEY=\"YOUR_API_KEY\"\n    ```\n\n2. **Create Tools**:\nDefine tools using the modern `@function_tool` decorator (recommended), or extend `BaseTool` (compatible):\n    ```python\n    from agency_swarm import function_tool\n\n    @function_tool\n    def my_custom_tool(example_field: str) -> str:\n        \"\"\"A brief description of what the custom tool does.\"\"\"\n        return f\"Result: {example_field}\"\n    ```\n\n    or with `BaseTool`:\n\n    ```python\n    from agency_swarm.tools import BaseTool\n    from pydantic import Field\n\n    class MyCustomTool(BaseTool):\n        \"\"\"\n        A brief description of what the custom tool does.\n        The docstring should clearly explain the tool's purpose and functionality.\n        It will be used by the agent to determine when to use this tool.\n        \"\"\"\n\n        # Define the fields with descriptions using Pydantic Field\n        example_field: str = Field(\n            ..., description=\"Description of the example field, explaining its purpose and usage for the Agent.\"\n        )\n\n        def run(self):\n            \"\"\"\n            The implementation of the run method, where the tool's main functionality is executed.\n            \"\"\"\n            # Your custom tool logic goes here\n            # do_something(self.example_field)\n\n            # Return the result of the tool's operation\n            return \"Result of MyCustomTool operation\"\n    ```\n\n    or convert from OpenAPI schemas:\n\n    ```python\n    from agency_swarm.tools import ToolFactory\n    # using local file\n    with open(\"schemas\u002Fyour_schema.json\") as f:\n        tools = ToolFactory.from_openapi_schema(\n            f.read(),\n        )\n\n    # using requests\n    import requests\n    tools = ToolFactory.from_openapi_schema(\n        requests.get(\"https:\u002F\u002Fapi.example.com\u002Fopenapi.json\").json(),\n    )\n    ```\n\n\n3. **Define Agent Roles**: Start by defining the roles of your agents. For example, a CEO agent for managing tasks and a developer agent for executing tasks.\n\n    ```python\n    from agency_swarm import Agent, ModelSettings\n\n    ceo = Agent(\n        name=\"CEO\",\n        description=\"Responsible for client communication, task planning and management.\",\n        instructions=\"You must converse with other agents to ensure complete task execution.\", # can be a file like .\u002Finstructions.md\n        files_folder=\".\u002Ffiles\", # files to be uploaded to OpenAI\n        schemas_folder=\".\u002Fschemas\", # OpenAPI schemas to be converted into tools\n        tools=[my_custom_tool],  # FunctionTool returned by @function_tool (or adapt BaseTool via ToolFactory)\n        model=\"gpt-5.4-mini\",\n        model_settings=ModelSettings(\n            max_tokens=25000,\n        ),\n    )\n    ```\n\n    Working from examples:\n\n    - Browse `.\u002Fexamples` for runnable demos and patterns you can adapt.\n    - Use the `.cursorrules` file at the repo root with your AI coding agent (Cursor, Claude Code, etc.).\n    - Follow the Cursor IDE guide: https:\u002F\u002Fagency-swarm.ai\u002Fwelcome\u002Fgetting-started\u002Fcursor-ide\n\n\n4. **Define Agency Communication Flows**:\nEstablish how your agents will communicate with each other.\n\n    ```python\n    from agency_swarm import Agency\n    # if importing from local files\n    from Developer import Developer\n    from VirtualAssistant import VirtualAssistant\n\n    dev = Developer()\n    va = VirtualAssistant()\n\n    agency = Agency(\n        ceo,  # CEO will be the entry point for communication with the user\n        communication_flows=[\n            ceo > dev,  # CEO can initiate communication with Developer\n            ceo > va,   # CEO can initiate communication with Virtual Assistant\n            dev > va    # Developer can initiate communication with Virtual Assistant\n        ],\n        shared_instructions='agency_manifesto.md', # shared instructions for all agents\n    )\n    ```\n\n     In Agency Swarm, communication flows are directional. The `>` operator defines allowed initiations (left can initiate a chat with right).\n\n5. **Run a Demo**\n\nWeb UI:\n```python\nagency.copilot_demo()\n```\n\nTerminal:\n```python\nagency.tui()\n```\n\nOn first run, Agency Swarm sets up the terminal app automatically, shows a short setup message, and reuses it on later runs.\n\nSee the terminal workflow guide: https:\u002F\u002Fagency-swarm.ai\u002Fcore-framework\u002Fagencies\u002Fagent-swarm-cli\n\nProgrammatic (async):\n```python\nimport asyncio\n\nasync def main():\n    resp = await agency.get_response(\"Create a project skeleton.\")\n    print(resp.final_output)\n\nasyncio.run(main())\n```\n\nNeed sync? `agency.get_response_sync(...)` exists, but async is recommended.\n\n### Folder Structure\n\nRecommended agent folder structure:\n\n```\n\u002Fyour-specified-path\u002F\n│\n├── agency_manifesto.md or .txt # Agency's guiding principles (created if not present)\n└── AgentName\u002F                  # Directory for the specific agent\n    ├── files\u002F                  # Directory for files that will be uploaded to OpenAI\n    ├── schemas\u002F                # Directory for OpenAPI schemas to be converted into tools\n    ├── tools\u002F                  # Directory for tools to be imported by default.\n    ├── AgentName.py            # The main agent class file\n    ├── __init__.py             # Initializes the agent folder as a Python package\n    ├── instructions.md or .txt # Instruction document for the agent\n    └── tools.py                # Custom tools specific to the agent's role.\n\n```\n\nThis structure ensures that each agent has its dedicated space with all necessary files to start working on its specific tasks. The `tools.py` can be customized to include tools and functionalities specific to the agent's role.\n\n## Learn More\n\n- Installation: https:\u002F\u002Fagency-swarm.ai\u002Fwelcome\u002Finstallation\n- From Scratch guide: https:\u002F\u002Fagency-swarm.ai\u002Fwelcome\u002Fgetting-started\u002Ffrom-scratch\n- Cursor IDE workflow: https:\u002F\u002Fagency-swarm.ai\u002Fwelcome\u002Fgetting-started\u002Fcursor-ide\n- Tools overview: https:\u002F\u002Fagency-swarm.ai\u002Fcore-framework\u002Ftools\u002Foverview\n- Agents overview: https:\u002F\u002Fagency-swarm.ai\u002Fcore-framework\u002Fagents\u002Foverview\n- Agencies overview: https:\u002F\u002Fagency-swarm.ai\u002Fcore-framework\u002Fagencies\u002Foverview\n- Communication flows: https:\u002F\u002Fagency-swarm.ai\u002Fcore-framework\u002Fagencies\u002Fcommunication-flows\n- Running an agency: https:\u002F\u002Fagency-swarm.ai\u002Fcore-framework\u002Fagencies\u002Frunning-agency\n- Agent Swarm CLI: https:\u002F\u002Fagency-swarm.ai\u002Fcore-framework\u002Fagencies\u002Fagent-swarm-cli\n- Observability: https:\u002F\u002Fagency-swarm.ai\u002Fadditional-features\u002Fobservability\n\n## Contributing\n\nFor details on how to contribute to Agency Swarm, please refer to the [Contributing Guide](CONTRIBUTING.md).\n\n## License\n\nAgency Swarm is open-source and licensed under [MIT](https:\u002F\u002Fopensource.org\u002Flicenses\u002FMIT).\n\n\n\n## Need Help?\n\nIf you need help creating custom agent swarms for your business, check out our [Agents-as-a-Service](https:\u002F\u002Fagents.vrsen.ai\u002F) subscription, or schedule a consultation with me at https:\u002F\u002Fcalendly.com\u002Fvrsen\u002Fai-readiness-call\n","# 🐝 代理集群\n\n![框架](https:\u002F\u002Ffirebasestorage.googleapis.com\u002Fv0\u002Fb\u002Fvrsen-ai\u002Fo\u002Fpublic%2Fgithub%2FLOGO_BG_large_bold_shadow%20(1).jpg?alt=media&token=8c681331-2a7a-4a69-b21b-3ab1f9bf1a23)\n\n## 概述\n\n**代理集群**是一个用于构建多智能体应用的框架。它基于并扩展了 [OpenAI Agents SDK](https:\u002F\u002Fgithub.com\u002Fopenai\u002Fopenai-agents-python)，提供了专门的功能来创建、编排和管理协作性的智能体群。\n\n该框架延续了 Arsenii Shatokhin（又名 VRSEN）的原始愿景，即通过将自动化视为现实世界的组织结构，从而简化 AI 机构的创建过程，使智能体和用户都能直观地理解和使用。\n\n**从 v0.x 迁移吗？** 请参阅我们的 [迁移指南](https:\u002F\u002Fagency-swarm.ai\u002Fmigration\u002Fguide)，了解如何将您的项目适配到这个基于 SDK 的新版本。\n\n[![文档](https:\u002F\u002Fimg.shields.io\u002Fwebsite?label=Docs&up_message=available&url=https:\u002F\u002Fagency-swarm.ai\u002F)](https:\u002F\u002Fagency-swarm.ai)\n[![覆盖率](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fcoverage-92%25-brightgreen)](https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Factions?query=branch%3Amain+event%3Apush)\n[![订阅 YouTube 频道](https:\u002F\u002Fimg.shields.io\u002Fyoutube\u002Fchannel\u002Fsubscribers\u002FUCSv4qL8vmoSH7GaPjuqRiCQ)](https:\u002F\u002Fyoutube.com\u002F@vrsen\u002F)\n[![关注 Twitter](https:\u002F\u002Fimg.shields.io\u002Ftwitter\u002Ffollow\u002F__vrsen__.svg?style=social&label=Follow%20%40__vrsen__)](https:\u002F\u002Ftwitter.com\u002F__vrsen__)\n[![加入我们的 Discord!](https:\u002F\u002Fimg.shields.io\u002Fdiscord\u002F1200037936352202802?label=Discord)](https:\u002F\u002Fdiscord.gg\u002Fcw2xBaWfFM)\n[![代理即服务](https:\u002F\u002Fimg.shields.io\u002Fwebsite?label=Agents-as-a-Service&up_message=For%20Business&url=https%3A%2F%2Fvrsen.ai)](https:\u002F\u002Fagents.vrsen.ai)\n\n### 核心特性\n\n- **可自定义的智能体角色**：在代理集群框架中，利用底层的 OpenAI Agents SDK，定义不同的智能体角色（如 CEO、虚拟助理、开发者），并为其配备定制化的指令、工具和能力。\n- **完全掌控提示\u002F指令**：您可以完全控制每个智能体的引导提示（指令），以实现精确的行为定制。\n- **类型安全的工具**：使用 Pydantic 模型开发工具，实现自动参数验证，兼容 OpenAI Agents SDK 的 `FunctionTool` 格式。\n- **协调的智能体通信**：智能体通过专用的 `send_message` 工具进行通信，交互由 `Agency` 上明确、定向的 `communication_flows` 规则管理。\n- **灵活的状态持久化**：通过为 `Agency` 提供 `load_threads_callback` 和 `save_threads_callback`，管理对话历史，实现跨会话的持久化存储（如数据库或文件存储）。\n- **多智能体编排**：在 OpenAI Agents SDK 的基础上构建智能体工作流，并通过代理集群的结构化编排层进一步增强。\n- **面向生产环境**：专为可靠性设计，易于部署于实际环境中。\n\n## 安装\n\n```bash\npip install -U agency-swarm\n```\n\n> **v1.x 注意事项：** 该框架针对的是 OpenAI Agents SDK + Responses API。\n> 从 v0.x 迁移吗？请参阅 [迁移指南](https:\u002F\u002Fagency-swarm.ai\u002Fmigration\u002Fguide)。\n\n### 兼容性\n- **Python**：3.12 及以上版本\n- **模型后端：**\n  - **OpenAI（原生）：** GPT-5 系列、GPT-4o 等\n  - **通过 LiteLLM（路由器）：** Anthropic（Claude）、Google（Gemini）、Grok（xAI）、Azure OpenAI、**OpenRouter（网关）** 等\n- **操作系统**：macOS、Linux、Windows\n\n如果您遇到环境问题，请参阅 [安装指南](https:\u002F\u002Fagency-swarm.ai\u002Fwelcome\u002Finstallation)。\n\n## 开始使用\n\n> **推荐**：在自定义任何内容之前，请先从 [Agency Starter Template](https:\u002F\u002Fgithub.com\u002Fagency-ai-solutions\u002Fagency-starter-template) 开始。\n\n1. **设置 OpenAI API 密钥**：\n    - 创建一个包含 `OPENAI_API_KEY=your_key` 的 `.env` 文件（自动加载），或者在你的 shell 中导出它：\n    ```bash\n    export OPENAI_API_KEY=\"YOUR_API_KEY\"\n    ```\n\n2. **创建工具**：\n使用现代的 `@function_tool` 装饰器定义工具（推荐），或扩展 `BaseTool`（兼容）：\n    ```python\n    from agency_swarm import function_tool\n\n    @function_tool\n    def my_custom_tool(example_field: str) -> str:\n        \"\"\"自定义工具的简要描述。\"\"\"\n        return f\"结果：{example_field}\"\n    ```\n\n    或者使用 `BaseTool`：\n\n    ```python\n    from agency_swarm.tools import BaseTool\n    from pydantic import Field\n\n    class MyCustomTool(BaseTool):\n        \"\"\"\n        自定义工具的简要描述。\n        文档字符串应清晰说明工具的目的和功能。\n        代理将根据此文档决定何时使用该工具。\n        \"\"\"\n\n        # 使用 Pydantic Field 定义带有描述的字段\n        example_field: str = Field(\n            ..., description=\"示例字段的描述，解释其用途及对代理的意义。\"\n        )\n\n        def run(self):\n            \"\"\"\n            run 方法的实现，执行工具的主要功能。\n            \"\"\"\n            # 在这里编写自定义工具的逻辑\n            # do_something(self.example_field)\n\n            # 返回工具操作的结果\n            return \"MyCustomTool 操作的结果\"\n    ```\n\n    或者从 OpenAPI 模式转换：\n\n    ```python\n    from agency_swarm.tools import ToolFactory\n    # 使用本地文件\n    with open(\"schemas\u002Fyour_schema.json\") as f:\n        tools = ToolFactory.from_openapi_schema(\n            f.read(),\n        )\n\n    # 使用 requests\n    import requests\n    tools = ToolFactory.from_openapi_schema(\n        requests.get(\"https:\u002F\u002Fapi.example.com\u002Fopenapi.json\").json(),\n    )\n    ```\n\n\n3. **定义代理角色**：首先定义你的代理角色。例如，一个 CEO 代理用于管理任务，一个开发者代理用于执行任务。\n\n    ```python\n    from agency_swarm import Agent, ModelSettings\n\n    ceo = Agent(\n        name=\"CEO\",\n        description=\"负责客户沟通、任务规划与管理。\",\n        instructions=\"您必须与其他代理对话，以确保任务的完整执行。\"，# 可以是像 .\u002Finstructions.md 这样的文件\n        files_folder=\".\u002Ffiles\", # 将上传到 OpenAI 的文件\n        schemas_folder=\".\u002Fschemas\", # 将转换为工具的 OpenAPI 模式\n        tools=[my_custom_tool],  # 由 @function_tool 返回的 FunctionTool（或通过 ToolFactory 转换 BaseTool）\n        model=\"gpt-5.4-mini\",\n        model_settings=ModelSettings(\n            max_tokens=25000,\n        ),\n    )\n    ```\n\n    从示例入手：\n\n    - 浏览 `.\u002Fexamples` 查看可运行的演示和模式，以便进行调整。\n    - 使用仓库根目录下的 `.cursorrules` 文件，配合你的 AI 编码代理（Cursor、Claude Code 等）。\n    - 参阅 Cursor IDE 指南：https:\u002F\u002Fagency-swarm.ai\u002Fwelcome\u002Fgetting-started\u002Fcursor-ide\n\n\n4. **定义机构通信流程**：\n确定你的代理之间如何相互沟通。\n\n    ```python\n    from agency_swarm import Agency\n    # 如果从本地文件导入\n    from Developer import Developer\n    from VirtualAssistant import VirtualAssistant\n\n    dev = Developer()\n    va = VirtualAssistant()\n\n    agency = Agency(\n        ceo,  # CEO 将作为与用户沟通的入口点\n        communication_flows=[\n            ceo > dev,  # CEO 可以发起与开发者之间的沟通\n            ceo > va,   # CEO 可以发起与虚拟助理之间的沟通\n            dev > va    # 开发者可以发起与虚拟助理之间的沟通\n        ],\n        shared_instructions='agency_manifesto.md', # 所有代理共享的指令\n    )\n    ```\n\n     在 Agency Swarm 中，通信流程是单向的。`>` 运算符定义了允许的发起方（左侧可以发起与右侧的对话）。\n\n5. **运行演示**\n\nWeb UI：\n```python\nagency.copilot_demo()\n```\n\n终端：\n```python\nagency.tui()\n```\n\n首次运行时，Agency Swarm 会自动设置终端应用，显示简短的设置信息，并在后续运行中重复使用。\n\n查看终端工作流指南：https:\u002F\u002Fagency-swarm.ai\u002Fcore-framework\u002Fagencies\u002Fagent-swarm-cli\n\n程序化（异步）：\n```python\nimport asyncio\n\nasync def main():\n    resp = await agency.get_response(\"创建一个项目框架。\")\n    print(resp.final_output)\n\nasyncio.run(main())\n```\n\n需要同步吗？`agency.get_response_sync(...)` 也存在，但建议使用异步方式。\n\n### 文件夹结构\n\n推荐的代理文件夹结构如下：\n\n```\n\u002Fyour-specified-path\u002F\n│\n├── agency_manifesto.md 或 .txt # 机构的指导原则（如不存在则会创建）\n└── AgentName\u002F                  # 特定代理的目录\n    ├── files\u002F                  # 将上传到 OpenAI 的文件目录\n    ├── schemas\u002F                # 将转换为工具的 OpenAPI 模式的目录\n    ├── tools\u002F                  # 默认导入工具的目录\n    ├── AgentName.py            # 主代理类文件\n    ├── __init__.py             # 初始化代理文件夹为 Python 包\n    ├── instructions.md 或 .txt # 代理的指令文档\n    └── tools.py                # 专属于该代理角色的自定义工具。\n\n```\n\n这种结构确保每个代理都有独立的空间，包含开始执行特定任务所需的所有文件。`tools.py` 可以根据代理的角色定制工具和功能。\n\n## 了解更多\n\n- 安装：https:\u002F\u002Fagency-swarm.ai\u002Fwelcome\u002Finstallation\n- 从零开始指南：https:\u002F\u002Fagency-swarm.ai\u002Fwelcome\u002Fgetting-started\u002Ffrom-scratch\n- Cursor IDE 工作流：https:\u002F\u002Fagency-swarm.ai\u002Fwelcome\u002Fgetting-started\u002Fcursor-ide\n- 工具概述：https:\u002F\u002Fagency-swarm.ai\u002Fcore-framework\u002Ftools\u002Foverview\n- 代理概述：https:\u002F\u002Fagency-swarm.ai\u002Fcore-framework\u002Fagents\u002Foverview\n- 机构概述：https:\u002F\u002Fagency-swarm.ai\u002Fcore-framework\u002Fagencies\u002Foverview\n- 通信流程：https:\u002F\u002Fagency-swarm.ai\u002Fcore-framework\u002Fagencies\u002Fcommunication-flows\n- 运营机构：https:\u002F\u002Fagency-swarm.ai\u002Fcore-framework\u002Fagencies\u002Frunning-agency\n- Agent Swarm CLI：https:\u002F\u002Fagency-swarm.ai\u002Fcore-framework\u002Fagencies\u002Fagent-swarm-cli\n- 可观测性：https:\u002F\u002Fagency-swarm.ai\u002Fadditional-features\u002Fobservability\n\n## 贡献\n\n有关如何为 Agency Swarm 做贡献的详细信息，请参阅 [贡献指南](CONTRIBUTING.md)。\n\n## 许可证\n\nAgency Swarm 是开源项目，采用 [MIT 许可证](https:\u002F\u002Fopensource.org\u002Flicenses\u002FMIT)。\n\n## 需要帮助吗？\n\n如果您需要帮助为您的企业创建自定义智能体集群，请查看我们的[智能体即服务](https:\u002F\u002Fagents.vrsen.ai\u002F)订阅计划，或通过 https:\u002F\u002Fcalendly.com\u002Fvrsen\u002Fai-readiness-call 预约与我进行咨询。","# Agency Swarm 快速上手指南\n\nAgency Swarm 是一个基于 OpenAI Agents SDK 构建的多智能体（Multi-Agent）应用框架。它允许你通过定义真实的组织架构（如 CEO、开发者、助理等角色）来编排和协作 AI 智能体群。\n\n## 1. 环境准备\n\n在开始之前，请确保你的开发环境满足以下要求：\n\n*   **操作系统**：macOS, Linux, 或 Windows\n*   **Python 版本**：3.12 或更高版本（必须）\n*   **API Key**：你需要一个有效的 OpenAI API Key。\n    *   *注：该框架原生支持 OpenAI 模型。若需使用 Anthropic (Claude)、Google (Gemini) 或国内大模型，可通过 LiteLLM 路由进行配置。*\n\n## 2. 安装步骤\n\n使用 pip 安装最新版本的 `agency-swarm`：\n\n```bash\npip install -U agency-swarm\n```\n\n> **提示**：如果你在中国大陆地区遇到下载速度慢的问题，建议使用国内镜像源加速安装：\n> ```bash\n> pip install -U agency-swarm -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n> ```\n\n## 3. 基本使用\n\n以下是构建一个简单双智能体协作系统的最小化流程。\n\n### 第一步：配置 API Key\n\n在项目根目录创建 `.env` 文件，或在终端中导出环境变量：\n\n```bash\nexport OPENAI_API_KEY=\"YOUR_API_KEY\"\n```\n\n### 第二步：定义工具 (Tools)\n\n智能体通过工具执行具体任务。你可以使用 `@function_tool` 装饰器快速定义：\n\n```python\nfrom agency_swarm import function_tool\n\n@function_tool\ndef get_weather(city: str) -> str:\n    \"\"\"获取指定城市的天气信息。\"\"\"\n    return f\"The weather in {city} is sunny.\"\n```\n\n或者使用更复杂的 Pydantic 类定义（适用于需要严格参数校验的场景）：\n\n```python\nfrom agency_swarm.tools import BaseTool\nfrom pydantic import Field\n\nclass SearchDatabase(BaseTool):\n    \"\"\"在内部数据库中搜索相关信息。\"\"\"\n    query: str = Field(..., description=\"搜索关键词\")\n\n    def run(self):\n        # 在此处编写具体的搜索逻辑\n        return f\"Search results for: {self.query}\"\n```\n\n### 第三步：定义智能体角色 (Agents)\n\n创建具体的智能体实例，赋予其指令、工具和模型配置：\n\n```python\nfrom agency_swarm import Agent, ModelSettings\n\n# 定义一个 CEO 智能体\nceo = Agent(\n    name=\"CEO\",\n    description=\"负责与客户沟通并规划任务。\",\n    instructions=\"你必须与其他智能体协作以确保任务完成。\",\n    tools=[get_weather],  # 挂载之前定义的工具\n    model=\"gpt-4o\",       # 指定模型\n    model_settings=ModelSettings(max_tokens=25000),\n)\n\n# 定义一个开发者智能体\ndeveloper = Agent(\n    name=\"Developer\",\n    description=\"负责执行具体的代码开发任务。\",\n    instructions=\"你负责编写代码和解决技术问题。\",\n    tools=[],             # 可以挂载不同的工具集\n    model=\"gpt-4o\",\n)\n```\n\n### 第四步：组建机构并定义通信流 (Agency & Flows)\n\n将智能体组合成 `Agency`，并使用 `>` 操作符定义谁可以主动发起对话（通信流向）：\n\n```python\nfrom agency_swarm import Agency\n\nagency = Agency(\n    ceo,  # CEO 作为用户交互的入口\n    communication_flows=[\n        ceo > developer,  # CEO 可以指派任务给开发者\n        # developer > ceo, # 如果需要开发者主动汇报，可取消注释此行\n    ],\n    shared_instructions='agency_manifesto.md', # 可选：所有智能体共享的指令文件\n)\n```\n\n### 第五步：运行与交互\n\n你可以选择三种方式运行你的智能体机构：\n\n**方式 A：终端交互界面 (推荐)**\n自动启动一个交互式终端 UI：\n```python\nagency.tui()\n```\n\n**方式 B：Web 演示界面**\n启动一个简单的 Web Copilot 界面：\n```python\nagency.copilot_demo()\n```\n\n**方式 C：代码编程调用 (异步)**\n在你的应用中直接获取响应：\n```python\nimport asyncio\n\nasync def main():\n    # 发送指令并获取最终结果\n    resp = await agency.get_response(\"查询北京天气并制定开发计划。\")\n    print(resp.final_output)\n\nasyncio.run(main())\n```\n\n---\n*更多高级功能（如持久化存储、OpenAPI 工具导入、多智能体复杂编排）请参考官方文档。*","一家中型电商公司的技术团队需要构建一个自动化系统，用于每日监控社交媒体舆情、分析用户反馈并自动生成危机公关报告。\n\n### 没有 agency-swarm 时\n- **角色混乱**：开发者需手动编写复杂的条件判断代码来区分“情感分析员”和“报告撰写员”，导致逻辑耦合严重，难以维护。\n- **沟通失控**：多个 AI 实例之间缺乏标准化的消息传递机制，经常出现信息丢失或重复处理同一任务的情况。\n- **状态难持久**：一旦程序中断，之前的对话历史和中间分析结果全部丢失，无法跨会话继续执行长流程任务。\n- **工具开发繁琐**：为每个智能体定制专用工具时，缺乏统一的参数验证机制，经常因输入格式错误导致运行崩溃。\n\n### 使用 agency-swarm 后\n- **架构清晰**：利用自定义角色功能，轻松定义 CEO、分析师和写手等独立智能体，各自拥有专属指令和工具集，结构一目了然。\n- **流转可控**：通过预定义的 `communication_flows` 规范智能体间的单向通信，确保舆情数据按“采集→分析→报告”的顺序精准流转。\n- **断点续传**：内置的状态持久化回调机制自动将对话线程保存至数据库，即使服务重启也能无缝恢复之前的工作进度。\n- **类型安全**：基于 Pydantic 模型开发工具，自动校验输入参数，大幅减少了运行时错误，提升了系统的稳定性。\n\nagency-swarm 将原本杂乱无章的脚本堆砌转化为秩序井然的虚拟组织，让多智能体协作像管理真实团队一样高效可靠。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FVRSEN_agency-swarm_2c950b52.png","VRSEN","Arseny Shatokhin","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002FVRSEN_94bfc83a.jpg",null,"__vrsen__","https:\u002F\u002Fagents.vrsen.ai\u002F","https:\u002F\u002Fgithub.com\u002FVRSEN",[81,85,89,93,97,101,105],{"name":82,"color":83,"percentage":84},"Python","#3572A5",97.5,{"name":86,"color":87,"percentage":88},"JavaScript","#f1e05a",1.2,{"name":90,"color":91,"percentage":92},"CSS","#663399",0.5,{"name":94,"color":95,"percentage":96},"TypeScript","#3178c6",0.4,{"name":98,"color":99,"percentage":100},"HTML","#e34c26",0.2,{"name":102,"color":103,"percentage":104},"Makefile","#427819",0.1,{"name":106,"color":107,"percentage":108},"Shell","#89e051",0,4211,1015,"2026-04-15T18:31:20","MIT","macOS, Linux, Windows","未说明",{"notes":116,"python":117,"dependencies":118},"该工具是基于 OpenAI Agents SDK 的框架，主要依赖 API 调用而非本地模型推理，因此通常不需要高性能 GPU。支持通过 LiteLLM 连接多种模型后端（如 Anthropic, Google, Azure OpenAI 等）。首次运行需配置 OPENAI_API_KEY 环境变量。","3.12+",[119,120,121],"openai-agents","pydantic","litellm",[14,13],"2026-03-27T02:49:30.150509","2026-04-16T08:20:13.759535",[126,131,136,141,146,151],{"id":127,"question_zh":128,"answer_zh":129,"source_url":130},35607,"如何将 agency-swarm 连接到自己的类 GPT LLM API（如本地运行的 Llama 3）？","目前 API 集成在理论上是可热替换的，但需要底层代理服务器支持才能实现无缝集成。一旦代理服务器支持完成，官方文档（https:\u002F\u002Fvrsen.github.io\u002Fagency-swarm\u002Fadvanced-usage\u002Fopen-source-models\u002F）将会更新具体的配置方法。建议关注该文档以获取最新进展。","https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fissues\u002F112",{"id":132,"question_zh":133,"answer_zh":134,"source_url":135},35608,"使用 Azure OpenAI 时遇到 'Unknown parameter: tools[0].function.strict' 错误怎么办？","该问题已在主分支（main branch）中修复。请直接从 GitHub 主分支安装最新版本进行测试：\n1. 卸载当前版本。\n2. 运行命令从主分支安装：pip install git+https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm.git@main\n3. 如果问题依旧，尝试更新 Azure OpenAI API 版本至支持结构化输出的版本（如 2024-08-01-preview）。","https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fissues\u002F177",{"id":137,"question_zh":138,"answer_zh":139,"source_url":140},35609,"Genesis 工具生成的 Agent 导入失败或需要频繁人工干预如何解决？","这是一个已知问题，通常由消息历史记录中的小 bug 导致超时或路径解析错误引起。维护者建议从主分支（main branch）安装最新版本以修复此问题。如果仍然遇到问题，可能需要手动修正生成的 Python 文件中的导入语句和文件夹命名格式（确保与 markdown 指令中的名称逻辑一致，去除空格并使用下划线）。","https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fissues\u002F115",{"id":142,"question_zh":143,"answer_zh":144,"source_url":145},35610,"Browsing Agent 无法工作并报错 'invalid syntax' 或找不到目录怎么办？","请检查以下几点：\n1. 确保导入路径正确：使用 from agency_swarm.agents.BrowsingAgent import BrowsingAgent。\n2. 不要在 selenium_config 中填写已经打开的 Chrome 配置文件路径（chrome_profile_path），否则会导致冲突报错。如果不需要特定配置，可注释掉该行。\n3. 某些旧版本存在解析工具文件的 bug，建议尝试使用包含修复补丁的 Fork 版本或等待官方更新。","https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fissues\u002F128",{"id":147,"question_zh":148,"answer_zh":149,"source_url":150},35611,"agency-swarm 是否支持第三方 LLM 提供商（如 Anthropic Claude 或 Google Gemini）？","社区非常期待支持第三方模型（如 Claude 3 Opus 的 200K 上下文或 Gemini 1.5 Pro 的 1M 上下文），以提升性能。目前官方尚未完全内置这些提供商的原生支持，主要依赖 OpenAI 兼容接口。用户可以通过构建符合 OpenAI 标准的代理层来间接使用这些模型，或者直接关注官方关于多模型支持的后续更新。","https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fissues\u002F92",{"id":152,"question_zh":153,"answer_zh":154,"source_url":135},35612,"在使用 Azure OpenAI 配合 Tools（工具）时遇到兼容性问题的根本原因是什么？","根本原因是 Azure OpenAI 的某些 API 版本不支持 OpenAI 最新的 'strict' 参数（用于结构化输出）。当代码中定义了工具（Tools）时，该参数会被自动添加，从而导致请求失败。解决方法是升级到支持结构化输出的新 API 版本（例如 2024-08-01-preview），或者暂时不使用工具功能直到环境升级。",[156,161,166,171,176,181,186,191,196,201,206,211,216,221,226,231,236,241,246,251],{"id":157,"version":158,"summary_zh":159,"released_at":160},280836,"v1.1.0","## Features\r\n- feat(fastapi): auto-generate chat titles from request context and harden guardrail\u002Foutput_type interplay by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F362  \r\n- cli: add agent template creation workflow for new projects by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F380  \r\n- cli(v0 migration): port `settings.json` agents into v1 projects by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F374  \r\n- terminal demo: add \u002Fresume persistence command by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F340\r\n- terminal demo: add show_reasoning parameter to control reasoning traces display by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F342\r\n\r\n_Note: some terminal demo improvements were introduced in 1.0.2._\r\n\r\n## Improvements & Fixes\r\n- guardrails: preserve prior assistant output across retry cycles without `run_data` by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F365  \r\n- guardrails: ensure input wrappers apply regardless of guardrail name by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F370  \r\n- handoffs: inject a reminder system message whenever agents transfer control by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F366  \r\n- metadata: surface the running agency-swarm version on the FastAPI metadata endpoint by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F373  \r\n- files: reuse existing vector-store directories and skip utility files during ingestion by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F385 and https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F382  \r\n- cli: tighten error handling, exit codes, and AGENTS.md compliance for migrate-agent flows by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F386  \r\n- terminal: fix reference routing for agents with repeating name patterns by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F363  \r\n- fix: add tsx runner support and resolve build failures by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F392\r\n\r\n## Docs\r\n- docs: remove legacy v0 mentions and polish navigation by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F371  \r\n- docs: refresh MCP demo guidance, agent initialization, and `max_tokens` usage by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F375  \r\n- docs: align BaseTool examples with the async `run` context pattern by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F384  \r\n- docs: add coverage badge and trim outdated Colab reference by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcommit\u002F57928303925204edbf324ba72fdc086cff4ce64c  \r\n\r\n## Tests\r\n- tests: broaden coverage across tool factory, BaseTool, MCP server, persistence, and FastAPI logging by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F378  \r\n- tests: refactor FastAPI logging middleware integration to run in-process and disable tracing noise by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F379  \r\n- tests: stabilize file attachment expectations in integration suite by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F377  \r\n\r\n## Security\r\n- security: upgrade Next.js dependency to 15.4.7+ to address upstream vulnerability by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcommit\u002Faac33f036c61da67107fdafb76c9de15dbe198c3  \r\n\r\nFull Changelog: https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcompare\u002Fv1.0.2...v1.1.0\r\n","2025-10-08T01:46:10",{"id":162,"version":163,"summary_zh":164,"released_at":165},280837,"v1.0.2","## Improvements\r\n- feat(mcp): add persistent MCP manager to maintain connections with MCP servers by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F355\r\n- feat(terminal): add \u002Fresume persistence command; split demo into focused modules by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F340\r\n- feat: add show_reasoning parameter to control reasoning traces display by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F342\r\n- agency: reject global model parameter by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F353\r\n- guardrails: avoid double-wrapping in FastAPI by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F343\r\n- guardrails: adjust how guardrail messages are added to history by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F346\r\n- prompts: ensure shared instructions lead agent prompts by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F351\r\n- tools: update BaseTool context usage; add reasoning event warning by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F352\r\n- messages: add message_origin parameter to system messages by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F347\r\n## Refactors\r\n- refactor(agency): agent execution system for better modularity and maintainability; refactor tests by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F349\r\n\r\n## Fixes\r\n- fix(mcp): persistent MCP manager driver reuse by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F356\r\n- fix(handoff): incorrect agent assignment for handoff tool call output by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F341\r\n- fix(ui): launcher chat persistence; strict compact client by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F354\r\n\r\n## Tests\r\n- tests: verify message_origin assignment by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F348\r\n\r\n## Handoff\r\n- improve handoff handling and name parsing logic by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F360\r\n\r\nFull Changelog: https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcompare\u002Fv1.0.1...v1.0.2","2025-09-23T04:04:21",{"id":167,"version":168,"summary_zh":169,"released_at":170},280838,"v1.0.1","## What's Changed\r\n\r\n### Agent API & Defaults\r\n* [backwards-compatible] agents: rename return_input_guardrail_errors -> throw_input_guardrail_error by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F330\r\n* Added default model assignment to agent init by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F338\r\n* Updated handoff tool to include unedited agent name and recipient_agent field by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F337\r\n\r\n### Terminal & UI\r\n* Improve Terminal Demo: Support commands: compact summarize, reset by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F331\r\n* terminal_demo: enable reasoning summaries and render in console by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F335\r\n* Fixed reasoning duplication in terminal demo by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F336\r\n* Updated handoff name display in terminal demo by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F332\r\n* ui(console): improve ConsoleEventAdapter rendering by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcommit\u002Fa775841\r\n* ui(launcher): reuse entry agent sync client in \u002Fcompact; add unit test to ensure model passthrough and non-OpenAI path by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcommit\u002F434c9dc\r\n* ui\u002Fterminal: compact uses OpenAI minimal reasoning when model contains 'gpt'; keep provider string intact; retain ID sanitization; integration test unchanged by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcommit\u002F89b93b3\r\n* fix: enable terminal_demo slash menu with prompt-toolkit dependency by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcommit\u002F9981feb\r\n\r\n### Streaming & Messaging\r\n* refactor: make streaming event handling non-destructive by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcommit\u002F13840d2\r\n* streaming: add ensure_event_agent_metadata and use in SendMessage to avoid overwriting agent attribution during sub-agent forwarding by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcommit\u002F6c8e023\r\n* feat: add ExtraParams support to SendMessage for clean field customization by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcommit\u002F8356ee6\r\n\r\n### FastAPI Integration\r\n* fix: FastAPI integration additional_instructions parameter support by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F328\r\n\r\n### Tests & Stability\r\n* tests: stabilize integration test flakes by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F333\r\n* tests: migrate fin_agency agents from v0.x to v1.x pattern by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcommit\u002Fd471366\r\n* test: fix ConsoleEventAdapter handoff_agent initialization in test fixture by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcommit\u002F06d97a9\r\n\r\n### Docs & Examples\r\n* Updated examples and docs by @ArtemShatokhin and @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F326 and https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F327\r\n* Updated multi_agent_workflow example by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F329\r\n* docs(AGENTS): prioritize running relevant tests first; reserve make ci for final verification by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcommit\u002Fa7ab748\r\n\r\n### Other\r\n* Bug fixes and updates by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F334\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcompare\u002Fv1.0.0...v1.0.1","2025-09-13T03:35:04",{"id":172,"version":173,"summary_zh":174,"released_at":175},280839,"v1.0.0","![Framework](https:\u002F\u002Ffirebasestorage.googleapis.com\u002Fv0\u002Fb\u002Fvrsen-ai\u002Fo\u002Fpublic%2Fgithub%2FLOGO_BG_large_bold_shadow%20(1).jpg?alt=media&token=8c681331-2a7a-4a69-b21b-3ab1f9bf1a23)\r\n\r\n**Agency Swarm v1.0: The complete rewrite on the OpenAI Agents SDK and Responses API is now production-ready. This release preserves Agency Swarm’s orchestrator-workers pattern, runs natively async, moves off the legacy Assistants API, and retains production-grade persistence.**\r\n\r\n## What's New: Stable Agents SDK + Responses API\r\n\r\n* **Production-Ready Orchestrator Pattern**  \r\n  Agency Swarm brings its proven orchestrator-workers pattern to the OpenAI Agents SDK. Agents communicate through defined pathways in `communication_flows`, enabling coordinated multi-agent execution with full thread and run state control.\r\n\r\n* **Built on OpenAI Agents SDK** \r\n  Migrated from OpenAI's Assistants API to the **OpenAI Agents SDK** for explicit thread and run state control.\r\n\r\n* **OpenAI Responses API Integration**  \r\n  Uses OpenAI's Responses API by default, enabling support for the latest OpenAI models (including the GPT-5 family). Provides lower latency, more reliable execution, and third-party model compatibility.\r\n\r\n* **Continuity & Migration**  \r\n  v1.0 retains core v0.x capabilities where possible and provides warnings and a [migration guide](https:\u002F\u002Fagency-swarm.ai\u002Fmigration\u002Fguide) to ease upgrades.\r\n\r\n## Major Features\r\n\r\n* **Async-first Architecture**  \r\n  Main methods expose async entry points (e.g., `await agency.get_response()`). Synchronous wrappers remain available, but async is recommended for concurrency.\r\n\r\n* **Improved Communication & Multi-Agency Support**\r\n  * `communication_flows` parameter replaces nested lists in `agency_chart`\r\n  * Agents can belong to multiple agencies with improved per-flow handling\r\n  * Customizable `send_message` tool for inter-agent messaging\r\n  * Parent run ID tracking for end-to-end traceability\r\n  * Per-recipient `send_message` blocking to prevent parallel sends to the same recipient while allowing different recipients to run in parallel\r\n\r\n* **Modern Tool System**\r\n  * `@function_tool` decorator for concise tool creation (BaseTool` still supported)\r\n  * ToolFactory for dynamic tool creation from OpenAPI schemas\r\n  * Context-aware tools with `RunContextWrapper` access\r\n\r\n* **Advanced State Management**\r\n  * Full conversation persistence with complete history management\r\n  * `ThreadManager` and `MessageStore` with `RunHooks`\r\n  * Shared `MasterContext` across agents\r\n\r\n* **Improved Validation System**\r\n  * Improves upon the Agents SDK’s Input Guardrails and Output Guardrails; supports Pydantic validators via BaseTool\r\n  * Native Pydantic models via `output_type` for structured outputs\r\n  * Automatic re-tries on output guardrail failures with system guidance (configurable via `validation_attempts`)\r\n  * Surface input guardrail guidance instead of raising (set return_input_guardrail_errors)\r\n\r\n* **FastAPI Integration**\r\n  * `run_fastapi` method to expose agencies as authenticated HTTP APIs\r\n  * File upload support with image processing\r\n  * Run logs endpoint and metadata APIs\r\n  * Real-time streaming support\r\n\r\n* **Agency Visualization**\r\n  * ReactFlow-based interactive visualization\r\n  * `get_agency_structure`, `plot_agency_chart`, and `visualize` methods\r\n  * HTML templates with agency statistics\r\n\r\n* **Model Context Protocol (MCP) Support**\r\n  * `run_mcp` method with FastMCP server integration\r\n  * Terminal demo support with lifecycle management\r\n  * Concurrency control with `one_call_at_a_time`\r\n\r\n* **Citation System**\r\n  * Methods to extract vector store citations and direct file annotations\r\n  * Complete file handling with automatic path resolution\r\n\r\n## Breaking Changes\r\n\r\n**v0.x code requires migration.**\r\n\r\n* **Agent.get_response** no longer supports delegation to other agents when used without an Agency context\r\n* `response_validator` → `output_guardrails` and `input_guardrails`\r\n* `response_format` parameter → `output_type` on Agent\r\n* Thread callbacks now handle complete conversation data, not just IDs\r\n* `agency_chart` syntax → positional arguments + `communication_flows`\r\n* `threads_callbacks` dict → separate `load_threads_callback` and `save_threads_callback`\r\n\r\nSee the [Migration Guide](https:\u002F\u002Fagency-swarm.ai\u002Fmigration\u002Fguide) for detailed code examples.\r\n\r\n## New Capabilities\r\n\r\n* **Web Search & Computer Use**: Native OpenAI Responses API integration\r\n* **Latest Models**: Support for the GPT-5 family and current OpenAI models\r\n* **Third-Party Models**: Use any LiteLLM-compatible provider (Anthropic, Google, etc.)\r\n* **Direct Thread Control**: Complete control over conversation threads and runs\r\n* **Enhanced Streaming**: Improved real-time streaming with better event ordering\r\n\r\n## Usage Examples\r\n\r\nThe [\u002Fexamples](https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Ftree\u002Fmain\u002Fexamples) directory contains 9 comprehensive examples demonstrating v1.0 capabilities:\r\n\r\n* Multi-agent workflows with th","2025-09-03T23:43:53",{"id":177,"version":178,"summary_zh":179,"released_at":180},280840,"v1.0.0-beta.6","This update builds on v1.0.0-beta.1-5 and introduces significant improvements to agent orchestration, streaming stability, and framework architecture with 350+ commits.\r\n\r\n## Breaking Changes\r\n- **Agent.get_response** no longer supports delegation to other agents when used without an Agency context\r\n\r\n## Major Changes\r\n\r\n### 🧩 Orchestration & Flows\r\n- **Multi-Agency Support** — Agents can belong to multiple agencies with improved per-flow handling https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F104, https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F155 — @ArtemShatokhin\r\n- **Parent Run ID Tracking** — End-to-end traceability across nested calls and streaming sequences https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F153, https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F154 — @bonk1t\r\n- **Thread Management** — Shared user conversations across agents for smoother context continuity https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F156 — @bonk1t\r\n- **Unified Send Message Tool** — Single `send_message` interface https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F93 — @bonk1t\r\n- **Flat Message Structure** — Centralized flat message structure with filtering for better persistence https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F72 — @bonk1t\r\n\r\n### 📡 Streaming & MCP\r\n- **Nested Agent Event Streaming** — Fixed critical issue where nested agent events weren't visible in streaming https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F88 — @bonk1t\r\n- **Streaming Order Consistency** — Fixed event ordering; third-party model streaming reliability (LiteLLM\u002FClaude) https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F105 https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F152\r\n- **MCP Server Integration** — Ported run_mcp with FastMCP for model context protocol support https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F54 — @ArtemShatokhin\r\n- **MCP Lifecycle & Stability** — Simplified `get_response_stream`, explicit MCP lifecycle, and cleanup fixes https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F117, https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F148 — @ArtemShatokhin\r\n- **Terminal Demo** — Fixed special-character rendering https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fcommit\u002F80cb8da — @ArtemShatokhin\r\n\r\n### ⚙️ Tools & Execution\r\n- **ToolFactory for v1.x** — Ported ToolFactory for dynamic tool creation  https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F61 — @ArtemShatokhin\r\n- **Citation Extraction System** — Comprehensive citation extraction for FileSearch and file attachments https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F58 — @bonk1t\r\n- **Concurrency Control** — `one_call_at_a_time` for safer sequential tool execution  https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F150 — @ArtemShatokhin\r\n- **Tool Context Initialization** — Lifecycle-aware context init for BaseTools  https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F126 — @ArtemShatokhin\r\n- **Execution Script Sync** — Synced `execute.py` with dev to avoid feature loss  https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F122 — @ArtemShatokhin\r\n\r\n### 🛠️ Developer Experience\r\n- **Agent Module Refactoring** — Split monolithic agent.py (1335 lines) into organized modules https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F66 — @bonk1t\r\n- **Framework Refactoring** — Broad cleanup for maintainability across execution\u002Fagency\u002Ffile management https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F159 — @ArtemShatokhin\r\n- **FastAPI File Uploads** — Added file upload capability with image processing https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F64 — @ArtemShatokhin\r\n- **FastAPI** — Added run-logs endpoint and flattened `\u002Fget_metadata` response https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F116 https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fcommit\u002F137f39f\r\n- **Instruction File Paths** — Load `.md` instructions with automatic path resolution https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F94 — @ArtemShatokhin\r\n- **File Handling** — Removed auto-adding FileSearch for attachments; fixed relative filepaths https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F149 https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F101 — @ArtemShatokhin\r\n- **dotenv Autoload** — Environment variables now load automatically https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F129 — @bonk1t\r\n- **Agency Metadata (Rich)** — Restored detailed metadata for structure visualization https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F134 — @bonk1t\r\n- **Makefile Ergonomics** — `make prime` target for developer workflows https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fcommit\u002F5492a7c — @bonk1t\r\n\r\n### 🔒 Quality & Stability\r\n- **Type Annotations & MyPy** — Resolved mypy issues and broadened typing coverage https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F160 — @bonk1t\r\n- **Defaults** — Removed the max-turns default limit to avoid unintended caps https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F151 — @ArtemShatokhin\r\n\r\n### 📖 Examples & Docs\r\n- Guardrails demo; tightened `CLAUDE.md`; improved staging guidance  https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fcommit\u002F817a6ee  https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fcommit\u002F9012d1c https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fcommit\u002Fa173a86\r\n\r\n➡️ See [v1.0.0-beta.5 release notes](https:\u002F\u002Fgi","2025-08-26T01:10:52",{"id":182,"version":183,"summary_zh":184,"released_at":185},280841,"v0.7.2","## What's Changed\r\n* Handle Azure rate limit in thread retry by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F309\r\n* Increase retry attempts from 3 to 10 for rate limiting resilience by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F310\r\n* Fix run_async_sync handling in running loops by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F313\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcompare\u002Fv0.7.1...v0.7.2","2025-07-29T23:34:40",{"id":187,"version":188,"summary_zh":189,"released_at":190},280842,"v1.0.0-beta.5","This update builds on v1.0.0-beta.1-4 and introduces new capabilities for the Agents SDK + Responses API foundation.\r\n\r\n## New Features\r\n\r\n- **FastMCP server integration**: Launch a FastMCP server exposing BaseTool and FunctionTool instances via the `run_mcp` method with directory path inputs and terminal demo examples\r\n- **Citation extraction system for FileSearch and direct file attachments**: Added `citation_extractor.py` module with `extract_vector_store_citations()` and `extract_direct_file_annotations()` functions, plus `include_search_results` parameter support\r\n- **Custom SendMessage tool classes**: Added Agency-level and Agent-level support for enhanced inter-agent communication with context passing\r\n\r\n## PRs\r\n\r\n* Ported run_mcp for the new framework by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F54\r\n* feat(citations): Add citation extraction system for FileSearch and direct file attachments by @bonk1t in https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F58\r\n* Added terminal demo method by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002Fbonk1t\u002Fagency-swarm\u002Fpull\u002F42\r\n* SendMessage Enhancement, Documentation Fixes, and Code Cleanup by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F307\r\n* feat(observability): Update observability demo and docs by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F306\r\n* Add examples\u002Fobservability_demo.py; Minor fixes and improvements by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F305\r\n\r\n➡️ See [v1.0.0-beta.1 release notes](https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Freleases\u002Ftag\u002Fv1.0.0-beta.1) for core Agents SDK details.\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcompare\u002Fv1.0.0-beta.4...v1.0.0-beta.5","2025-07-17T04:40:08",{"id":192,"version":193,"summary_zh":194,"released_at":195},280843,"v1.0.0-beta.4","This update builds on v1.0.0-beta.1-3 and introduces targeted improvements for the new Agents SDK + Responses API foundation.\r\n\r\n- Renames `create_interactive_visualization` to `visualize` throughout the codebase for clarity and consistency.\r\n- Fixes agent schema loading and API parameter handling; moves `schemas_folder`, `api_headers`, `api_params` to the correct config section and removes false deprecation warnings.\r\n- Switches to flexible dependency version ranges and resolves tool validation\u002Ftype issues for broader compatibility.\r\n- All 193 tests passing; all examples verified.\r\n\r\n## PRs:\r\n* refactor(visualization): rename create_interactive_visualization to visualize by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F303\r\n* fix(agent): correct schemas_folder, api_headers, api_params param handling, update dependencies by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F304\r\n\r\n➡️ See [v1.0.0-beta.1 release notes](https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Freleases\u002Ftag\u002Fv1.0.0-beta.1) for core Agents SDK details.\r\n\r\n**Full Changelog:** https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcompare\u002Fv1.0.0-beta.3...v1.0.0-beta.4","2025-07-05T01:30:17",{"id":197,"version":198,"summary_zh":199,"released_at":200},280844,"v1.0.0-beta.3","This update builds on v1.0.0-beta.1 and v1.0.0-beta.2 and introduces key enhancements to the new Agents SDK + Responses API foundation.\r\n\r\n## What's Changed\r\n- Implemented ReactFlow-based Agency Visualization:\r\n![Screenshot 2025-07-01 at 00 26 56](https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002Fc32a5927-e0b2-4d48-9628-8443c9825264)\r\n\r\n- Added AG-UI Support + CopilotKit demo:\r\n![Screenshot 2025-07-01 at 00 28 03](https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F0377230d-5784-447e-bf6b-dc474765bd84)\r\n\r\n## PRs:\r\n* AG-UI Support, CopilotKit demo, ReactFlow-based Agency Visualization in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F299\r\n* FastAPI streaming fix: serialization for dataclasses in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F296\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcompare\u002Fv1.0.0-beta.2...v1.0.0-beta.3\r\n\r\n➡️ See [v1.0.0-beta.1 release notes](https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Freleases\u002Ftag\u002Fv1.0.0-beta.1) for the core Agents SDK architecture and migration overview.","2025-07-01T01:32:30",{"id":202,"version":203,"summary_zh":204,"released_at":205},280825,"v1.8.1","## 变更\n- 将 `agency.tui()` 的首次安装版本锁定为 `agentswarm-cli 1.4.7`\n- 合并已发布的 CLI 中的终端 UI 认证\u002F引导流程修复\n\n## 验证\n- `uv run --python 3.13 pytest tests\u002Ftest_agency_modules\u002Ftest_agentswarm_cli_tui.py`\n- 已确认在本次发布之前，`agentswarm-cli 1.4.7` 已经发布到 npm 上。","2026-04-14T08:51:26",{"id":207,"version":208,"summary_zh":209,"released_at":210},280826,"v1.8.0","本次发布改进了首次消息的用户体验、跨提供商的历史兼容性以及元数据可见性，同时加强了 LiteLLM 发布验证机制，并升级了 OpenAI Agents SDK。简而言之：更优的默认配置、更清晰的迁移路径以及更安全的发布行为。\n\n## 破坏性变更\n\n本次发布继续推进 v1.x 版本中关于护栏\u002F运行时 API 的清理与标准化工作。\n\n- **输入护栏运行时标志迁移**，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F532 中完成  \n  - 标准化的运行时标志为 `raise_input_guardrail_error`。  \n  - `throw_input_guardrail_error` 仍作为已弃用的别名保留。  \n  - `return_input_guardrail_errors` 已被移除，现会快速失败并提供迁移指导。\n- **交接导出澄清**，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F532 中完成  \n  - `Handoff` 现在映射到 Agency Swarm 的交接工具导出。  \n  - OpenAI Agents SDK 的交接则以 `SDKHandoff` 的形式导出。\n\n## 功能特性\n\n这些新增功能提升了代理行为和 API 元数据方面的开箱即用体验。\n\n- **内置 `PresentFiles` 工具**，由 @ArtemShatokhin 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F520 中完成  \n  - 将文件移动到配置的预览目录中，同时确保覆盖与符号链接的安全性，并输出预览元数据。\n- **对话开场白 + 快速回复缓存优化**，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F519 和 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F526 中完成  \n  - 为缓存的首次消息重放添加了 `quick_replies` 支持。  \n  - 加强了缓存指纹识别（包括共享指令和流式上下文）以及针对覆盖\u002F钩子路径的缓存跳过规则。\n- **在元数据中暴露工具输入 Schema**，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F523 中完成  \n  - 函数工具的 Schema 现在可在元数据中获取，从而支持更丰富的 UI 和客户端生成。\n- **默认启用 Web 搜索来源**，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F530 中完成  \n  - 默认启用来源包含功能，并新增共享来源提取工具。\n- **FastAPI 服务器示例优化**，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F529 中完成  \n  - 更新了服务器示例，以展示更清晰的代理间路由和类型化的计算工具使用。\n\n## 改进与修复\n\n本次发布重点关注跨模型后端的历史正确性以及发布检查的稳定性。\n\n- **历史协议兼容性加固**，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F525 中完成  \n  - 一致地持久化协议元数据，拒绝不兼容的混合历史，并改进 LiteLLM ↔ Responses 的重放行为。\n- **RunItem 持久化重构**，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F527 中完成  \n  - 在流式持久化路径中切换为使用类型化的 `RunItem.raw_item` 属性访问。\n- **OpenAI Agents SDK 升级**，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F535 中完成  \n  - 将 `openai-agents` 升级至 `0.9.3`，并更新锁定元数据。\n- **发布检查 u","2026-02-25T01:22:52",{"id":212,"version":213,"summary_zh":214,"released_at":215},280827,"v1.7.0","本次发布引入了带有缓存和元数据支持的对话开场白、更强大的验证机制，以及更整洁的 FastAPI\u002FMCP 行为，并对测试和文档进行了稳定性优化。\n\n## 重大变更\n本版本通过移除已弃用的 v0.x API，完成了 v1.x 的过渡。如果您仍在使用旧模式，请参阅[迁移指南](https:\u002F\u002Fagency-swarm.ai\u002Fmigration\u002Fguide)。\n\n* 移除了已弃用的 Agency API（`agency_chart`、补全代理、`message_files`），强制执行严格的 Agent 关键字参数，并移除了遗留的提醒覆盖行为。\n* `SendMessage` 不再接受 `my_primary_instructions` 字段。请将其从工具负载中移除，以避免模式错误。\n* `get_agency_structure` 已弃用。请改用 `get_agency_graph` 获取节点和边信息，而需要完整元数据包时则使用 `get_metadata`。\n* `SendMessageHandoff` 已弃用，并更名为 `Handoff`。\n\n## 功能特性\n* 特性：对话开场白的缓存与元数据支持，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F516 中实现：\n  * 将对话开场白暴露在 agency graph 和 FastAPI 元数据中，新增基于指纹识别和预热的缓存式开场白提示系统，适用于同步和流式运行；同时移除了已弃用的补全路径，并更新了相关文档和测试。\n* 向 agency 类添加共享资源参数，由 @ArtemShatokhin 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F502 中实现：\n  * 现在可以一次性传递共享工具、工具文件夹、文件文件夹以及 MCP 服务器，并将它们应用于每个代理（支持共享向量存储并完成 FastAPI 配置）。\n* 为文件 URL 添加本地文件路径支持，由 @ArtemShatokhin 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F498 中实现：\n  * 当设置了 `allowed_local_file_dirs` 时，FastAPI 现在允许在 `file_urls` 中使用本地文件路径，并实施严格的白名单机制、改进错误报告，同时更新了 AG-UI 的处理逻辑。\n* 为 FastAPI 端点添加动态客户端配置覆盖功能，由 @ArtemShatokhin 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F503 中实现：\n  * 现在每次请求都可以通过 `client_config` 覆盖 `base_url` 和 `api_key`（以及 LiteLLM 提供商密钥）。\n\n## 改进\n* 移除 `my_primary_instructions`，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F499 中实现：简化了 `SendMessage` 模式，移除了冗余的 `my_primary_instructions` 字段。\n* 杂项：将 openai-agents 升级至 0.6.4，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F497 中完成：升级了 `openai-agents`（以及 `openai`），并加强了运行时工具类型检查，以便更早地捕获无效的托管工具输入。\n* 将 `SendMessageHandoff` 重命名为 `Handoff`，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F507 中实现：正式推出 `Handoff` 作为推荐的工具名称，保留 `SendMessageHandoff` 作为已弃用的别名，并相应更新了文档和测试。\n* 修复：强化工具验证错误提示，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F501 中完成：Agent 工具验证现在会更早地拒绝无效的工具条目（例如未初始化的托管工具类或 `FunctionTool` 类）。\n* 重命名 a","2026-01-20T09:16:53",{"id":217,"version":218,"summary_zh":219,"released_at":220},280828,"v1.6.0","本次发布重点聚焦于**用量与费用跟踪**和**流式响应取消**（FastAPI + 终端演示），同时改进了**热重载**功能，提升了**事件\u002F历史记录处理**的可靠性（包括时间戳、孤儿清理、LiteLLM ID 规范化），并进一步稳定了文档和测试。\n\n如果你是 Agency Swarm 的新用户，以下内容将帮助你理解这些更新的意义：\n\n- **流式响应取消**：当你以逐 token 的方式获取流式响应时，可以提前停止该流，这在你已获得所需答案或希望避免额外消耗 token 时非常有用。\n- **热重载（终端演示）**：类似于 FastAPI，终端演示现在会在本地 `.py` 或 `.md` 文件发生更改时自动重启，从而让你更快地进行迭代开发。\n- **用量与费用跟踪**：FastAPI 响应和终端演示现会包含 token 总数以及**估算**的 USD 费用（支持多智能体运行）。我们使用真实的模型定价，包括 LiteLLM 支持的模型。\n- **更可靠的历史记录**：消息会保留其发出时的时间戳，移除孤立的工具调用及输出，并对 LiteLLM 占位符 ID 进行规范化处理。\n\n---\n\n## 功能特性\n* **新增 Token 用量与费用跟踪**（FastAPI + 终端演示；支持多智能体场景）——由 @ArtemShatokhin 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F489 中实现  \n  - **你将获得**：一个包含请求数量、token 分布及估算费用的 `usage` 摘要。  \n  - **使用场景**：FastAPI 端点会返回该信息，终端演示中也会进行跟踪（使用 `\u002Fcost` 命令查看，程序退出时也会打印）。  \n  - **多智能体感知**：不仅统计主智能体的 token 和费用，还会计入运行过程中所有子智能体调用所产生的 token 和费用。\n* **为 FastAPI 和终端演示热重载添加流式响应取消支持**——由 @ArtemShatokhin 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F469 中实现  \n  - **FastAPI 流式响应**：流式响应会提前发送一个 `run_id`；客户端可通过取消端点来中断正在进行的流式请求。若客户端断开连接，则流式请求会自动取消。  \n  - **取消模式**：`immediate`（立即停止）和 `after_turn`（完成当前一轮后停止）。  \n  - **终端演示热重载**：监听 `.py` 和 `.md` 文件的变化，并在本地开发时自动重启。\n* **为终端演示添加取消功能**——由 @ArtemShatokhin 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F474 中实现  \n  - **使用方法**：在流式响应过程中按下 `ESC` 键即可取消当前响应。  \n  - **取消后更整洁的历史记录**：演示会过滤掉重复或孤立的工具调用消息，确保保存的聊天记录始终有效且可回放。\n\n---\n\n## 改进内容\n* 改进终端工具格式——由 @ArtemShatokhin 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F471 中实现  \n* 在事件发出时捕获并分配时间戳——由 @ArtemShatokhin 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F485 中实现  \n  - **变更内容**：每个事件\u002F消息现在都会在其流式传输过程中发出时就附带时间戳，而非在后续持久化时再添加。  \n  - **重要性**：这使得在流式传输、保存的线程以及不同 UI 之间对历史记录进行排序和回放时更加一致。  \n  - **工具时","2025-12-24T16:34:03",{"id":222,"version":223,"summary_zh":224,"released_at":225},280829,"v1.5.0","## 变更内容\n\n### 功能\n* 添加 `ToolFactory.from_mcp` 方法，用于将 MCP 服务器转换为经过验证的工具，由 @ArtemShatokhin 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F453 中实现。\n* 启用工具的 OpenAPI 模式生成功能，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F446 中实现。\n* 在 `run_fastapi` 中添加 `server_url` 参数，以改善模式生成效果，由 @ArtemShatokhin 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F465 中实现。\n\n### 修复与改进\n* 修复：修正护栏流式处理的持久化问题，并重构执行模块，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F434 中完成。\n* 修复：规范化基类工具的错误输出，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F455 中完成。\n* 修复：防止流式上下文覆盖导致的变异问题，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F457 中完成。\n* 改进：将超时参数移至工具输入字段中，由 @ArtemShatokhin 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F456 中实现。\n* 改进：简化了 Litellm 文档中的模型定义，由 @ArtemShatokhin 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F463 中完成。\n\n### 依赖项\n* 将 `openai-agents` 升级至 0.6.0 版本，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F458 中完成。\n\n### 文档\n* 添加 n8n 集成文档，由 @dumitruPuggle 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F464 中完成。\n* 更新 n8n 集成文档，加入关于自…的相关说明，由 @dumitruPuggle 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F466 中完成。\n\n**完整变更日志**：https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcompare\u002Fv1.4.1...v1.5.0","2025-12-01T00:50:36",{"id":227,"version":228,"summary_zh":229,"released_at":230},280830,"v1.4.1","## 变更内容\n* 修复：由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F454 中打包可视化模板\n* 文档：由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F443 中在快速入门中添加启动模板链接\n\n**完整变更日志**：https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcompare\u002Fv1.4.0...v1.4.1","2025-11-15T02:52:08",{"id":232,"version":233,"summary_zh":234,"released_at":235},280831,"v1.4.0","## 🚀 功能特性\n\n### 内置工具与 CLI 导入\n引入了内置工具——**IPythonInterpreter**、**PersistentShellTool** 和 **LoadFileAttachment**——以及一个新的 `import-tool` CLI 命令。  \n同时对文件处理、代理模板生成，以及内置工具的文档和导航进行了优化。\n\n- 添加内置工具和 CLI 导入命令（[#442](https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F442)），由 @ArtemShatokhin 实现  \n  - CLI：新增 `import-tool` 命令，用于列出并将内置工具复制到项目中  \n  - 文档：新增 [内置工具](https:\u002F\u002Fagency-swarm.ai\u002Fcore-framework\u002Ftools\u002Fbuilt-in-tools) 页面，并添加交叉链接\n\n### 终端演示改进\n对终端演示（Agency.terminal_demo()）进行了重大 UX 升级，加入了交互式下拉菜单和自动补全功能。\n\n- 终端演示改进（[#432](https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F432)），由 @ArtemShatokhin 实现  \n  - 为 `\u002F` 命令和 `@agent` 提及添加了可通过键盘导航的下拉菜单  \n  - 改进了代理解析、错误处理和任务交接逻辑  \n  - 新增自动补全功能、新的快捷键绑定（Tab、Esc、Enter），并优化了样式  \n  - 更新了针对所有新行为的集成测试\n\n---\n\n## 🛠 修复与改进\n- 改进 IPython 工具的隔离机制（[#448](https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F448)），由 @ArtemShatokhin 实现  \n- 将工作目录作为独立的 IPython 工具输入项添加（[#452](https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F452)），由 @ArtemShatokhin 实现  \n- 其他改进：通过跳过子文件夹、检测空目录以及优化向量存储的创建，进一步提升代理的文件处理能力。\n\n---\n\n## 📘 文档\n- 新增 [**Widget 集成** 文档](https:\u002F\u002Fagency-swarm.ai\u002Fplatform\u002Fintegrations\u002Fwidget)（[#427](https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F427)），由 @MykhailoShchuka 完成  \n\n---\n\n## 🧑‍💻 新贡献者\n- @MykhailoShchuka 在 [#427](https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F427) 中完成了首次贡献  \n\n---\n\n**完整变更日志：** [v1.3.1 → v1.4.0](https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcompare\u002Fv1.3.1...v1.4.0)","2025-11-12T16:49:50",{"id":237,"version":238,"summary_zh":239,"released_at":240},280832,"v1.3.1","## 变更内容\n* 修复：确保 OpenAPI 工具序列化 Schema 请求体，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F430 中完成\n* 修复：为可视化函数添加对象序列化功能，并将 `serialize()` 方法移至工具模块，由 @ArtemShatokhin 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F426 中完成\n* 重构：采用 SDK 响应模型，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F424 中完成\n* 重构：简化引用提取逻辑，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F431 中完成\n\n\n**完整变更日志**：https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcompare\u002Fv1.3.0...v1.3.1","2025-10-31T03:00:07",{"id":242,"version":243,"summary_zh":244,"released_at":245},280833,"v1.3.0","## 功能特性\n* 支持多模态工具输出，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F383 中将 Agents SDK 升级至 0.4.1。\n* 使用代理名称作为默认跟踪名称，由 @ArtemShatokhin 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F404 中实现。\n* 将运行分组到单个跟踪中，由 @ArtemShatokhin 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F411 中完成。\n* `get_response_stream` 现在返回运行响应包装器，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F400 中实现。更多详情请参阅下方的“流式 API 更新”部分。\n* 扩展了从 Agents SDK 导入到框架中的内容，由 @ArtemShatokhin 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F402 中完成。\n\n## 改进与修复\n* 改进了终端演示中的提及解析，并更新了 `show_reasoning` 的默认值，由 @ArtemShatokhin 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F403 中完成。\n* 修复：简化向量存储导入时的等待流程，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F423 中实现。\n* 修复：传播 `run_config` 中的跟踪 ID，由 @ArtemShatokhin 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F413 中完成。\n* 修复：解除流式处理最终 Future 的阻塞，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F414 中实现。\n* 修复：在代理间调用过程中为托管工具系统消息设置 `callerAgent`，由 @ArtemShatokhin 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F420 中完成。\n\n## 文档\n* 文档：更新 API 参考以匹配 Agency 和 Agent API，由 @ArtemShatokhin 完成，链接为 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F416。\n* 修订 README 文件以提高清晰度并进行更新，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F422 中完成。\n* 修复验证文档中的格式问题，由 @ArtemShatokhin 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F412 中完成。\n\n## 流式 API 更新\n- `Agency.get_response_stream` 和 `Agent.get_response_stream` 现在会立即返回 `StreamingRunResponse`，而不是可等待的协程。\n- 现有的 `async for` 循环仍可正常工作，但对这些方法直接调用 `await` 现在会抛出 `TypeError`，因为它们是同步的。\n\n更新后的用法如下：\n```python\nstream = agent.get_response_stream(...)\nasync for event in stream:\n...\nfinal_result = await stream.wait_final_result()\n```\n获取流时请移除开头的 `await`。之前编写过 `await agent.get_response_stream(...)` 的调用方，需采用上述模式。\n\n**完整变更日志**：https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcompare\u002Fv1.2.1...v1.3.0","2025-10-28T02:54:14",{"id":247,"version":248,"summary_zh":249,"released_at":250},280834,"v1.2.1","## 变更内容\n* 修复：在 MCP 代理中支持异步上下文，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F406 中完成\n* 准备 1.2.1 版本发布，由 @bonk1t 在 https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F407 中完成\n\n\n**完整变更日志**：https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcompare\u002Fv1.2.0...v1.2.1","2025-10-15T02:06:19",{"id":252,"version":253,"summary_zh":254,"released_at":255},280835,"v1.2.0","## Features\r\n- FastAPI integration: provide per-agent capability metadata (tools, reasoning, hosted tools) by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F397\r\n- FastAPI integration: support `DRY_RUN` env variable so agent metadata can be served without initialization \u002F API keys by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F398\r\n- handoffs (backwards-compatible): move reminder customization onto `Agent.handoff_reminder`, add a dedicated `examples\u002Fhandoffs.py`, and update communication-flow docs by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F388 and by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F399\r\n\r\n## Improvements & Fixes\r\n- fix: isolate agent runtime state per agency by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F396  \r\n- fix: isolate send_message pending guard per thread manager by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F394  \r\n- fixes: add vector-store file cleanup, timestamp reconciliation, and resilient ID parsing so stale artifacts and OpenAI ID mismatches no longer break sync jobs by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F393 and @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F395  \r\n- fix: harden mcp shutdown handling by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F401\r\n\r\n## Dependencies\r\n- deps: upgrade the OpenAI Agents SDK from 0.2.9 to 0.3.3 (plus pinned transitive updates) to include streaming fixes and other improvements by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F369\r\n\r\n## Docs\r\n- docs: rename \"open-source models\" to \"third-party models\", fix GitHub references by @bonk1t in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F391  \r\n- docs\u002Fexamples: expand hybrid communication flow guidance and align tutorials with the new handoff reminder APIs by @ArtemShatokhin in https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fpull\u002F388  \r\n\r\n\r\nFull Changelog: https:\u002F\u002Fgithub.com\u002FVRSEN\u002Fagency-swarm\u002Fcompare\u002Fv1.1.0...v1.2.0\r\n","2025-10-14T01:48:20"]