[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-langchain-ai--deepagents":3,"tool-langchain-ai--deepagents":62},[4,18,26,36,46,54],{"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",158594,2,"2026-04-16T23:34:05",[14,13,35],"语言模型",{"id":37,"name":38,"github_repo":39,"description_zh":40,"stars":41,"difficulty_score":42,"last_commit_at":43,"category_tags":44,"status":17},8272,"opencode","anomalyco\u002Fopencode","OpenCode 是一款开源的 AI 编程助手（Coding Agent），旨在像一位智能搭档一样融入您的开发流程。它不仅仅是一个代码补全插件，而是一个能够理解项目上下文、自主规划任务并执行复杂编码操作的智能体。无论是生成全新功能、重构现有代码，还是排查难以定位的 Bug，OpenCode 都能通过自然语言交互高效完成，显著减少开发者在重复性劳动和上下文切换上的时间消耗。\n\n这款工具专为软件开发者、工程师及技术研究人员设计，特别适合希望利用大模型能力来提升编码效率、加速原型开发或处理遗留代码维护的专业人群。其核心亮点在于完全开源的架构，这意味着用户可以审查代码逻辑、自定义行为策略，甚至私有化部署以保障数据安全，彻底打破了传统闭源 AI 助手的“黑盒”限制。\n\n在技术体验上，OpenCode 提供了灵活的终端界面（Terminal UI）和正在测试中的桌面应用程序，支持 macOS、Windows 及 Linux 全平台。它兼容多种包管理工具，安装便捷，并能无缝集成到现有的开发环境中。无论您是追求极致控制权的资深极客，还是渴望提升产出的独立开发者，OpenCode 都提供了一个透明、可信",144296,1,"2026-04-16T14:50:03",[13,45],"插件",{"id":47,"name":48,"github_repo":49,"description_zh":50,"stars":51,"difficulty_score":32,"last_commit_at":52,"category_tags":53,"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":55,"name":56,"github_repo":57,"description_zh":58,"stars":59,"difficulty_score":32,"last_commit_at":60,"category_tags":61,"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",[45,13,15,14],{"id":63,"github_repo":64,"name":65,"description_en":66,"description_zh":67,"ai_summary_zh":68,"readme_en":69,"readme_zh":70,"quickstart_zh":71,"use_case_zh":72,"hero_image_url":73,"owner_login":74,"owner_name":75,"owner_avatar_url":76,"owner_bio":77,"owner_company":78,"owner_location":78,"owner_email":79,"owner_twitter":75,"owner_website":80,"owner_url":81,"languages":82,"stars":95,"forks":96,"last_commit_at":97,"license":98,"difficulty_score":32,"env_os":99,"env_gpu":99,"env_ram":99,"env_deps":100,"category_tags":106,"github_topics":107,"view_count":32,"oss_zip_url":78,"oss_zip_packed_at":78,"status":17,"created_at":109,"updated_at":110,"faqs":111,"releases":141},8227,"langchain-ai\u002Fdeepagents","deepagents","Agent harness built with LangChain and LangGraph. Equipped with a planning tool, a filesystem backend, and the ability to spawn subagents - well-equipped to handle complex agentic tasks.","deepagents 是一个基于 LangChain 和 LangGraph 构建的“开箱即用”智能体框架，旨在帮助开发者快速搭建能处理复杂任务的 AI 助手。它解决了传统开发中需要手动编排提示词、工具调用及上下文管理的繁琐痛点，让用户无需从零开始即可拥有一个具备完整能力的智能体。\n\n该工具特别适合希望高效构建自动化工作流、代码辅助或研究助手的 AI 开发者与研究人员。其核心亮点在于内置了丰富的原生能力：包括自动任务规划与进度追踪、完整的文件系统操作（读写、搜索）、安全的沙箱命令执行，以及独特的“子智能体”机制——可将大任务拆解并委托给拥有独立上下文的子代理并行处理。此外，deepagents 还具备智能上下文管理功能，能在对话过长时自动摘要或将大输出保存至文件，有效突破长度限制。\n\n除了作为 Python 库集成，deepagents 还提供功能强大的命令行界面（CLI），支持交互式终端操作、实时网络搜索及无头模式运行，体验类似 Claude Code 或 Cursor。无论是需要快速原型验证，还是构建生产级应用，deepagents 都提供了灵活的自定义接口，允许用户轻松替换模型、","deepagents 是一个基于 LangChain 和 LangGraph 构建的“开箱即用”智能体框架，旨在帮助开发者快速搭建能处理复杂任务的 AI 助手。它解决了传统开发中需要手动编排提示词、工具调用及上下文管理的繁琐痛点，让用户无需从零开始即可拥有一个具备完整能力的智能体。\n\n该工具特别适合希望高效构建自动化工作流、代码辅助或研究助手的 AI 开发者与研究人员。其核心亮点在于内置了丰富的原生能力：包括自动任务规划与进度追踪、完整的文件系统操作（读写、搜索）、安全的沙箱命令执行，以及独特的“子智能体”机制——可将大任务拆解并委托给拥有独立上下文的子代理并行处理。此外，deepagents 还具备智能上下文管理功能，能在对话过长时自动摘要或将大输出保存至文件，有效突破长度限制。\n\n除了作为 Python 库集成，deepagents 还提供功能强大的命令行界面（CLI），支持交互式终端操作、实时网络搜索及无头模式运行，体验类似 Claude Code 或 Cursor。无论是需要快速原型验证，还是构建生产级应用，deepagents 都提供了灵活的自定义接口，允许用户轻松替换模型、添加专属工具或调整系统提示，是构建下一代自主智能体的坚实基石。","\u003Cdiv align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fdocs.langchain.com\u002Foss\u002Fpython\u002Fdeepagents\u002Foverview#deep-agents-overview\">\n    \u003Cpicture>\n      \u003Csource media=\"(prefers-color-scheme: dark)\" srcset=\".github\u002Fimages\u002Flogo-dark.svg\">\n      \u003Csource media=\"(prefers-color-scheme: light)\" srcset=\".github\u002Fimages\u002Flogo-light.svg\">\n      \u003Cimg alt=\"Deep Agents Logo\" src=\".github\u002Fimages\u002Flogo-dark.svg\" width=\"50%\">\n    \u003C\u002Fpicture>\n  \u003C\u002Fa>\n\u003C\u002Fdiv>\n\n\u003Cdiv align=\"center\">\n  \u003Ch3>The batteries-included agent harness.\u003C\u002Fh3>\n\u003C\u002Fdiv>\n\n\u003Cdiv align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fopensource.org\u002Flicenses\u002FMIT\" target=\"_blank\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fl\u002Fdeepagents\" alt=\"PyPI - License\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fpypistats.org\u002Fpackages\u002Fdeepagents\" target=\"_blank\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fpepy\u002Fdt\u002Fdeepagents\" alt=\"PyPI - Downloads\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fpypi.org\u002Fproject\u002Fdeepagents\u002F#history\" target=\"_blank\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fv\u002Fdeepagents?label=%20\" alt=\"Version\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fx.com\u002Flangchain\" target=\"_blank\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Ftwitter\u002Furl\u002Fhttps\u002Ftwitter.com\u002Flangchain.svg?style=social&label=Follow%20%40LangChain\" alt=\"Twitter \u002F X\">\u003C\u002Fa>\n\u003C\u002Fdiv>\n\n\u003Cbr>\n\nDeep Agents is an agent harness. An opinionated, ready-to-run agent out of the box. Instead of wiring up prompts, tools, and context management yourself, you get a working agent immediately and customize what you need.\n\n**What's included:**\n\n- **Planning** — `write_todos` for task breakdown and progress tracking\n- **Filesystem** — `read_file`, `write_file`, `edit_file`, `ls`, `glob`, `grep` for reading and writing context\n- **Shell access** — `execute` for running commands (with sandboxing)\n- **Sub-agents** — `task` for delegating work with isolated context windows\n- **Smart defaults** — Prompts that teach the model how to use these tools effectively\n- **Context management** — Auto-summarization when conversations get long, large outputs saved to files\n\n> [!NOTE]\n> Looking for the JS\u002FTS library? Check out [deepagents.js](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagentsjs).\n\n## Quickstart\n\n```bash\npip install deepagents\n# or\nuv add deepagents\n```\n\n```python\nfrom deepagents import create_deep_agent\n\nagent = create_deep_agent()\nresult = agent.invoke({\"messages\": [{\"role\": \"user\", \"content\": \"Research LangGraph and write a summary\"}]})\n```\n\nThe agent can plan, read\u002Fwrite files, and manage its own context. Add tools, customize prompts, or swap models as needed.\n\n> [!TIP]\n> For developing, debugging, and deploying AI agents and LLM applications, see [LangSmith](https:\u002F\u002Fdocs.langchain.com\u002Flangsmith\u002Fhome).\n\n## Customization\n\nAdd your own tools, swap models, customize prompts, configure sub-agents, and more. See the [documentation](https:\u002F\u002Fdocs.langchain.com\u002Foss\u002Fpython\u002Fdeepagents\u002Foverview) for full details.\n\n```python\nfrom langchain.chat_models import init_chat_model\n\nagent = create_deep_agent(\n    model=init_chat_model(\"openai:gpt-4o\"),\n    tools=[my_custom_tool],\n    system_prompt=\"You are a research assistant.\",\n)\n```\n\nMCP is supported via [`langchain-mcp-adapters`](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Flangchain-mcp-adapters).\n\n## Deep Agents CLI\n\nA pre-built coding agent in your terminal — similar to Claude Code or Cursor — powered by any LLM. One install command and you're up and running.\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Flangchain-ai_deepagents_readme_24f8f3b1f035.png\" alt=\"Deep Agents CLI\" width=\"600\"\u002F>\n\u003C\u002Fp>\n\n```bash\ncurl -LsSf https:\u002F\u002Fraw.githubusercontent.com\u002Flangchain-ai\u002Fdeepagents\u002Fmain\u002Flibs\u002Fcli\u002Fscripts\u002Finstall.sh | bash\n```\n\n**Highlights:**\n\n- **Interactive TUI** — rich terminal interface with streaming responses\n- **Web search** — ground responses in live information\n- **Headless mode** — run non-interactively for scripting and CI\n- Plus all SDK features out of the box — remote sandboxes, persistent memory, custom skills, and human-in-the-loop approval\n\nSee the [CLI documentation](https:\u002F\u002Fdocs.langchain.com\u002Foss\u002Fpython\u002Fdeepagents\u002Fcli\u002Foverview) for the full feature set.\n\n## LangGraph Native\n\n`create_deep_agent` returns a compiled [LangGraph](https:\u002F\u002Fdocs.langchain.com\u002Foss\u002Fpython\u002Flanggraph\u002Foverview) graph. Use it with streaming, Studio, checkpointers, or any LangGraph feature.\n\n## FAQ\n\n### Why should I use this?\n\n- **100% open source** — MIT licensed, fully extensible\n- **Provider agnostic** — Works with any Large Language Model that supports tool calling, including both frontier and open models\n- **Built on LangGraph** — Production-ready runtime with streaming, persistence, and checkpointing\n- **Batteries included** — Planning, file access, sub-agents, and context management work out of the box\n- **Get started in seconds** — `uv add deepagents` and you have a working agent\n- **Customize in minutes** — Add tools, swap models, tune prompts when you need to\n\n---\n\n## Documentation\n\n- [docs.langchain.com](https:\u002F\u002Fdocs.langchain.com\u002Foss\u002Fpython\u002Fdeepagents\u002Foverview) – Comprehensive documentation, including conceptual overviews and guides\n- [reference.langchain.com\u002Fpython](https:\u002F\u002Freference.langchain.com\u002Fpython\u002Fdeepagents\u002F) – API reference docs for Deep Agents packages\n- [Chat LangChain](https:\u002F\u002Fchat.langchain.com\u002F) – Chat with the LangChain documentation and get answers to your questions\n\n**Discussions**: Visit the [LangChain Forum](https:\u002F\u002Fforum.langchain.com) to connect with the community and share all of your technical questions, ideas, and feedback.\n\n## Additional resources\n\n- **[Examples](examples\u002F)** — Working agents and patterns\n- [Contributing Guide](https:\u002F\u002Fdocs.langchain.com\u002Foss\u002Fpython\u002Fcontributing\u002Foverview) – Learn how to contribute to LangChain projects and find good first issues.\n- [Code of Conduct](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Flangchain\u002F?tab=coc-ov-file) – Our community guidelines and standards for participation.\n\n---\n\n## Acknowledgements\n\nThis project was primarily inspired by Claude Code, and initially was largely an attempt to see what made Claude Code general purpose, and make it even more so.\n\n## Security\n\nDeep Agents follows a \"trust the LLM\" model. The agent can do anything its tools allow. Enforce boundaries at the tool\u002Fsandbox level, not by expecting the model to self-police. See the [security policy](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents?tab=security-ov-file) for more information.\n","\u003Cdiv align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fdocs.langchain.com\u002Foss\u002Fpython\u002Fdeepagents\u002Foverview#deep-agents-overview\">\n    \u003Cpicture>\n      \u003Csource media=\"(prefers-color-scheme: dark)\" srcset=\".github\u002Fimages\u002Flogo-dark.svg\">\n      \u003Csource media=\"(prefers-color-scheme: light)\" srcset=\".github\u002Fimages\u002Flogo-light.svg\">\n      \u003Cimg alt=\"Deep Agents Logo\" src=\".github\u002Fimages\u002Flogo-dark.svg\" width=\"50%\">\n    \u003C\u002Fpicture>\n  \u003C\u002Fa>\n\u003C\u002Fdiv>\n\n\u003Cdiv align=\"center\">\n  \u003Ch3>开箱即用的智能体框架。\u003C\u002Fh3>\n\u003C\u002Fdiv>\n\n\u003Cdiv align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fopensource.org\u002Flicenses\u002FMIT\" target=\"_blank\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fl\u002Fdeepagents\" alt=\"PyPI - 许可证\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fpypistats.org\u002Fpackages\u002Fdeepagents\" target=\"_blank\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fpepy\u002Fdt\u002Fdeepagents\" alt=\"PyPI - 下载量\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fpypi.org\u002Fproject\u002Fdeepagents\u002F#history\" target=\"_blank\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fv\u002Fdeepagents?label=%20\" alt=\"版本\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fx.com\u002Flangchain\" target=\"_blank\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Ftwitter\u002Furl\u002Fhttps\u002Ftwitter.com\u002Flangchain.svg?style=social&label=关注%20@LangChain\" alt=\"Twitter \u002F X\">\u003C\u002Fa>\n\u003C\u002Fdiv>\n\n\u003Cbr>\n\nDeep Agents 是一个智能体框架，提供了一套开箱即用、高度集成的解决方案。您无需手动配置提示词、工具和上下文管理逻辑，即可快速获得一个可用的智能体，并根据需求进行定制。\n\n**包含的功能：**\n\n- **规划** — `write_todos` 用于任务分解和进度跟踪\n- **文件系统** — `read_file`、`write_file`、`edit_file`、`ls`、`glob`、`grep` 用于读取和写入上下文\n- **Shell 访问** — `execute` 用于执行命令（并带有沙盒隔离）\n- **子智能体** — `task` 用于在独立的上下文窗口中委派工作\n- **智能默认设置** — 提示词会引导模型高效使用这些工具\n- **上下文管理** — 当对话过长时自动摘要，大型输出会保存到文件中\n\n> [!注意]\n> 如果您正在寻找 JS\u002FTS 版本的库，请查看 [deepagents.js](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagentsjs)。\n\n## 快速入门\n\n```bash\npip install deepagents\n# 或者\nuv add deepagents\n```\n\n```python\nfrom deepagents import create_deep_agent\n\nagent = create_deep_agent()\nresult = agent.invoke({\"messages\": [{\"role\": \"user\", \"content\": \"研究 LangGraph 并撰写总结\"}]})\n```\n\n该智能体能够进行规划、读写文件以及管理自身上下文。您可以根据需要添加工具、自定义提示词或更换模型。\n\n> [!提示]\n> 如需开发、调试和部署 AI 智能体及 LLM 应用程序，请参阅 [LangSmith](https:\u002F\u002Fdocs.langchain.com\u002Flangsmith\u002Fhome)。\n\n## 自定义\n\n您可以添加自己的工具、更换模型、自定义提示词、配置子智能体等。完整细节请参阅[文档](https:\u002F\u002Fdocs.langchain.com\u002Foss\u002Fpython\u002Fdeepagents\u002Foverview)。\n\n```python\nfrom langchain.chat_models import init_chat_model\n\nagent = create_deep_agent(\n    model=init_chat_model(\"openai:gpt-4o\"),\n    tools=[my_custom_tool],\n    system_prompt=\"你是一名研究助理。\",\n)\n```\n\nMCP 支持通过 [`langchain-mcp-adapters`](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Flangchain-mcp-adapters) 实现。\n\n## Deep Agents CLI\n\n终端中的预构建编码智能体——类似于 Claude Code 或 Cursor——由任何 LLM 驱动。只需一条安装命令，即可立即使用。\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Flangchain-ai_deepagents_readme_24f8f3b1f035.png\" alt=\"Deep Agents CLI\" width=\"600\"\u002F>\n\u003C\u002Fp>\n\n```bash\ncurl -LsSf https:\u002F\u002Fraw.githubusercontent.com\u002Flangchain-ai\u002Fdeepagents\u002Fmain\u002Flibs\u002Fcli\u002Fscripts\u002Finstall.sh | bash\n```\n\n**亮点：**\n\n- **交互式 TUI** — 丰富的终端界面，支持流式响应\n- **网络搜索** — 基于实时信息生成回复\n- **无头模式** — 可以非交互式运行，适用于脚本和 CI\n- 此外，还具备 SDK 的所有功能——远程沙盒、持久化存储、自定义技能以及人工审核流程\n\n完整的功能列表请参阅[CLI 文档](https:\u002F\u002Fdocs.langchain.com\u002Foss\u002Fpython\u002Fdeepagents\u002Fcli\u002Foverview)。\n\n## LangGraph 原生支持\n\n`create_deep_agent` 返回一个编译好的 [LangGraph](https:\u002F\u002Fdocs.langchain.com\u002Foss\u002Fpython\u002Flanggraph\u002Foverview) 图。您可以将其与流式处理、Studio、检查点等功能结合使用。\n\n## 常见问题解答\n\n### 为什么应该使用它？\n\n- **100% 开源** — MIT 许可，完全可扩展\n- **平台无关** — 适用于任何支持工具调用的大语言模型，包括前沿模型和开源模型\n- **基于 LangGraph 构建** — 生产级运行时，支持流式处理、持久化和检查点\n- **开箱即用** — 规划、文件访问、子智能体和上下文管理功能直接可用\n- **秒级上手** — 使用 `uv add deepagents` 即可获得一个可用的智能体\n- **分钟级定制** — 根据需求添加工具、更换模型或调整提示词\n\n---\n\n## 文档\n\n- [docs.langchain.com](https:\u002F\u002Fdocs.langchain.com\u002Foss\u002Fpython\u002Fdeepagents\u002Foverview) — 全面的文档，包括概念概述和指南\n- [reference.langchain.com\u002Fpython](https:\u002F\u002Freference.langchain.com\u002Fpython\u002Fdeepagents\u002F) — Deep Agents 包的 API 参考文档\n- [Chat LangChain](https:\u002F\u002Fchat.langchain.com\u002F) — 与 LangChain 文档聊天，获取问题解答\n\n**讨论区**：访问 [LangChain 论坛](https:\u002F\u002Fforum.langchain.com)，与社区交流技术问题、分享想法和反馈。\n\n## 更多资源\n\n- **[示例](examples\u002F)** — 可运行的智能体和常见模式\n- [贡献指南](https:\u002F\u002Fdocs.langchain.com\u002Foss\u002Fpython\u002Fcontributing\u002Foverview) — 学习如何为 LangChain 项目做出贡献，并找到适合初学者的问题\n- [行为准则](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Flangchain\u002F?tab=coc-ov-file) — 我们的社区参与规范和标准\n\n---\n\n## 致谢\n\n本项目的主要灵感来源于 Claude Code，最初主要是为了探究 Claude Code 的通用性，并在此基础上进一步增强其能力。\n\n## 安全性\n\nDeep Agents 采用“信任 LLM”的模式。智能体可以执行其工具允许的任何操作。应在工具或沙盒层面强制实施边界限制，而不是期望模型自我约束。更多信息请参阅[安全策略](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents?tab=security-ov-file)。","# Deep Agents 快速上手指南\n\nDeep Agents 是一个“开箱即用”的智能体框架（Agent Harness）。它内置了任务规划、文件系统操作、Shell 执行、子代理委托及上下文管理等核心能力，让开发者无需手动编排提示词和工具，即可快速构建功能强大的 AI 智能体。\n\n## 环境准备\n\n*   **操作系统**：支持 Linux、macOS 和 Windows。\n*   **Python 版本**：建议 Python 3.10 或更高版本。\n*   **前置依赖**：\n    *   推荐使用 `uv` 进行包管理（速度更快），也可使用标准的 `pip`。\n    *   需要配置好大模型 API Key（如 OpenAI、Anthropic 等），Deep Agents 本身不限制模型提供商。\n\n> **国内加速建议**：\n> 如果安装速度较慢，可使用国内镜像源：\n> *   **pip**: 添加 `-i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple` 参数。\n> *   **uv**: 设置环境变量 `export UV_INDEX_URL=https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple`。\n\n## 安装步骤\n\n你可以选择以下任意一种方式进行安装：\n\n**方式一：使用 uv（推荐）**\n```bash\nuv add deepagents\n```\n\n**方式二：使用 pip**\n```bash\npip install deepagents\n# 或使用国内镜像\npip install deepagents -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n```\n\n**可选：安装 CLI 命令行工具**\n如果你希望在终端直接体验类似 Claude Code 的交互式编程助手，可运行：\n```bash\ncurl -LsSf https:\u002F\u002Fraw.githubusercontent.com\u002Flangchain-ai\u002Fdeepagents\u002Fmain\u002Flibs\u002Fcli\u002Fscripts\u002Finstall.sh | bash\n```\n\n## 基本使用\n\nDeep Agents 的核心是通过 `create_deep_agent` 快速创建一个具备规划、文件读写和上下文管理能力的智能体。\n\n### 1. 最小化示例\n\n以下代码展示了如何创建一个默认智能体并执行任务（研究 LangGraph 并撰写总结）：\n\n```python\nfrom deepagents import create_deep_agent\n\n# 创建默认智能体\nagent = create_deep_agent()\n\n# 调用智能体执行任务\nresult = agent.invoke({\"messages\": [{\"role\": \"user\", \"content\": \"Research LangGraph and write a summary\"}]})\n\n# 打印结果\nprint(result[\"messages\"][-1].content)\n```\n\n### 2. 自定义配置\n\n你可以轻松替换底层模型、添加自定义工具或修改系统提示词：\n\n```python\nfrom langchain.chat_models import init_chat_model\nfrom deepagents import create_deep_agent\n\n# 假设你有一个自定义工具 my_custom_tool\n# from my_tools import my_custom_tool \n\nagent = create_deep_agent(\n    model=init_chat_model(\"openai:gpt-4o\"),  # 指定模型\n    tools=[],                                # 添加额外工具列表\n    system_prompt=\"You are a research assistant.\", # 自定义系统提示词\n)\n\nresult = agent.invoke({\"messages\": [{\"role\": \"user\", \"content\": \"Analyze the current directory structure\"}]})\n```\n\n### 3. 核心特性说明\n\n创建后的智能体默认拥有以下能力，无需额外代码：\n*   **任务规划**：自动使用 `write_todos` 拆解复杂任务。\n*   **文件操作**：内置 `read_file`, `write_file`, `edit_file`, `ls`, `grep` 等工具。\n*   **Shell 访问**：可通过 `execute` 运行命令（默认在沙箱环境中）。\n*   **子代理**：支持通过 `task` 工具将子任务委托给具有独立上下文的子代理。\n*   **上下文管理**：对话过长时自动摘要，大输出自动保存为文件。\n\n> **提示**：`create_deep_agent` 返回的是一个编译好的 **LangGraph** 图，因此你可以直接使用 LangGraph 的高级功能，如流式输出（streaming）、检查点（checkpointing）和 LangSmith 调试。","某全栈开发者需要在两天内将遗留的单体 Python 项目重构为微服务架构，涉及大量代码分析、文件拆分及依赖调整。\n\n### 没有 deepagents 时\n- **手动规划易遗漏**：开发者需人工梳理数百个文件的依赖关系并制定迁移步骤，极易因疏忽导致逻辑断层或循环依赖。\n- **上下文窗口受限**：面对庞大的代码库，大模型无法一次性读取所有相关文件，开发者需反复手动复制粘贴代码片段，效率极低。\n- **子任务协调困难**：将“数据库迁移”与\"API 重写”等子任务分派给不同 AI 会话时，缺乏统一的上下文隔离与状态同步机制，导致信息割裂。\n- **文件操作风险高**：通过脚本批量修改文件时缺乏沙箱保护，一旦生成错误代码直接覆盖源文件，回滚成本巨大。\n\n### 使用 deepagents 后\n- **自动任务拆解**：deepagents 内置的 `write_todos` 工具自动分析项目结构，生成详细的迁移待办清单并实时追踪进度，确保无死角覆盖。\n- **智能文件系统交互**：利用 `read_file`、`grep` 等原生工具，deepagents 自主检索跨文件上下文，自动管理长对话摘要，无需人工干预数据投喂。\n- **高效子代理协作**：通过 `task` 工具动态 spawn 子代理，分别处理独立模块重构，各子任务拥有隔离上下文窗，最终由主代理统一合并成果。\n- **安全沙箱执行**：所有代码修改与 shell 命令均在沙箱环境中运行，支持 `edit_file` 原子化操作，确保在验证通过前不破坏原始代码库。\n\ndeepagents 将繁琐的工程化协调工作转化为自动化的多代理协作流，让开发者从“提示词工程师”回归到真正的架构决策者。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Flangchain-ai_deepagents_24f8f3b1.png","langchain-ai","LangChain","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Flangchain-ai_8e6aaeef.png","",null,"support@langchain.dev","https:\u002F\u002Fwww.langchain.com","https:\u002F\u002Fgithub.com\u002Flangchain-ai",[83,87,91],{"name":84,"color":85,"percentage":86},"Python","#3572A5",99.3,{"name":88,"color":89,"percentage":90},"Makefile","#427819",0.4,{"name":92,"color":93,"percentage":94},"Shell","#89e051",0.3,20942,2906,"2026-04-16T13:02:34","MIT","未说明",{"notes":101,"python":99,"dependencies":102},"该工具是一个基于 LangGraph 的 Agent 框架，本身不包含模型权重，因此无特定 GPU 或显存需求（取决于用户选择的大模型）。支持通过 pip 或 uv 安装。提供 CLI 命令行工具（类似 Claude Code），可通过脚本一键安装。安全模型基于‘信任 LLM'，建议在工具或沙箱层面执行权限隔离，而非依赖模型自我约束。",[103,104,105],"langchain","langgraph","langchain-mcp-adapters (可选)",[13,14,15],[65,103,104,108],"ai","2026-03-27T02:49:30.150509","2026-04-17T09:53:24.475797",[112,117,122,127,132,137],{"id":113,"question_zh":114,"answer_zh":115,"source_url":116},36818,"如何为子代理（subagent）启用检查点（checkpointer）并传递配置参数？","目前的实现中，直接在调用时传递配置参数给子代理以启用检查点较为困难，因为任务方法每次都会创建新状态。最干净的解决方案是为子代理显式启用 `checkpointer`。一旦启用，可以在 `before_mode` 阶段检查存储的状态，并使用 `jump_to` 直接跳转到 `ToolNode`，从而实现受控恢复，避免从头重新执行整个任务或子代理。手动通过其他机制传递历史状态非常复杂，不推荐。","https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F354",{"id":118,"question_zh":119,"answer_zh":120,"source_url":121},36819,"当 ToolMessage 内容过大导致 artifact 属性丢失时，该如何处理？","这是为了避免占用模型的上下文窗口而设计的限制。如果下游逻辑依赖 `artifact` 字段且不能丢失结构化数据，建议将 `artifact` 持久化到本地目录中（例如与工具结果相同的文件夹 `\u002Flarge_tool_results\u002F{sanitized_id}`），而不是试图在内存中保留大对象。虽然目前注入和传播 artifact 会违背节省内存的初衷，但为了确保运行时语义和防止数据丢失，应当考虑将其作为消息的一部分进行传播或持久化存储。","https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F1450",{"id":123,"question_zh":124,"answer_zh":125,"source_url":126},36820,"如何查看或记录子代理的工具调用（tool calls）日志？","现在可以通过在流式传输时设置 `subgraphs=True` 来实现子代理工具调用的记录和查看。此外，得益于相关 PR 的合并，现在在追踪（traces）中也能看到更详细的子代理调用视图。如果需要将特定 UI 工具的消息保留在主状态中用于持久化，可以修改 DeepAgents 添加自定义参数（如 `preserve_message_tool_names`），指定哪些工具名称对应的消息应向上传递到主状态，以避免污染主代理的提示词上下文。","https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F177",{"id":128,"question_zh":129,"answer_zh":130,"source_url":131},36821,"同时使用 ShellToolMiddleware 和 MemorySaver checkpointer 时报错\"Type is not msgpack serializable: Send\"怎么办？","这是一个上游依赖问题，已在 LangChain 的相关 PR 中修复。如果不需要并行 shell 执行功能，建议暂时继续使用默认的 `FilesystemMiddleware`，它在生产环境中已被广泛使用且稳定。该错误是由于 `Send` 类型无法被 msgpack 序列化导致的，等待上游修复或避免组合使用这两个组件即可解决。","https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F941",{"id":133,"question_zh":134,"answer_zh":135,"source_url":136},36822,"DeepAgents 无法执行 skills 目录中的 Python 脚本是什么原因？","确保 `skills` 路径正确指向包含技能定义的目录，并且每个技能目录下有正确的 `SKILL.md` 描述文件。检查后端（Backend）配置是否允许执行外部脚本，特别是当使用 `FilesystemBackend` 时，需确认根目录权限和工作空间设置是否正确。如果使用的是虚拟模式（virtual_mode），请确认脚本路径相对于后端根目录是可达的。此外，验证模型是否能正确解析技能描述并生成合法的调用指令。","https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F884",{"id":138,"question_zh":139,"answer_zh":140,"source_url":126},36823,"如何在保持子代理工作范围隔离的同时，将某些关键工具调用结果回传给主代理？","默认情况下，子代理的详细信息会被汇总后返回给主代理以保持上下文简洁。但如果需要将特定工具（尤其是 UI 相关工具）的调用消息保留在主状态中，可以通过扩展 DeepAgents 实现自定义逻辑。例如，引入一个名为 `preserve_message_tool_names` 的参数，传入需要保留的工具名称数组，系统会将这些工具对应的消息向上传递到主状态。这样既保持了子代理的封装性，又满足了特定场景下的持久化和前端交互需求。",[142,147,152,157,162,167,172,177,182,187,192,197,202,207,212,217,222,227,232,237],{"id":143,"version":144,"summary_zh":145,"released_at":146},297220,"deepagents-cli==0.0.38","### 功能特性\n\n* 用户作用域内存（[#2708](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2708)）（[23bfca6](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F23bfca6e46e6f3e4fba6657d858ddd5a0b06626f)）\n\n\n### 性能改进\n\n* `MessageStore` 中的 `O(1)` 消息查找（[#2350](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2350)）（[d39fd5d](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002Fd39fd5d3651fd87d1eea8c02cbef2c2f62449e67)）\n\n---\n\n感谢我们的社区贡献者：@octo-patch、@hzylyq\n\n内部维护人员：@sydney-runkle、@mdrxy","2026-04-15T13:16:42",{"id":148,"version":149,"summary_zh":150,"released_at":151},297221,"deepagents==0.5.3","### 功能特性\n\n* 为子代理响应添加静态结构化输出 ([#2437](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2437)) ([6e57731](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F6e57731fc6d908ac1ebe131e782696a4776147e9))\n* 弃用 `create_deep_agent` 中的 `model=None` 参数 ([#2677](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2677)) ([149df41](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F149df415d17f3cf3b7eb0bd1e78460112bfa9b04))\n\n\n### Bug 修复\n\n* 技能加载应默认限制为 1000 行 ([#2721](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2721)) ([badc4d3](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002Fbadc4d3921ae0ede4305f44f85fa7266df9465e7))\n\n---\n\n内部维护人员：@sydney-runkle、@maahir30、@mdrxy","2026-04-15T13:06:48",{"id":153,"version":154,"summary_zh":155,"released_at":156},297222,"deepagents-cli==0.0.37","### 功能特性\n\n* 为 `deepagents deploy` 添加权限控制 ([#2651](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2651)) ([5d93b73](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F5d93b736af6ffb165f33569233d533ced95a6943))\n\n### Bug 修复\n\n* 向 `deepagents deploy` 打包工具中添加缺失的模型提供商依赖项 [关闭了 [#2647](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2647)] ([#2660](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2660)) ([b710a69](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002Fb710a69b12e49479045eaa54dfb709326473500b))\n* 系统提示中的 `AGENTS.md` 出现两次 ([#2652](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2652)) ([9052be9](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F9052be98d9f4ef9b11a88c9b1df3fae5e5ac666c))\n* 加强 `deepagents deploy` 配置解析并添加单元测试 ([#2636](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2636)) ([0469d14](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F0469d1429d129e604fc1b622263923162f719314))\n* 在进行部署或开发配置验证之前加载 `deepagents deploy` 项目的 `.env` 文件 ([#2644](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2644)) ([8299091](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F829909166606f8a9d9571b00da725845bad08da7))\n\n---\n\n内部维护人员：@sydney-runkle、@hinthornw、@mdrxy","2026-04-10T20:47:07",{"id":158,"version":159,"summary_zh":160,"released_at":161},297223,"deepagents==0.5.2","### 功能特性\n\n* 文件系统访问控制权限系统（[#2633](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2633)）（[41dc759](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F41dc7597deb3fc036f1f850e68edc3c0870f27da)）\n  * 将权限范围限定于路由，适用于具有沙箱默认配置的复合后端（[#2659](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2659)）（[6dd6122](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F6dd612237a7ee707726c4cafc4b691704e4cdb37)）\n  * 对于没有前导斜杠及包含路径遍历的权限路径，抛出 `ValueError` 异常（[#2665](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2665)）（[723d27d](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F723d27dcdce03cc9ffaa757c70533f0134a43a44)）\n* 为 `StateBackend` 实现 `upload_files` 方法（[#2661](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2661)）（[5798345](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F579834513a4ba1a024a52fc4edf918f526eab5f2)）\n\n### 错误修复\n\n* 在 `FilesystemBackend` 的 ripgrep 操作中捕获 `PermissionError` 异常（[#2571](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2571)）（[3d5d673](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F3d5d67349c8e88e33af98137db9634742f018cb0)）\n\n---\n\n内部维护人员：@nick-hollon-lc、@sydney-runkle、@ccurme、@mdrxy","2026-04-10T20:40:07",{"id":163,"version":164,"summary_zh":165,"released_at":166},297224,"deepagents==0.5.2a3","> [!警告]\n> 这是一个预发布版本。请使用以下命令安装：`pip install deepagents==0.5.2a3`\n\n---\n\n内部维护人员：@nick-hollon-lc、@sydney-runkle、@ccurme、@mdrxy","2026-04-10T19:27:20",{"id":168,"version":169,"summary_zh":170,"released_at":171},297225,"deepagents==0.5.2a2","> [!警告]\n> 这是一个预发布版本。请使用以下命令安装：`pip install deepagents==0.5.2a2`\n\n---\n\n内部维护人员：@nick-hollon-lc、@sydney-runkle、@ccurme、@mdrxy","2026-04-10T15:25:48",{"id":173,"version":174,"summary_zh":175,"released_at":176},297226,"deepagents==0.5.2a1","> [!警告]\r\n> 这是一个预发布版本。请使用以下命令安装：`pip install deepagents==0.5.2a1`","2026-04-10T14:40:35",{"id":178,"version":179,"summary_zh":180,"released_at":181},297227,"deepagents-cli==0.0.36","### 功能特性\n\n* `deepagents deploy` ([#2491](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2491)) ([01dc60e](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F01dc60e394ecb56bd5336e447d32caeed8a67ec2))\n* 在缺少 Tavily API 密钥时发出警告，并新增 `\u002Fnotifications` 端点 ([#2555](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2555)) ([3dff3ed](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F3dff3ed6835eae9f8b420b8a73c054127faaf7d2))\n\n### 错误修复\n\n* 在缺少提供商凭据时快速失败 ([#2554](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2554)) ([50fb8ae](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F50fb8aefe7e3065024e10f1d5ecd11a54d736641))\n* 修复 macOS 上调试脚本中的 `mktemp` 模板问题 ([#2603](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2603)) ([63fa537](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F63fa537e9995ca2ead492ee44902227567e9a130))\n\n---\n\n内部维护人员：@sydney-runkle，@mdrxy","2026-04-09T15:44:48",{"id":183,"version":184,"summary_zh":185,"released_at":186},297228,"deepagents==0.5.1","### 功能特性\n\n* **SDK:** `BASE_AGENT_PROMPT` 优化调整 ([#2541](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2541)) ([812eef1](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F812eef185ffda7bc9e6f11425eb5eddc3d3b32e8))\n* **SDK:** 向 `CompositeBackend` 和中间件添加 `artifacts_root` 配置项 ([#2490](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2490)) ([753ee56](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F753ee567f1cc4d544dc2afea7b414564fd07d37d))\n\n\n### Bug 修复\n\n* **SDK:** 多模态相关更新 ([#2514](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2514)) ([a2edf3e](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002Fa2edf3ed80e17a87027c41a46283387031ebd3e5))\n\n---\n\n内部维护人员：@vtrivedy、@nick-hollon-lc、@ccurme、@mdrxy","2026-04-07T22:58:14",{"id":188,"version":189,"summary_zh":190,"released_at":191},297229,"langchain-repl==0.0.1","初始发布\ncr\n热修复(repl)：存根包\n发布(deepagents)：0.5.0 (#2509)\n杂项(deps)：在 1 个目录中将 uv 组的版本提升，包含 3 次更新 (#2487)\n功能(repl)：添加用于测试的最小化 REPL (#2482)","2026-04-07T19:36:53",{"id":193,"version":194,"summary_zh":195,"released_at":196},297230,"langchain-runloop==0.0.4","Changes since langchain-runloop==0.0.3\nrelease(runloop): 0.0.4 (#2527)\nrelease(deepagents): 0.5.0 (#2509)\nrelease: deepagents 0.5.0a4 (#2444)\nchore(sdk): bump lock files (#2430)\nrelease(deepagents): 0.5.0a3 (#2385)\nchore: bump pygments to 2.20.0 everywhere (GHSA-5239-wwwm-4pmq) (#2349)\nchore(deps): bump cryptography from 46.0.5 to 46.0.6 in \u002Flibs\u002Fpartners\u002Frunloop (#2281)\nchore(deps): bump langchain-core from 1.2.21 to 1.2.22 in \u002Flibs\u002Fpartners\u002Frunloop (#2280)\nchore: sync lockfiles (#2279)\nchore(deps): bump the uv group across 4 directories with 1 update (#2222)\nrelease(deepagents): 0.5.0a2 (#2158)\nrelease(deepagents): 0.5.0a1 (#2157)\nchore(sdk): update langchain and langchain-anthropic deps (#2069)\nfeat(sdk): port `LangSmithSandbox` from CLI to SDK (#1983)\nrefactor(evals): consolidate eval suite into standalone `libs\u002Fevals` (#2032)\nci: suppress pytest streaming output in CI (#2026)\nchore: promote `PLR6201` to safe-fix across all packages (#2000)\ntest(sdk): add wall-time benchmarks for `create_deep_agent` (#1944)\nchore(deps): bump the uv group across 4 directories with 1 update (#1963)\nchore(sdk): increment version to 0.5 (#1908)\nchore(sdk,cli): make `langchain-anthropic` and `langchain-google-genai` explicit deps (#1886)\nrelease(sdk): 0.4.11 (#1868)\nchore(sdk): make `langsmith` explicit evals dep (#1867)\ntest(sdk): add `ChatBaseten` for evals (#1839)\nfeat(sdk): add MemoryAgentBench evaluation suite (#1807)\nrelease(sdk): prepare 0.4.10 patch release (#1795)\nrelease(deepagents): 0.4.9 (#1773)\nrelease(deepagents): 0.4.8 (#1764)\n\n---\n\nThanks to our community contributors: @sydney-runkle","2026-04-07T18:12:04",{"id":198,"version":199,"summary_zh":200,"released_at":201},297231,"langchain-modal==0.0.3","Changes since langchain-modal==0.0.2\nrelease(modal): 0.0.3 (#2525)\nrelease(deepagents): 0.5.0 (#2509)\nrelease: deepagents 0.5.0a4 (#2444)\nchore(sdk): bump lock files (#2430)\nrelease(deepagents): 0.5.0a3 (#2385)\nchore: bump pygments to 2.20.0 everywhere (GHSA-5239-wwwm-4pmq) (#2349)\nchore(deps): bump cryptography from 46.0.5 to 46.0.6 in \u002Flibs\u002Fpartners\u002Fmodal (#2286)\nchore(deps): bump langchain-core from 1.2.21 to 1.2.22 in \u002Flibs\u002Fpartners\u002Fmodal (#2285)\nchore: sync lockfiles (#2279)\nchore(deps): bump the uv group across 4 directories with 1 update (#2222)\nchore(deps): bump the uv group across 2 directories with 1 update (#2161)\nrelease(deepagents): 0.5.0a2 (#2158)\nrelease(deepagents): 0.5.0a1 (#2157)\nchore(sdk): update langchain and langchain-anthropic deps (#2069)\nfeat(sdk): port `LangSmithSandbox` from CLI to SDK (#1983)\nrefactor(evals): consolidate eval suite into standalone `libs\u002Fevals` (#2032)\nci: suppress pytest streaming output in CI (#2026)\nchore: promote `PLR6201` to safe-fix across all packages (#2000)\ntest(sdk): add wall-time benchmarks for `create_deep_agent` (#1944)\nchore(deps): bump the uv group across 4 directories with 1 update (#1963)\nchore(sdk): increment version to 0.5 (#1908)\nchore(sdk,cli): make `langchain-anthropic` and `langchain-google-genai` explicit deps (#1886)\nrelease(sdk): 0.4.11 (#1868)\nchore(sdk): make `langsmith` explicit evals dep (#1867)\ntest(sdk): add `ChatBaseten` for evals (#1839)\nfeat(sdk): add MemoryAgentBench evaluation suite (#1807)\nrelease(sdk): prepare 0.4.10 patch release (#1795)\nrelease(deepagents): 0.4.9 (#1773)\nrelease(deepagents): 0.4.8 (#1764)\n\n---\n\nThanks to our community contributors: @sydney-runkle","2026-04-07T18:02:35",{"id":203,"version":204,"summary_zh":205,"released_at":206},297232,"langchain-daytona==0.0.5","Changes since langchain-daytona==0.0.4\nrelease(daytona): 0.0.5 (#2523)\nrelease(deepagents): 0.5.0 (#2509)\nrelease: deepagents 0.5.0a4 (#2444)\nchore(sdk): bump lock files (#2430)\nrelease(deepagents): 0.5.0a3 (#2385)\nchore: bump pygments to 2.20.0 everywhere (GHSA-5239-wwwm-4pmq) (#2349)\nchore: bump requests (#2325)\nchore(deps): bump cryptography from 46.0.5 to 46.0.6 in \u002Flibs\u002Fpartners\u002Fdaytona (#2288)\nchore(deps): bump langchain-core from 1.2.21 to 1.2.22 in \u002Flibs\u002Fpartners\u002Fdaytona (#2287)\nchore: sync lockfiles (#2279)\nrelease(deepagents): 0.5.0a2 (#2158)\nrelease(deepagents): 0.5.0a1 (#2157)\nchore(sdk): update langchain and langchain-anthropic deps (#2069)\nfeat(sdk): port `LangSmithSandbox` from CLI to SDK (#1983)\nrefactor(evals): consolidate eval suite into standalone `libs\u002Fevals` (#2032)\nci: suppress pytest streaming output in CI (#2026)\nchore: promote `PLR6201` to safe-fix across all packages (#2000)\ntest(sdk): add wall-time benchmarks for `create_deep_agent` (#1944)\nchore(deps): bump pyasn1 from 0.6.2 to 0.6.3 in \u002Flibs\u002Fpartners\u002Fdaytona (#1965)\nchore(sdk): increment version to 0.5 (#1908)\nchore(sdk,cli): make `langchain-anthropic` and `langchain-google-genai` explicit deps (#1886)\nrelease(sdk): 0.4.11 (#1868)\nchore(sdk): make `langsmith` explicit evals dep (#1867)\ntest(sdk): add `ChatBaseten` for evals (#1839)\nfeat(sdk): add MemoryAgentBench evaluation suite (#1807)\nchore(deps): bump multipart from 1.3.0 to 1.3.1 in \u002Flibs\u002Fpartners\u002Fdaytona (#1842)","2026-04-07T17:46:32",{"id":208,"version":209,"summary_zh":210,"released_at":211},297233,"deepagents-acp==0.0.5","Changes since deepagents-acp==0.0.4\nrelease(acp): 0.0.5 (#2521)\nrelease(deepagents): 0.5.0 (#2509)\nfix(acp): block dangerous shell patterns in command auto-approve (#2308)\nrelease: deepagents 0.5.0a4 (#2444)\nchore(sdk): bump lock files (#2430)\nfix(acp): Fix issue in ACP demo agent (#2422)\nfix: deprecate backend factories (#2360)\nchore: bump pygments to 2.20.0 everywhere (GHSA-5239-wwwm-4pmq) (#2349)\nchore: bump requests (#2325)\nchore(deps): bump langchain-core from 1.2.19 to 1.2.22 in \u002Flibs\u002Facp (#2296)\nchore(deps): bump cryptography from 46.0.5 to 46.0.6 in \u002Flibs\u002Facp (#2295)\nfeat(acp): model switching (#2135)\nfix(acp): cwd for agent session context (#2130)\nci: suppress pytest streaming output in CI (#2026)\nchore: promote `PLR6201` to safe-fix across all packages (#2000)\nchore(deps): bump pyasn1 from 0.6.2 to 0.6.3 in \u002Flibs\u002Facp (#1969)\nfeat(cli): `--acp` mode to run CLI agent as ACP server (#1297)\nchore: bump lock file (#1674)\nchore: standardize makefile type targets\u002F add type CI check (#1581)\nchore(deps): bump the pip-dependencies group across 8 directories with 17 updates (#1564)\nchore(acp): Update example ACP agent lockfile (#1522)\nchore(deps): bump the pip-dependencies group across 8 directories with 19 updates (#1491)\ndocs: Add ACP example agent upgrade instructions (#1475)\nfix(acp): Remove CLI dep from ACP demo agent (#1474)\nchore: auto-generate makefile help from target annotations (#1417)\nchore: enable type checking for all other packages (#1365)\nfeat(acp): Show raw input in tool calls (#1349)\nfix(acp): Fix double approval bug (#1335)\nfeat(acp): Add local context middleware to ACP example agent, update script working dir (#1332)\nfix(acp): Fix mode switching for ACP client (#1331)\n\n---\n\nThanks to our community contributors: @sydney-runkle, @eyurtsev, @jacoblee93, @mdrxy","2026-04-07T17:31:32",{"id":213,"version":214,"summary_zh":215,"released_at":216},297234,"deepagents-cli==0.0.35","### Highlights\n\n* **Skills:** Invoke SDK skills directly from the CLI with `\u002Fskill:name` or at startup via `--skill`. Skills are a composable extension point for domain-specific agent behaviors.\n* **Themes & configuration:** New theme system with color overrides, global dotenv at `~\u002F.deepagents\u002F.env`, and a `DEEPAGENTS_CLI_` env var prefix for conflict-free configuration.\n* **Auto-updates:** Full update lifecycle — `\u002Fupdate` to upgrade in-place, `\u002Fauto-update` to toggle background checks, and a refreshed install script UX.\n* **Headless workflows:** Agent-friendly UX for scripted\u002Fheadless invocations, AgentCore Code Interpreter sandbox provider, and improved non-interactive tracing and guidance.\n* **Performance:** Sub-250ms first paint via aggressive import deferral (pydantic, adapters, heavy SDK modules), markdown stack prewarming, and reduced health-poll intervals.\n\n### Features\n\n* Load `~\u002F.deepagents\u002F.env` as global dotenv ([#1909](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F1909)) ([5a21d0a](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F5a21d0add1d2a885c6cd3cf36621d1da690c8db3))\n* Skill invocation via `\u002Fskill:name` ([#2037](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2037)) ([cc8cce7](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002Fcc8cce70e1d6c5f897a3cebe0388aa3d774de487))\n  * `--skill` startup invocation ([#2477](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2477)) ([5f0f1d4](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F5f0f1d4605de1f1545846e2dbcc29a24c6f138d7))\n* Auto-update lifecycle, `\u002Fupdate` command, install script ux ([#2095](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2095)) ([fd92f6e](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002Ffd92f6eaa87bb1a397f1b7fd216657f354c46e0f))\n  * `\u002Fauto-update` to toggle auto-updates ([#2276](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2276)) ([ad70bde](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002Fad70bde070c58057ff98871fcacd95f03487051e))\n* Themes ([#2134](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2134)) ([db67af0](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002Fdb67af07984b5d73711383cf717b5e3d96eac3b8))\n  * Allow color overrides on built-in themes, default dark to false ([#2275](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2275)) ([8f71865](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F8f718650ee1913f83e70aa7854e576edc816e694))\n* Add `DEEPAGENTS_CLI_` env var prefix and fix dotenv load order ([#2303](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2303)) ([29647bb](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F29647bb4cdd89a1de65f6adb518f708135d59e03))\n* Add `ls_integration` metadata to langsmith traces ([#2272](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2272)) ([5dd8098](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F5dd80983a57a5c48bb9c75c6760733882cb908c2))\n* Add animated spinner to non-interactive verbose mode ([#2001](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2001)) ([153f465](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F153f465937c09a792119a0e0c8656fa8df29d4e5))\n* Add async backend support to local context middleware ([#2118](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2118)) ([a0d623c](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002Fa0d623cf18d216b5de36cc7d40e7804d5cc4dfa3))\n* Agent-friendly ux for scripted\u002Fheadless workflows ([#2271](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2271)) ([386438f](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F386438f62a458baa367468cf746c3a5387a217d4))\n* AgentCore Code Interpreter sandbox provider ([#2120](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2120)) ([92556c7](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F92556c767aaab601d3ea753baf7f52642e5769a4))\n* Context-aware connecting banner for resume and local server ([#2092](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2092)) ([18b385b](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F18b385be979d461797a76b6d7217e259fe260b2c))\n* Default langsmith project to `'deepagents-cli'` ([#2277](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2277)) ([7178b87](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F7178b8729a03ee1e5508fdf6c111488d18e77e8f))\n* Enhance tool-call UI, add `Ctrl+U` shortcut for chat input ([#1757](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F1757)) ([800c552](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F800c55213aa4c6515759fb70d36af370feb86302))\n* Persist token count in graph state across sessions ([#2323](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2323)) ([5be352d](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002F5be352d8bb4d266758f3e934f7affbe8d42b0149))\n* Pop queued messages individually on `esc` instead of clearing all ([#2089](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2089)) ([c76d855](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fcommit\u002Fc76d855e4c73a448af17ade8ad54a2071f2c6bfe))\n* Render ask-user questions as markdown ([#2339](https:\u002F\u002Fgithub.com\u002Flangchain-ai\u002Fdeepagents\u002Fissues\u002F2339)) ([5fbb14a](https:\u002F\u002Fgithu","2026-04-07T16:34:46",{"id":218,"version":219,"summary_zh":220,"released_at":221},297235,"deepagents==0.5.0","See [changelog](https:\u002F\u002Fdocs.langchain.com\u002Foss\u002Fpython\u002Freleases\u002Fchangelog) for high-level summary.\r\n\r\nDetails below:\r\n\r\nChanges since deepagents==0.4.12\r\nrelease(deepagents): 0.5.0 (#2509)\r\nfix(sdk): match example_agent_descriptions closing tag in task tool prompt (#2462)\r\ndocs(sdk,cli): fill missing docstrings in model-specific code (#2481)\r\nchore: replace old OAI model strings (#2480)\r\ndocs(sdk): updates for `create_deep_agent` (#2479)\r\nfix(sdk): improvements in default sandbox.write and sandbox.read implementations (#2321)\r\nrelease: deepagents 0.5.0a4 (#2444)\r\nfix: remove legacy subagents API (#2443)\r\nchore(sdk): delete unused `base_prompt.md`, restore missing sections (#2436)\r\nchore(sdk): update unset logic (#2435)\r\nfix(sdk): inherit parent `interrupt_on` for subagents (#2334)\r\ntest(sdk): added a failing test for callback propagation to subagent … (#2361)\r\nchore(sdk): bump lock files (#2430)\r\nchore(sdk): update docs around async subagents (#2429)\r\nrelease(deepagents): 0.5.0a3 (#2385)\r\nfix: plumb through generics for `create_deep_agent` (#2383)\r\nfix(sdk): fix TypeError in async sub-agents (#2376)\r\nfix(sdk): `last_updated_at` field doesn't account for task status changes (#2370)\r\nfix: deprecate backend factories (#2360)\r\nchore: bump pygments to 2.20.0 everywhere (GHSA-5239-wwwm-4pmq) (#2349)\r\nfix(sdk): catch `UnicodeDecodeError` in `FilesystemBackend.read` (#2319)\r\nfix(sdk): restore deprecated protocol return types (#2342)\r\ndocs: add threat model for deepagents monorepo (#1639)\r\nfix(sdk): add new line after HEREDOC for edit inline sandbox (#2340)\r\nchore(sdk): disable benchmark when not running benchmarks (#2338)\r\ntest(sdk): assert default toolruntime config in end-to-end test (#2324)\r\nfix(sdk): fix offloading for state backend (#2266)\r\nchore: bump requests (#2325)\r\nchore(deps): bump cryptography from 46.0.5 to 46.0.6 in \u002Flibs\u002Fdeepagents (#2292)\r\nchore(deps): bump langchain-core from 1.2.21 to 1.2.22 in \u002Flibs\u002Fdeepagents (#2291)\r\nchore: sync lockfiles (#2279)\r\nfix(sdk): update recursion limit (#2273)\r\nchore(sdk): add end-to-end test for compaction tool (#2268)\r\nchore(sdk): declare schema up front for remaining tools (#2264)\r\nchore(sdk): speed up init for file system, summarization middleware (#2257)\r\nfix(sdk): normalize CRLF line endings in `FilesystemBackend.edit()` (#2256)\r\ntest(sdk): add a test for invalid args into a tool (#2254)\r\nchore(sdk): add all snapshots (#2253)\r\nchore(sdk): add snapshot of tool descriptions (#2252)\r\nfix(sdk): make created_at and modified_at NotRequired fields on FileData (#2248)\r\nfeat(sdk): evict large HumanMessages (#2183)\r\nfix(sdk): bump recursion limit to 10_000 in deepagents and set agent name (#2194)\r\nfix(sdk): route subagent model resolution to `resolve_model` (#2208)\r\nfeat(sdk,cli): add openrouter SDK attribution (#2205)\r\nchore(sdk): minor clean up (#2186)\r\nfix(sdk,cli): align error messages and clean up recent refactors (#2171)\r\nrefactor(sdk): type-only exception classification in `map_file_operation_error` (#2168)\r\nfix(sdk): use file transfer instead of command strings for sandbox write\u002Fedit (#2117)\r\ndocs(sdk): improve `sandbox.py` docstrings and refactor read template (#2169)\r\ndocs(sdk,evals): fix docstring formatting and harbor grep literal flag (#2163)\r\nrelease(deepagents): 0.5.0a2 (#2158)\r\nrelease(deepagents): 0.5.0a1 (#2157)\r\nfeat(sdk,cli): add package version metadata to traces (#2129)\r\nrefactor(sdk): rename tools for async subgents (#2103)\r\nfeat(sdk): add timestamp tracking to `AsyncSubAgentJob` (#2071)\r\nfix(sdk): overwrite `write` on LangSmithSandbox (#2097)\r\nfix(sdk): update middleware order (#2070)\r\nfeat(sdk): add system prompt snapshot test with sync and async subagents (#2072)\r\nchore(sdk): update langchain and langchain-anthropic deps (#2069)\r\nchore(sdk): merge async subagents and subagents into single top level param (#2064)\r\nfeat(sdk): port `LangSmithSandbox` from CLI to SDK (#1983)\r\nrefactor(evals): consolidate eval suite into standalone `libs\u002Fevals` (#2032)\r\nci: fix evals dataclass (#2034)\r\nci: suppress pytest streaming output in CI (#2026)\r\nfeat(sdk): add tau2 LICENSE (#2025)\r\nfeat(sdk): add tau2 airline eval harness (#1987)\r\ntest(sdk): gather `MemoryAgentBench` CI subset (#1993)\r\ntest(sdk): add curated external benchmark evals (#2018)\r\ntest(sdk): add evals for llm judge, tool selection, followup quality, and multi-turn memory (#1838)\r\nchore: promote `PLR6201` to safe-fix across all packages (#2000)\r\ntest(sdk): add wall-time benchmarks for `create_deep_agent` (#1944)\r\nchore(deps): bump pyasn1 from 0.6.2 to 0.6.3 in \u002Flibs\u002Fdeepagents (#1967)\r\nchore(quickjs): add some unit tests and disable tests from running (#1962)\r\nfeat(sdk): rename backend methods (#1907)\r\ntest(sdk): update evals for relational tool usage (#1940)\r\ntest(sdk): add fixture for summarization tests (#1627)\r\nfeat(sdk): add async subagent middleware for remote LangGraph servers (#1758)\r\nfix(sdk): simplify memory agent bench evals (#1873)\r\nfeat(sdk): update backend return types for ls, glob, gre","2026-04-07T16:08:41",{"id":223,"version":224,"summary_zh":225,"released_at":226},297236,"deepagents==0.5.0a4","Initial release\nrelease: deepagents 0.5.0a4 (#2444)\nfix: remove legacy subagents API (#2443)\nchore(sdk): delete unused `base_prompt.md`, restore missing sections (#2436)\nchore(sdk): update unset logic (#2435)\nfix(sdk): inherit parent `interrupt_on` for subagents (#2334)\ntest(sdk): added a failing test for callback propagation to subagent … (#2361)\nchore(sdk): bump lock files (#2430)\nchore(sdk): update docs around async subagents (#2429)\nrelease(deepagents): 0.5.0a3 (#2385)\nfix: plumb through generics for `create_deep_agent` (#2383)\nfix(sdk): fix TypeError in async sub-agents (#2376)\nfix(sdk): `last_updated_at` field doesn't account for task status changes (#2370)\nfix: deprecate backend factories (#2360)\nchore: bump pygments to 2.20.0 everywhere (GHSA-5239-wwwm-4pmq) (#2349)\nfix(sdk): catch `UnicodeDecodeError` in `FilesystemBackend.read` (#2319)\nfix(sdk): restore deprecated protocol return types (#2342)\ndocs: add threat model for deepagents monorepo (#1639)\nfix(sdk): add new line after HEREDOC for edit inline sandbox (#2340)\nchore(sdk): disable benchmark when not running benchmarks (#2338)\ntest(sdk): assert default toolruntime config in end-to-end test (#2324)\nfix(sdk): fix offloading for state backend (#2266)\nchore: bump requests (#2325)\nchore(deps): bump cryptography from 46.0.5 to 46.0.6 in \u002Flibs\u002Fdeepagents (#2292)\nchore(deps): bump langchain-core from 1.2.21 to 1.2.22 in \u002Flibs\u002Fdeepagents (#2291)\nchore: sync lockfiles (#2279)\nfix(sdk): update recursion limit (#2273)\nchore(sdk): add end-to-end test for compaction tool (#2268)\nchore(sdk): declare schema up front for remaining tools (#2264)\nchore(sdk): speed up init for file system, summarization middleware (#2257)\nfix(sdk): normalize CRLF line endings in `FilesystemBackend.edit()` (#2256)\ntest(sdk): add a test for invalid args into a tool (#2254)\nchore(sdk): add all snapshots (#2253)\nchore(sdk): add snapshot of tool descriptions (#2252)\nfix(sdk): make created_at and modified_at NotRequired fields on FileData (#2248)\nfeat(sdk): evict large HumanMessages (#2183)\nfix(sdk): bump recursion limit to 10_000 in deepagents and set agent name (#2194)\nfix(sdk): route subagent model resolution to `resolve_model` (#2208)\nfeat(sdk,cli): add openrouter SDK attribution (#2205)\nchore(sdk): minor clean up (#2186)\nfix(sdk,cli): align error messages and clean up recent refactors (#2171)\nrefactor(sdk): type-only exception classification in `map_file_operation_error` (#2168)\nfix(sdk): use file transfer instead of command strings for sandbox write\u002Fedit (#2117)\ndocs(sdk): improve `sandbox.py` docstrings and refactor read template (#2169)\ndocs(sdk,evals): fix docstring formatting and harbor grep literal flag (#2163)\nrelease(deepagents): 0.5.0a2 (#2158)\nrelease(deepagents): 0.5.0a1 (#2157)\nfeat(sdk,cli): add package version metadata to traces (#2129)\nrefactor(sdk): rename tools for async subgents (#2103)\nfeat(sdk): add timestamp tracking to `AsyncSubAgentJob` (#2071)\nfix(sdk): overwrite `write` on LangSmithSandbox (#2097)\nfix(sdk): update middleware order (#2070)\nfeat(sdk): add system prompt snapshot test with sync and async subagents (#2072)\nchore(sdk): update langchain and langchain-anthropic deps (#2069)\nchore(sdk): merge async subagents and subagents into single top level param (#2064)\nfeat(sdk): port `LangSmithSandbox` from CLI to SDK (#1983)\nrefactor(evals): consolidate eval suite into standalone `libs\u002Fevals` (#2032)\nci: fix evals dataclass (#2034)\nci: suppress pytest streaming output in CI (#2026)\nfeat(sdk): add tau2 LICENSE (#2025)\nfeat(sdk): add tau2 airline eval harness (#1987)\ntest(sdk): gather `MemoryAgentBench` CI subset (#1993)\ntest(sdk): add curated external benchmark evals (#2018)\ntest(sdk): add evals for llm judge, tool selection, followup quality, and multi-turn memory (#1838)\nchore: promote `PLR6201` to safe-fix across all packages (#2000)\ntest(sdk): add wall-time benchmarks for `create_deep_agent` (#1944)\nchore(deps): bump pyasn1 from 0.6.2 to 0.6.3 in \u002Flibs\u002Fdeepagents (#1967)\nchore(quickjs): add some unit tests and disable tests from running (#1962)\nfeat(sdk): rename backend methods (#1907)\ntest(sdk): update evals for relational tool usage (#1940)\ntest(sdk): add fixture for summarization tests (#1627)\nfeat(sdk): add async subagent middleware for remote LangGraph servers (#1758)\nfix(sdk): simplify memory agent bench evals (#1873)\nfeat(sdk): update backend return types for ls, glob, grep (#1870)\nfix(sdk): use Sequence instead of list for subagents parameter (#1905)\nchore(sdk): increment version to 0.5 (#1908)\nfeat(sdk): update FileData and backend.read return type to support multimodal data (#1869)\nchore(sdk,cli): make `langchain-anthropic` and `langchain-google-genai` explicit deps (#1886)\nrelease(sdk): 0.4.11 (#1868)\nchore(sdk): make `langsmith` explicit evals dep (#1867)\nchore(sdk): add docs link to `LocalShellBackend` `virtual_mode` deprecation warning (#1866)\ntest(sdk): add `ChatBaseten` for evals (#1839)\nfeat(sdk): Add LangSmith integration metadata to deep","2026-04-03T14:41:43",{"id":228,"version":229,"summary_zh":230,"released_at":231},297237,"deepagents==0.5.0a3","Initial release\nrelease(deepagents): 0.5.0a3 (#2385)\nfix: plumb through generics for `create_deep_agent` (#2383)\nfix(sdk): fix TypeError in async sub-agents (#2376)\nfix(sdk): `last_updated_at` field doesn't account for task status changes (#2370)\nfix: deprecate backend factories (#2360)\nchore: bump pygments to 2.20.0 everywhere (GHSA-5239-wwwm-4pmq) (#2349)\nfix(sdk): catch `UnicodeDecodeError` in `FilesystemBackend.read` (#2319)\nfix(sdk): restore deprecated protocol return types (#2342)\ndocs: add threat model for deepagents monorepo (#1639)\nfix(sdk): add new line after HEREDOC for edit inline sandbox (#2340)\nchore(sdk): disable benchmark when not running benchmarks (#2338)\ntest(sdk): assert default toolruntime config in end-to-end test (#2324)\nfix(sdk): fix offloading for state backend (#2266)\nchore: bump requests (#2325)\nchore(deps): bump cryptography from 46.0.5 to 46.0.6 in \u002Flibs\u002Fdeepagents (#2292)\nchore(deps): bump langchain-core from 1.2.21 to 1.2.22 in \u002Flibs\u002Fdeepagents (#2291)\nchore: sync lockfiles (#2279)\nfix(sdk): update recursion limit (#2273)\nchore(sdk): add end-to-end test for compaction tool (#2268)\nchore(sdk): declare schema up front for remaining tools (#2264)\nchore(sdk): speed up init for file system, summarization middleware (#2257)\nfix(sdk): normalize CRLF line endings in `FilesystemBackend.edit()` (#2256)\ntest(sdk): add a test for invalid args into a tool (#2254)\nchore(sdk): add all snapshots (#2253)\nchore(sdk): add snapshot of tool descriptions (#2252)\nfix(sdk): make created_at and modified_at NotRequired fields on FileData (#2248)\nfeat(sdk): evict large HumanMessages (#2183)\nfix(sdk): bump recursion limit to 10_000 in deepagents and set agent name (#2194)\nfix(sdk): route subagent model resolution to `resolve_model` (#2208)\nfeat(sdk,cli): add openrouter SDK attribution (#2205)\nchore(sdk): minor clean up (#2186)\nfix(sdk,cli): align error messages and clean up recent refactors (#2171)\nrefactor(sdk): type-only exception classification in `map_file_operation_error` (#2168)\nfix(sdk): use file transfer instead of command strings for sandbox write\u002Fedit (#2117)\ndocs(sdk): improve `sandbox.py` docstrings and refactor read template (#2169)\ndocs(sdk,evals): fix docstring formatting and harbor grep literal flag (#2163)\nrelease(deepagents): 0.5.0a2 (#2158)\nrelease(deepagents): 0.5.0a1 (#2157)\nfeat(sdk,cli): add package version metadata to traces (#2129)\nrefactor(sdk): rename tools for async subgents (#2103)\nfeat(sdk): add timestamp tracking to `AsyncSubAgentJob` (#2071)\nfix(sdk): overwrite `write` on LangSmithSandbox (#2097)\nfix(sdk): update middleware order (#2070)\nfeat(sdk): add system prompt snapshot test with sync and async subagents (#2072)\nchore(sdk): update langchain and langchain-anthropic deps (#2069)\nchore(sdk): merge async subagents and subagents into single top level param (#2064)\nfeat(sdk): port `LangSmithSandbox` from CLI to SDK (#1983)\nrefactor(evals): consolidate eval suite into standalone `libs\u002Fevals` (#2032)\nci: fix evals dataclass (#2034)\nci: suppress pytest streaming output in CI (#2026)\nfeat(sdk): add tau2 LICENSE (#2025)\nfeat(sdk): add tau2 airline eval harness (#1987)\ntest(sdk): gather `MemoryAgentBench` CI subset (#1993)\ntest(sdk): add curated external benchmark evals (#2018)\ntest(sdk): add evals for llm judge, tool selection, followup quality, and multi-turn memory (#1838)\nchore: promote `PLR6201` to safe-fix across all packages (#2000)\ntest(sdk): add wall-time benchmarks for `create_deep_agent` (#1944)\nchore(deps): bump pyasn1 from 0.6.2 to 0.6.3 in \u002Flibs\u002Fdeepagents (#1967)\nchore(quickjs): add some unit tests and disable tests from running (#1962)\nfeat(sdk): rename backend methods (#1907)\ntest(sdk): update evals for relational tool usage (#1940)\ntest(sdk): add fixture for summarization tests (#1627)\nfeat(sdk): add async subagent middleware for remote LangGraph servers (#1758)\nfix(sdk): simplify memory agent bench evals (#1873)\nfeat(sdk): update backend return types for ls, glob, grep (#1870)\nfix(sdk): use Sequence instead of list for subagents parameter (#1905)\nchore(sdk): increment version to 0.5 (#1908)\nfeat(sdk): update FileData and backend.read return type to support multimodal data (#1869)\nchore(sdk,cli): make `langchain-anthropic` and `langchain-google-genai` explicit deps (#1886)\nrelease(sdk): 0.4.11 (#1868)\nchore(sdk): make `langsmith` explicit evals dep (#1867)\nchore(sdk): add docs link to `LocalShellBackend` `virtual_mode` deprecation warning (#1866)\ntest(sdk): add `ChatBaseten` for evals (#1839)\nfeat(sdk): Add LangSmith integration metadata to deepagents (#1837)\nfix(sdk): strip leading slash from glob patterns before matching (#1846)\nfeat(sdk): add MemoryAgentBench evaluation suite (#1807)\nfix(sdk): accept all langsmith tracing env vars in evals conftest (#1847)\ntest(sdk): add 3 more evals for tool usage (longer trajectories) (#1840)\nchore(sdk): add evals `README`, ensure tracing (#1834)\nfeat(sdk): add `subagent_model` param to `create_deep_agent` (#1369)\nrefactor(","2026-04-01T15:51:41",{"id":233,"version":234,"summary_zh":235,"released_at":236},297238,"deepagents==0.5.0a2","Initial release\nrelease(deepagents): 0.5.0a2 (#2158)\nrelease(deepagents): 0.5.0a1 (#2157)\nfeat(sdk,cli): add package version metadata to traces (#2129)\nrefactor(sdk): rename tools for async subgents (#2103)\nfeat(sdk): add timestamp tracking to `AsyncSubAgentJob` (#2071)\nfix(sdk): overwrite `write` on LangSmithSandbox (#2097)\nfix(sdk): update middleware order (#2070)\nfeat(sdk): add system prompt snapshot test with sync and async subagents (#2072)\nchore(sdk): update langchain and langchain-anthropic deps (#2069)\nchore(sdk): merge async subagents and subagents into single top level param (#2064)\nfeat(sdk): port `LangSmithSandbox` from CLI to SDK (#1983)\nrefactor(evals): consolidate eval suite into standalone `libs\u002Fevals` (#2032)\nci: fix evals dataclass (#2034)\nci: suppress pytest streaming output in CI (#2026)\nfeat(sdk): add tau2 LICENSE (#2025)\nfeat(sdk): add tau2 airline eval harness (#1987)\ntest(sdk): gather `MemoryAgentBench` CI subset (#1993)\ntest(sdk): add curated external benchmark evals (#2018)\ntest(sdk): add evals for llm judge, tool selection, followup quality, and multi-turn memory (#1838)\nchore: promote `PLR6201` to safe-fix across all packages (#2000)\ntest(sdk): add wall-time benchmarks for `create_deep_agent` (#1944)\nchore(deps): bump pyasn1 from 0.6.2 to 0.6.3 in \u002Flibs\u002Fdeepagents (#1967)\nchore(quickjs): add some unit tests and disable tests from running (#1962)\nfeat(sdk): rename backend methods (#1907)\ntest(sdk): update evals for relational tool usage (#1940)\ntest(sdk): add fixture for summarization tests (#1627)\nfeat(sdk): add async subagent middleware for remote LangGraph servers (#1758)\nfix(sdk): simplify memory agent bench evals (#1873)\nfeat(sdk): update backend return types for ls, glob, grep (#1870)\nfix(sdk): use Sequence instead of list for subagents parameter (#1905)\nchore(sdk): increment version to 0.5 (#1908)\nfeat(sdk): update FileData and backend.read return type to support multimodal data (#1869)\nchore(sdk,cli): make `langchain-anthropic` and `langchain-google-genai` explicit deps (#1886)\nrelease(sdk): 0.4.11 (#1868)\nchore(sdk): make `langsmith` explicit evals dep (#1867)\nchore(sdk): add docs link to `LocalShellBackend` `virtual_mode` deprecation warning (#1866)\ntest(sdk): add `ChatBaseten` for evals (#1839)\nfeat(sdk): Add LangSmith integration metadata to deepagents (#1837)\nfix(sdk): strip leading slash from glob patterns before matching (#1846)\nfeat(sdk): add MemoryAgentBench evaluation suite (#1807)\nfix(sdk): accept all langsmith tracing env vars in evals conftest (#1847)\ntest(sdk): add 3 more evals for tool usage (longer trajectories) (#1840)\nchore(sdk): add evals `README`, ensure tracing (#1834)\nfeat(sdk): add `subagent_model` param to `create_deep_agent` (#1369)\nrefactor(sdk): extract model resolution helpers into `_models` module (#1825)\nrelease(sdk): prepare 0.4.10 patch release (#1795)\nfix(daytona): fix execute implementation (#1756)\ntest(sdk): add memory evals (#1775)\nrelease(deepagents): 0.4.9 (#1773)\nfeat(sdk): add prompt guidance for large tool results (#1763)\nrelease(deepagents): 0.4.8 (#1764)\ntest(sdk): bump langsmith-sdk for fixes experiments (#1752)\nfeat(sdk): add factory function for summarization tool middleware (#1749)\nfeat(sdk): add solve_rate metric to evals reporter (#1711)\nfeat(cli): `--acp` mode to run CLI agent as ACP server (#1297)\ntest(sdk): add metadata for experiments (#1652)\ntest(sdk): reuse model fixture in summarization evals (#1697)\nfeat(sdk): add baseten-hosted models to evals workflow (#1682)\nrelease(sdk): bump deepagents to 0.4.7 (#1688)\nfix(deepagents): restore full routed path in write\u002Fedit results (#1681)\nfeat(infra): add fireworks as eval provider (#1684)\nrelease(sdk): bump deepagents to 0.4.6 (#1677)\nchore: bump lock file (#1674)\nfix(sdk): extract text from all content blocks for large tool result eviction (#1673)\nfix(sdk): enforce path boundaries in `CompositeBackend` routing (#1659)\nrefactor(sdk): split eval assertions into correctness and efficiency tiers (#1670)\ntest(sdk): remove an ambiguous eval test (#1661)\nchore(sdk): add tool usage evals (#1637)\ntest(deepagents): fix one more file operation test (#1653)\ntest(sdk): update eval test to be a bit less ambiguous (#1643)\ndocs(sdk): middleware module docstring (#1620)\nchore(sdk): expand eval matrix and fix model name access for groq (#1623)\nrelease(deepagents): 0.4.5 (#1622)\nchore(sdk): expand eval model matrix to 28 models across 7 providers (#1617)\nchore: remove unnecessary `@pytest.mark.asyncio` decorators (#1616)\nchore(sdk): update evals to use chat model instance  (#1614)\nfix(sdk): make evals more robust across model output variations (#1608)\nfix(infra): streamline evals workflow and summary (#1602)\ndocs(sdk): improve docstrings around summarization conditions (#1597)\nchore(sdk): add json report to evals (#1594)\nchore(sdk,cli): switch `claude-sonnet-4-5-20250929` -> `claude-sonnet-4-6` (#1590)\nfeat(cli): `\u002Fcompact` command (#1579)\nchore: standardize makefile type targets\u002F add type CI c","2026-03-23T19:58:22",{"id":238,"version":239,"summary_zh":240,"released_at":241},297239,"deepagents==0.5.0a1","Initial release\nrelease(deepagents): 0.5.0a1 (#2157)\nfeat(sdk,cli): add package version metadata to traces (#2129)\nrefactor(sdk): rename tools for async subgents (#2103)\nfeat(sdk): add timestamp tracking to `AsyncSubAgentJob` (#2071)\nfix(sdk): overwrite `write` on LangSmithSandbox (#2097)\nfix(sdk): update middleware order (#2070)\nfeat(sdk): add system prompt snapshot test with sync and async subagents (#2072)\nchore(sdk): update langchain and langchain-anthropic deps (#2069)\nchore(sdk): merge async subagents and subagents into single top level param (#2064)\nfeat(sdk): port `LangSmithSandbox` from CLI to SDK (#1983)\nrefactor(evals): consolidate eval suite into standalone `libs\u002Fevals` (#2032)\nci: fix evals dataclass (#2034)\nci: suppress pytest streaming output in CI (#2026)\nfeat(sdk): add tau2 LICENSE (#2025)\nfeat(sdk): add tau2 airline eval harness (#1987)\ntest(sdk): gather `MemoryAgentBench` CI subset (#1993)\ntest(sdk): add curated external benchmark evals (#2018)\ntest(sdk): add evals for llm judge, tool selection, followup quality, and multi-turn memory (#1838)\nchore: promote `PLR6201` to safe-fix across all packages (#2000)\ntest(sdk): add wall-time benchmarks for `create_deep_agent` (#1944)\nchore(deps): bump pyasn1 from 0.6.2 to 0.6.3 in \u002Flibs\u002Fdeepagents (#1967)\nchore(quickjs): add some unit tests and disable tests from running (#1962)\nfeat(sdk): rename backend methods (#1907)\ntest(sdk): update evals for relational tool usage (#1940)\ntest(sdk): add fixture for summarization tests (#1627)\nfeat(sdk): add async subagent middleware for remote LangGraph servers (#1758)\nfix(sdk): simplify memory agent bench evals (#1873)\nfeat(sdk): update backend return types for ls, glob, grep (#1870)\nfix(sdk): use Sequence instead of list for subagents parameter (#1905)\nchore(sdk): increment version to 0.5 (#1908)\nfeat(sdk): update FileData and backend.read return type to support multimodal data (#1869)\nchore(sdk,cli): make `langchain-anthropic` and `langchain-google-genai` explicit deps (#1886)\nrelease(sdk): 0.4.11 (#1868)\nchore(sdk): make `langsmith` explicit evals dep (#1867)\nchore(sdk): add docs link to `LocalShellBackend` `virtual_mode` deprecation warning (#1866)\ntest(sdk): add `ChatBaseten` for evals (#1839)\nfeat(sdk): Add LangSmith integration metadata to deepagents (#1837)\nfix(sdk): strip leading slash from glob patterns before matching (#1846)\nfeat(sdk): add MemoryAgentBench evaluation suite (#1807)\nfix(sdk): accept all langsmith tracing env vars in evals conftest (#1847)\ntest(sdk): add 3 more evals for tool usage (longer trajectories) (#1840)\nchore(sdk): add evals `README`, ensure tracing (#1834)\nfeat(sdk): add `subagent_model` param to `create_deep_agent` (#1369)\nrefactor(sdk): extract model resolution helpers into `_models` module (#1825)\nrelease(sdk): prepare 0.4.10 patch release (#1795)\nfix(daytona): fix execute implementation (#1756)\ntest(sdk): add memory evals (#1775)\nrelease(deepagents): 0.4.9 (#1773)\nfeat(sdk): add prompt guidance for large tool results (#1763)\nrelease(deepagents): 0.4.8 (#1764)\ntest(sdk): bump langsmith-sdk for fixes experiments (#1752)\nfeat(sdk): add factory function for summarization tool middleware (#1749)\nfeat(sdk): add solve_rate metric to evals reporter (#1711)\nfeat(cli): `--acp` mode to run CLI agent as ACP server (#1297)\ntest(sdk): add metadata for experiments (#1652)\ntest(sdk): reuse model fixture in summarization evals (#1697)\nfeat(sdk): add baseten-hosted models to evals workflow (#1682)\nrelease(sdk): bump deepagents to 0.4.7 (#1688)\nfix(deepagents): restore full routed path in write\u002Fedit results (#1681)\nfeat(infra): add fireworks as eval provider (#1684)\nrelease(sdk): bump deepagents to 0.4.6 (#1677)\nchore: bump lock file (#1674)\nfix(sdk): extract text from all content blocks for large tool result eviction (#1673)\nfix(sdk): enforce path boundaries in `CompositeBackend` routing (#1659)\nrefactor(sdk): split eval assertions into correctness and efficiency tiers (#1670)\ntest(sdk): remove an ambiguous eval test (#1661)\nchore(sdk): add tool usage evals (#1637)\ntest(deepagents): fix one more file operation test (#1653)\ntest(sdk): update eval test to be a bit less ambiguous (#1643)\ndocs(sdk): middleware module docstring (#1620)\nchore(sdk): expand eval matrix and fix model name access for groq (#1623)\nrelease(deepagents): 0.4.5 (#1622)\nchore(sdk): expand eval model matrix to 28 models across 7 providers (#1617)\nchore: remove unnecessary `@pytest.mark.asyncio` decorators (#1616)\nchore(sdk): update evals to use chat model instance  (#1614)\nfix(sdk): make evals more robust across model output variations (#1608)\nfix(infra): streamline evals workflow and summary (#1602)\ndocs(sdk): improve docstrings around summarization conditions (#1597)\nchore(sdk): add json report to evals (#1594)\nchore(sdk,cli): switch `claude-sonnet-4-5-20250929` -> `claude-sonnet-4-6` (#1590)\nfeat(cli): `\u002Fcompact` command (#1579)\nchore: standardize makefile type targets\u002F add type CI check (#1581)\nci(sdk): skip some evals","2026-03-23T19:06:46"]