[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-hatchet-dev--icepick":3,"tool-hatchet-dev--icepick":64},[4,17,27,35,43,56],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":16},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,3,"2026-04-05T11:01:52",[13,14,15],"开发框架","图像","Agent","ready",{"id":18,"name":19,"github_repo":20,"description_zh":21,"stars":22,"difficulty_score":23,"last_commit_at":24,"category_tags":25,"status":16},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",138956,2,"2026-04-05T11:33:21",[13,15,26],"语言模型",{"id":28,"name":29,"github_repo":30,"description_zh":31,"stars":32,"difficulty_score":23,"last_commit_at":33,"category_tags":34,"status":16},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",107662,"2026-04-03T11:11:01",[13,14,15],{"id":36,"name":37,"github_repo":38,"description_zh":39,"stars":40,"difficulty_score":23,"last_commit_at":41,"category_tags":42,"status":16},3704,"NextChat","ChatGPTNextWeb\u002FNextChat","NextChat 是一款轻量且极速的 AI 助手，旨在为用户提供流畅、跨平台的大模型交互体验。它完美解决了用户在多设备间切换时难以保持对话连续性，以及面对众多 AI 模型不知如何统一管理的痛点。无论是日常办公、学习辅助还是创意激发，NextChat 都能让用户随时随地通过网页、iOS、Android、Windows、MacOS 或 Linux 端无缝接入智能服务。\n\n这款工具非常适合普通用户、学生、职场人士以及需要私有化部署的企业团队使用。对于开发者而言，它也提供了便捷的自托管方案，支持一键部署到 Vercel 或 Zeabur 等平台。\n\nNextChat 的核心亮点在于其广泛的模型兼容性，原生支持 Claude、DeepSeek、GPT-4 及 Gemini Pro 等主流大模型，让用户在一个界面即可自由切换不同 AI 能力。此外，它还率先支持 MCP（Model Context Protocol）协议，增强了上下文处理能力。针对企业用户，NextChat 提供专业版解决方案，具备品牌定制、细粒度权限控制、内部知识库整合及安全审计等功能，满足公司对数据隐私和个性化管理的高标准要求。",87618,"2026-04-05T07:20:52",[13,26],{"id":44,"name":45,"github_repo":46,"description_zh":47,"stars":48,"difficulty_score":23,"last_commit_at":49,"category_tags":50,"status":16},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",84991,"2026-04-05T10:45:23",[14,51,52,53,15,54,26,13,55],"数据工具","视频","插件","其他","音频",{"id":57,"name":58,"github_repo":59,"description_zh":60,"stars":61,"difficulty_score":10,"last_commit_at":62,"category_tags":63,"status":16},3128,"ragflow","infiniflow\u002Fragflow","RAGFlow 是一款领先的开源检索增强生成（RAG）引擎，旨在为大语言模型构建更精准、可靠的上下文层。它巧妙地将前沿的 RAG 技术与智能体（Agent）能力相结合，不仅支持从各类文档中高效提取知识，还能让模型基于这些知识进行逻辑推理和任务执行。\n\n在大模型应用中，幻觉问题和知识滞后是常见痛点。RAGFlow 通过深度解析复杂文档结构（如表格、图表及混合排版），显著提升了信息检索的准确度，从而有效减少模型“胡编乱造”的现象，确保回答既有据可依又具备时效性。其内置的智能体机制更进一步，使系统不仅能回答问题，还能自主规划步骤解决复杂问题。\n\n这款工具特别适合开发者、企业技术团队以及 AI 研究人员使用。无论是希望快速搭建私有知识库问答系统，还是致力于探索大模型在垂直领域落地的创新者，都能从中受益。RAGFlow 提供了可视化的工作流编排界面和灵活的 API 接口，既降低了非算法背景用户的上手门槛，也满足了专业开发者对系统深度定制的需求。作为基于 Apache 2.0 协议开源的项目，它正成为连接通用大模型与行业专有知识之间的重要桥梁。",77062,"2026-04-04T04:44:48",[15,14,13,26,54],{"id":65,"github_repo":66,"name":67,"description_en":68,"description_zh":69,"ai_summary_zh":69,"readme_en":70,"readme_zh":71,"quickstart_zh":72,"use_case_zh":73,"hero_image_url":74,"owner_login":75,"owner_name":76,"owner_avatar_url":77,"owner_bio":78,"owner_company":79,"owner_location":79,"owner_email":80,"owner_twitter":79,"owner_website":81,"owner_url":82,"languages":83,"stars":100,"forks":101,"last_commit_at":102,"license":103,"difficulty_score":10,"env_os":104,"env_gpu":105,"env_ram":105,"env_deps":106,"category_tags":109,"github_topics":110,"view_count":120,"oss_zip_url":79,"oss_zip_packed_at":79,"status":16,"created_at":121,"updated_at":122,"faqs":123,"releases":129},1102,"hatchet-dev\u002Ficepick","icepick","Build agents that scale with a zero-cost abstraction.","Icepick 是一个基于 TypeScript 的 AI 代理开发库，专注于帮助开发者构建可扩展、可靠的 AI 系统。它通过简化任务调度、错误恢复和资源管理，让开发者能更高效地聚焦核心业务逻辑。传统方案常因故障恢复复杂、资源浪费或扩展性不足而受限，Icepick 通过内置的持久化任务队列（Hatchet）实现自动恢复，避免任务因失败或等待外部事件而浪费资源。\n\n适合需要构建分布式 AI 应用的开发者，尤其是那些希望将 AI 功能无缝集成到现有系统中的团队。其核心优势在于代码优先的设计理念，所有代理和工具均以函数形式定义，便于与业务逻辑深度融合。支持分布式部署，可跨多台机器调度任务，自动处理节点故障。同时提供丰富的配置选项，如重试策略、并发控制等，兼容多种云平台。Icepick 的轻量化抽象降低了开发门槛，让 AI 代理的构建和维护更高效。","\u003Cdiv align=\"center\">\n\u003Cpicture>\n  \u003Csource media=\"(prefers-color-scheme: dark)\" srcset=\".\u002Fstatic\u002Ficepick_dark.png\">\n  \u003Cimg width=\"200\" alt=\"Hatchet Logo\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fhatchet-dev_icepick_readme_d8394ae202bb.png\">\n\u003C\u002Fpicture>\n\u003C\u002Fa>\n\n### Icepick: A Typescript library for building AI agents that scale\n\n[![Docs](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fdocs-icepick.hatchet.run-E64327)](https:\u002F\u002Ficepick.hatchet.run) [![License: MIT](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-purple.svg)](https:\u002F\u002Fopensource.org\u002Flicenses\u002FMIT) [![NPM Downloads](https:\u002F\u002Fimg.shields.io\u002Fnpm\u002Fdm\u002F%40hatchet-dev%2Ficepick)](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002F@hatchet-dev\u002Ficepick)\n\n[![Discord](https:\u002F\u002Fimg.shields.io\u002Fdiscord\u002F1088927970518909068?style=social&logo=discord)](https:\u002F\u002Fhatchet.run\u002Fdiscord)\n[![Twitter](https:\u002F\u002Fimg.shields.io\u002Ftwitter\u002Furl\u002Fhttps\u002Ftwitter.com\u002Fhatchet-dev.svg?style=social&label=Follow%20%40hatchet-dev)](https:\u002F\u002Ftwitter.com\u002Fhatchet_dev)\n[![GitHub Repo stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fhatchet-dev\u002Ficepick?style=social)](https:\u002F\u002Fgithub.com\u002Fhatchet-dev\u002Ficepick)\n\n\u003C\u002Fdiv>\n\nIcepick is a simple Typescript library for building AI agents that are fault-tolerant and scalable. It handles the complexities of durable execution, queueing and scheduling, allowing you to focus on writing core business logic. [It is not a framework](#philosophy).\n\nEverything in Icepick is just a function that **you have written**, which makes it easy to integrate with your existing codebase and business logic. You can build agents that call tools, other agents, or any other functions you define:\n\nhttps:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002Fb28fc406-f501-4427-9574-e4c756b29dd4\n\n_Not sure if Icepick is a good fit? [Book office hours](https:\u002F\u002Fcal.com\u002Fteam\u002Fhatchet\u002Foffice-hours)_\n\n## Code Example\n\n```ts\nimport { icepick } from \"@hatchet-dev\u002Ficepick\";\nimport z from \"zod\";\nimport { myTool1, myTool2 } from \"@\u002Ftools\";\n\nconst MyAgentInput = z.object({\n  message: z.string(),\n});\n\nconst MyAgentOutput = z.object({\n  message: z.string(),\n});\n\nexport const myToolbox = icepick.toolbox({\n  tools: [myTool1, myTool2],\n});\n\nexport const myAgent = icepick.agent({\n  name: \"my-agent\",\n  executionTimeout: \"15m\",\n  inputSchema: MyAgentInput,\n  outputSchema: MyAgentOutput,\n  description: \"Description of what this agent does\",\n  fn: async (input, ctx) => {\n    const result = await myToolbox.pickAndRun({\n      prompt: input.message,\n    });\n\n    switch (result.name) {\n      case \"myTool1\":\n        return {\n          message: `Result: ${result.output}`,\n        };\n      case \"myTool2\":\n        return {\n          message: `Another result: ${result.output}`,\n        };\n      default:\n        return myToolbox.assertExhaustive(result);\n    }\n  },\n});\n```\n\n## Get started\n\nGetting started is as easy as two commands:\n\n```\npnpm i -g @hatchet-dev\u002Ficepick-cli\nicepick create first-agent\n```\n\nThis will prompt you to create a new Icepick project from a template to see an end to end example of Icepick in action.\n\nFor a full quickstart, check out our [documentation](https:\u002F\u002Ficepick.hatchet.run\u002Fquickstart).\n\n## Benefits\n\nIcepick is centered around the benefit of **durable execution**, which creates automatic checkpoints for agents so that they can easily recover from failure or wait for external events for a very long time without consuming resources. This is achieved by using a durable task queue called [Hatchet](https:\u002F\u002Fgithub.com\u002Fhatchet-dev\u002Fhatchet).\n\nAdditionally, Icepick agents are:\n\n- **💻 Code-first** - agents are defined as code and are designed to integrate with your business logic.\n- **🌐 Distributed** - all agents and tools run across a fleet of machines, where scheduling is handled gracefully by Icepick. When your underlying machine fails, Icepick takes care of rescheduling and resuming the agent on a different machine.\n- **⚙️ Configurable** - simple configuration for retries, rate limiting, concurrency control, and much more\n- **☁️ Runnable anywhere** - Icepick agents can run on any container-based platform (Hatchet, Railway, Fly.io, Porter, Kubernetes, AWS ECS, GCP Cloud Run)\n\n## Scalability\n\nIcepick is designed for scale: specifically, massive throughput and parallelism. Hatchet has run agentic workloads which spawn hundreds of thousands of tasks for a single execution, and runs billions of tasks per month.\n\n## Philosophy\n\nIcepick is not a framework. Agents and tools are simply functions that you have written. This means you can choose or build the best memory, knowledge, reasoning, or integrations. It does not impose any constraints on how you design your tools, call LLMs, or implement features like agent memory. Icepick is opinionated about the infrastructure layer of your agents, but not about the implementation details of your agents.\n\n## Documentation\n\n### Concepts\n\n- [**Overview**](https:\u002F\u002Ficepick.hatchet.run\u002Fconcepts\u002Foverview) - an overview of the Icepick execution model\n- [**Agents**](https:\u002F\u002Ficepick.hatchet.run\u002Fconcepts\u002Fagents) - agents are functions which call other tools and agents.\n- [**Tools**](https:\u002F\u002Ficepick.hatchet.run\u002Fconcepts\u002Ftools) - tools are functions that perform specific tasks and can be called by agents.\n- [**Toolbox**](https:\u002F\u002Ficepick.hatchet.run\u002Fconcepts\u002Ftoolbox) - a toolbox is a collection of tools with AI-powered selection capabilities.\n\n### API Reference\n\n- [`icepick.start`](https:\u002F\u002Ficepick.hatchet.run\u002Fapi-reference\u002Fstart)\n- [`icepick.agent`](https:\u002F\u002Ficepick.hatchet.run\u002Fapi-reference\u002Fagent)\n- [`icepick.tool`](https:\u002F\u002Ficepick.hatchet.run\u002Fapi-reference\u002Ftool)\n- [`icepick.toolbox`](https:\u002F\u002Ficepick.hatchet.run\u002Fapi-reference\u002Ftoolbox)\n\n### Use-Cases and Patterns\n\n- [Prompt chaining](https:\u002F\u002Ficepick.hatchet.run\u002Fpatterns\u002Fprompt-chaining)\n- [Routing](https:\u002F\u002Ficepick.hatchet.run\u002Fpatterns\u002Frouting)\n- [Parallelization](https:\u002F\u002Ficepick.hatchet.run\u002Fpatterns\u002Fparallelization)\n- [Evaluator-optimizer](https:\u002F\u002Ficepick.hatchet.run\u002Fpatterns\u002Fevaluator-optimizer)\n- [Multi-agent](https:\u002F\u002Ficepick.hatchet.run\u002Fpatterns\u002Fmulti-agent)\n- [Human-in-the-loop](https:\u002F\u002Ficepick.hatchet.run\u002Fpatterns\u002Fhuman-in-the-loop)\n\n## Comparison to Existing Tools\n\n### vs Frameworks (Mastra, Voltagent)\n\nIcepick is **not a framework**. It is not opinionated on how you structure your LLM calls, business logic, prompts, or context; we expect you to write these yourself (though Icepick does have a few utilities for tool-picking and bundles the AI SDK for calling LLMs).\n\nIcepick is designed to be extended and modified -- for example, you could build your own agent library on top of Icepick.\n\n### vs Temporal\n\nIcepick's execution model is most similar to [Temporal](https:\u002F\u002Fgithub.com\u002Ftemporalio\u002Ftemporal) with a simplified execution model and with more control for workflow scheduling:\n\n| Feature                                     | Icepick | Temporal |\n| ------------------------------------------- | ------- | -------- |\n| **Durable Execution**                       | ✅      | ✅       |\n| **Event Listeners within Workflows**        | ✅      | ✅       |\n| **Code-First Workflow Definitions**         | ✅      | ✅       |\n| **Cron Jobs**                               | ✅      | ✅       |\n| **One-Time Scheduling**                     | ✅      | ✅       |\n| **Flow Control**                            | ✅      | ✅       |\n| **Durable Sleep**                           | ✅      | ✅       |\n| **Global Rate Limits**                      | ✅      | ❌       |\n| **Event-Based Triggering**                  | ✅      | ❌       |\n| **Event Streaming**                         | ✅      | ❌       |\n| **DAG Support**                             | ✅      | ❌       |\n| **Priority Queues**                         | ✅      | ❌       |\n| **Sticky Assignment\u002FComplex Routing Logic** | ✅      | ❌       |\n\n## Agent Best Practices\n\nWhen writing agents with Icepick, it's useful to follow these rules:\n\n1. Agents should be **stateless reducers** with **no side effects**. They should not depend on external API calls, database calls, or local disk calls; their entire state should be determined by the results of their tool calls. See the [technical deep-dive](#technical-deep-dive) for more information.\n\n2. All quanta of work should be invoked as a task or a tool call.\n\n3. Treat **LLM calls as libraries** and **own your data lookups**: applications should not permit unconstrained agentic tool calling with data lookup. All tool calls should validate user permissions and separate data lookup from LLM calls for security reasons.\n\n## Contributions\n\nContributions are welcome! Please start a discussion in [Discord](https:\u002F\u002Fhatchet.run\u002Fdiscord) before tackling anything larger than a simple bug fix.\n\n## Technical Deep-Dive\n\nIcepick is a utility layer built on top of [Hatchet](https:\u002F\u002Fgithub.com\u002Fhatchet-dev\u002Fhatchet). It is built on the concept of a **durable task queue**, which means that every task which gets called in Hatchet is stored in a database. This is useful because tasks can easily be replayed and recover from failure, even when the underlying hardware crashes. Another way to look at it: Hatchet makes _distributed systems incredibly easy to deploy and maintain_.\n\nFor agents, this is particularly useful because they are extremely long-running, and thus need to be resilient to hardware failure. Agents also need to manage third-party rate limits and need concurrency control to prevent the system from getting overwhelmed.\n\nThe first rule of agents is that they should be _stateless reducers with no side effects_. To understand why, it's necessary to understand some concepts of durable execution. At its core, a function which executes durably stores an event log of all functions it has executed up to that point. Let's say an agent has called the tools `search_documents`, `get_document`, and is in the middle of processing `extract_from_document`. Its execution history looks like:\n\n```\nEvent log:\n-> Start search_documents\n-> Finish search_documents\n-> Start get_document\n-> Finish get_document\n-> Start extract_from_document...\n```\n\nNow, let's say that the machine which the agent is running on crashes during the last step. In order to recover from failure, Icepick will automatically replay all steps up to this point in the execution history:\n\n```\nEvent log:\n-> Start search_documents (replayed)\n-> Finish search_documents (replayed)\n-> Start get_document (replayed)\n-> Finish get_document (replayed)\n-> Start extract_from_document (replayed)\n-> (later) Finish extract_from_document\n```\n\nIn other words, the execution history is cached by Icepick, which allows the agent to recover gracefully from failure, instead of having to replay a bunch of work. Another way to think about it is that the agent automatically \"checkpoints\" its state.\n\nThis execution model is much more powerful when there's a requirement to wait for external systems, like a human reviewer or external event. Building a system that's resilient to failure becomes much more difficult, because if the agent starts from scratch, it may have lost the event which allowed execution to continue. In this model, the event automatically gets stored and replayed.\n\nBeyond Hatchet, there are two other points of inspiration for Icepick:\n\n- [12-factor agents](https:\u002F\u002Fgithub.com\u002Fhumanlayer\u002F12-factor-agents) -- this is the foundation for why Icepick advocates owning your control flow, context window, and prompts\n- Anthropic's [Building Effective Agents](https:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Fbuilding-effective-agents) -- we have ensured that each pattern documented in Anthropic's post are compatible with Icepick\n","\u003Cdiv align=\"center\">\n\u003Cpicture>\n  \u003Csource media=\"(prefers-color-scheme: dark)\" srcset=\".\u002Fstatic\u002Ficepick_dark.png\">\n  \u003Cimg width=\"200\" alt=\"Hatchet Logo\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fhatchet-dev_icepick_readme_d8394ae202bb.png\">\n\u003C\u002Fpicture>\n\u003C\u002Fa>\n\n### Icepick：用于构建可扩展AI代理的TypeScript库\n\n[![Docs](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fdocs-icepick.hatchet.run-E64327)](https:\u002F\u002Ficepick.hatchet.run) [![License: MIT](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-purple.svg)](https:\u002F\u002Fopensource.org\u002Flicenses\u002FMIT) [![NPM Downloads](https:\u002F\u002Fimg.shields.io\u002Fnpm\u002Fdm\u002F%40hatchet-dev%2Ficepick)](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002F@hatchet-dev\u002Ficepick)\n\n[![Discord](https:\u002F\u002Fimg.shields.io\u002Fdiscord\u002F1088927970518909068?style=social&logo=discord)](https:\u002F\u002Fhatchet.run\u002Fdiscord)\n[![Twitter](https:\u002F\u002Fimg.shields.io\u002Ftwitter\u002Furl\u002Fhttps\u002Ftwitter.com\u002Fhatchet-dev.svg?style=social&label=Follow%20%40hatchet-dev)](https:\u002F\u002Ftwitter.com\u002Fhatchet_dev)\n[![GitHub Repo stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fhatchet-dev\u002Ficepick?style=social)](https:\u002F\u002Fgithub.com\u002Fhatchet-dev\u002Ficepick)\n\nIcepick 是一个简单的 TypeScript 库，用于构建具有容错性和可扩展性的 AI 代理。它处理了持久执行、队列和调度的复杂性，让你能够专注于编写核心业务逻辑。[它不是一个框架](#philosophy)。\n\nIcepick 中的一切都是你所编写的功能，这使得它容易与现有代码库和业务逻辑集成。你可以构建调用工具、其他代理或你定义的任何其他函数的代理：\n\nhttps:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002Fb28fc406-f501-4427-9574-e4c756b29dd4\n\n不确定 Icepick 是否适合你？[预约办公室时间](https:\u002F\u002Fcal.com\u002Fteam\u002Fhatchet\u002Foffice-hours)\n\n## 代码示例\n\n```ts\nimport { icepick } from \"@hatchet-dev\u002Ficepick\";\nimport z from \"zod\";\nimport { myTool1, myTool2 } from \"@\u002Ftools\";\n\nconst MyAgentInput = z.object({\n  message: z.string(),\n});\n\nconst MyAgentOutput = z.object({\n  message: z.string(),\n});\n\nexport const myToolbox = icepick.toolbox({\n  tools: [myTool1, myTool2],\n});\n\nexport const myAgent = icepick.agent({\n  name: \"my-agent\",\n  executionTimeout: \"15m\",\n  inputSchema: MyAgentInput,\n  outputSchema: MyAgentOutput,\n  description: \"描述这个代理的功能\",\n  fn: async (input, ctx) => {\n    const result = await myToolbox.pickAndRun({\n      prompt: input.message,\n    });\n\n    switch (result.name) {\n      case \"myTool1\":\n        return {\n          message: `结果：${result.output}`,\n        };\n      case \"myTool2\":\n        return {\n          message: `另一个结果：${result.output}`,\n        };\n      default:\n        return myToolbox.assertExhaustive(result);\n    }\n  },\n});\n```\n\n## 开始使用\n\n开始使用只需两个命令：\n\n```\npnpm i -g @hatchet-dev\u002Ficepick-cli\nicepick create first-agent\n```\n\n这将提示你从模板创建一个新的 Icepick 项目，以查看 Icepick 的完整使用示例。\n\n如需完整的快速入门，请查看我们的[文档](https:\u002F\u002Ficepick.hatchet.run\u002Fquickstart)。\n\n## 优势\n\nIcepick 围绕**持久执行**这一优势展开，这为代理创建自动检查点，使其能够轻松从失败中恢复或等待外部事件很长时间而不会消耗资源。这是通过使用名为[Hatchet](https:\u002F\u002Fgithub.com\u002Fhatchet-dev\u002Fhatchet)的持久化任务队列实现的。\n\n此外，Icepick 代理具有以下特点：\n\n- **💻 以代码为中心** - 代理作为代码定义，设计为与你的业务逻辑集成。\n- **🌐 分布式** - 所有代理和工具在一组机器上运行，Icepick 优雅地处理调度。当底层机器失败时，Icepick 会负责重新调度和在另一台机器上恢复代理。\n- **⚙️ 可配置** - 简单的配置用于重试、速率限制、并发控制等。\n- **☁️ 可在任何地方运行** - Icepick 代理可以在任何容器化平台（Hatchet、Railway、Fly.io、Porter、Kubernetes、AWS ECS、GCP Cloud Run）上运行。\n\n## 可扩展性\n\nIcepick 专为大规模设计：具体而言，是巨大的吞吐量和并行性。Hatchet 已经运行了单次执行生成数万任务的代理工作负载，并且每月运行数十亿任务。\n\n## 哲学\n\nIcepick 不是一个框架。代理和工具只是你所编写的功能。这意味着你可以选择或构建最佳的记忆、知识、推理或集成。它不强制要求你如何设计工具、调用 LLM 或实现代理记忆等功能。Icepick 对代理的基础设施层持观点，但不对代理的实现细节持观点。\n\n## 文档\n\n### 概念\n\n- [**概览**](https:\u002F\u002Ficepick.hatchet.run\u002Fconcepts\u002Foverview) - Icepick 执行模型的概述\n- [**代理**](https:\u002F\u002Ficepick.hatchet.run\u002Fconcepts\u002Fagents) - 代理是调用其他工具和代理的功能\n- [**工具**](https:\u002F\u002Ficepick.hatchet.run\u002Fconcepts\u002Ftools) - 工具是执行特定任务的功能，可以被代理调用\n- [**工具箱**](https:\u002F\u002Ficepick.hatchet.run\u002Fconcepts\u002Ftoolbox) - 工具箱是包含工具的集合，具有 AI 驱动的选型能力\n\n### API 参考\n\n- [`icepick.start`](https:\u002F\u002Ficepick.hatchet.run\u002Fapi-reference\u002Fstart)\n- [`icepick.agent`](https:\u002F\u002Ficepick.hatchet.run\u002Fapi-reference\u002Fagent)\n- [`icepick.tool`](https:\u002F\u002Ficepick.hatchet.run\u002Fapi-reference\u002Ftool)\n- [`icepick.toolbox`](https:\u002F\u002Ficepick.hatchet.run\u002Fapi-reference\u002Ftoolbox)\n\n### 使用场景和模式\n\n- [提示链](https:\u002F\u002Ficepick.hatchet.run\u002Fpatterns\u002Fprompt-chaining)\n- [路由](https:\u002F\u002Ficepick.hatchet.run\u002Fpatterns\u002Frouting)\n- [并行化](https:\u002F\u002Ficepick.hatchet.run\u002Fpatterns\u002Fparallelization)\n- [评估器-优化器](https:\u002F\u002Ficepick.hatchet.run\u002Fpatterns\u002Fevaluator-optimizer)\n- [多代理](https:\u002F\u002Ficepick.hatchet.run\u002Fpatterns\u002Fmulti-agent)\n- [人机协作](https:\u002F\u002Ficepick.hatchet.run\u002Fpatterns\u002Fhuman-in-the-loop)\n\n## 与其他工具的比较\n\n### 与框架（Mastra、Voltagent）对比\n\nIcepick **不是框架**。它不强制要求你如何结构 LLM 调用、业务逻辑、提示或上下文；我们期望你自行编写这些内容（尽管 Icepick 有一些工具用于工具选择，并捆绑了调用 LLM 的 AI SDK）。Icepick 设计为可扩展和可修改——例如，你可以在 Icepick 上构建自己的代理库。\n\n### vs Temporal\n\nIcepick 的执行模型与 [Temporal](https:\u002F\u002Fgithub.com\u002Ftemporalio\u002Ftemporal) 最相似，具有简化了的执行模型，并提供了更精细的流程调度控制：\n\n| 特性                                     | Icepick | Temporal |\n| ------------------------------------------- | ------- | -------- |\n| **持久执行**                       | ✅      | ✅       |\n| **流程内事件监听器**        | ✅      | ✅       |\n| **基于代码的流程定义**         | ✅      | ✅       |\n| **定时任务**                               | ✅      | ✅       |\n| **单次调度**                     | ✅      | ✅       |\n| **流控**                            | ✅      | ✅       |\n| **持久休眠**                           | ✅      | ✅       |\n| **全局速率限制**                      | ✅      | ❌       |\n| **基于事件的触发**                  | ✅      | ❌       |\n| **事件流**                         | ✅      | ❌       |\n| **有向无环图支持**                             | ✅      | ❌       |\n| **优先队列**                         | ✅      | ❌       |\n| **粘性分配\u002F复杂路由逻辑** | ✅      | ❌       |\n\n## Agent 最佳实践\n\n在使用 Icepick 编写 Agent 时，建议遵循以下规则：\n\n1. Agent 应该是 **无状态的缩减器**，且 **无副作用**。它们不应依赖外部 API 调用、数据库调用或本地磁盘调用；其全部状态应由工具调用的结果决定。详见 [技术深度解析](#technical-deep-dive) 以获取更多信息。\n\n2. 所有工作量应作为任务或工具调用进行调用。\n\n3. 将 **LLM 调用视为库**，并 **负责数据查找**：应用程序不应允许无约束的代理工具调用与数据查找结合使用。所有工具调用应验证用户权限，并将数据查找与 LLM 调用分离以确保安全。\n\n## 贡献\n\n欢迎贡献！在处理任何大于简单 bug 修复的问题前，请先在 [Discord](https:\u002F\u002Fhatchet.run\u002Fdiscord) 开始讨论。\n\n## 技术深度解析\n\nIcepick 是建立在 [Hatchet](https:\u002F\u002Fgithub.com\u002Fhatchet-dev\u002Fhatchet) 之上的工具层。它基于 **持久任务队列** 的概念，这意味着 Hatchet 中调用的每个任务都会存储在数据库中。这很有用，因为任务可以轻松回放并从失败中恢复，即使底层硬件崩溃。另一种说法是：Hatchet 使 _分布式系统极其容易部署和维护_。\n\n对于 Agent 来说，这特别有用，因为它们运行时间极长，因此需要具备对硬件故障的容错能力。Agent 还需要管理第三方速率限制，并通过并发控制防止系统过载。\n\nAgent 的第一条规则是：它们应是 _无状态的缩减器，无副作用_。要理解原因，需要了解持久执行的一些概念。在核心层面，一个执行持久化的函数会存储其执行到目前为止的所有事件日志。假设一个 Agent 已调用工具 `search_documents`、`get_document`，并在处理 `extract_from_document` 时处于中间状态。其执行历史如下：\n\n```\n事件日志：\n-> 开始 search_documents\n-> 完成 search_documents\n-> 开始 get_document\n-> 完成 get_document\n-> 开始 extract_from_document...\n```\n\n现在，假设运行 Agent 的机器在最后一步崩溃。为了从失败中恢复，Icepick 会自动回放执行历史中的所有步骤：\n\n```\n事件日志：\n-> 开始 search_documents（回放）\n-> 完成 search_documents（回放）\n-> 开始 get_document（回放）\n-> 完成 get_document（回放）\n-> 开始 extract_from_document（回放）\n-> （稍后）完成 extract_from_document\n```\n\n换句话说，Icepick 缓存了执行历史，这使得 Agent 可以从失败中优雅恢复，而不是重新执行大量工作。另一种说法是：Agent 会自动“检查点”其状态。\n\n当需要等待外部系统（如人工审核或外部事件）时，这种执行模型会更加强大。构建容错系统变得更加困难，因为如果 Agent 从头开始，可能会丢失允许继续执行的事件。在这种模型中，事件会自动存储并回放。","# Icepick 快速上手指南\n\n## 环境准备\n- **系统要求**：Node.js 16+（推荐使用 LTS 版本）\n- **前置依赖**：\n  - TypeScript（用于类型声明）\n  - zod（用于输入输出校验）\n  - icepick-cli（命令行工具）\n  - Hatchet（底层任务队列，通过 icepick 自动管理）\n\n## 安装步骤\n1. 安装全局依赖（推荐使用 pnpm）：\n   ```bash\n   pnpm i -g @hatchet-dev\u002Ficepick-cli\n   ```\n   *若需使用国内镜像源，可先配置：*\n   ```bash\n   nrm use npmmirror\n   ```\n\n2. 创建项目：\n   ```bash\n   icepick create first-agent\n   ```\n   该命令会初始化项目模板并生成示例代码\n\n## 基本使用\n```ts\nimport { icepick } from \"@hatchet-dev\u002Ficepick\";\nimport z from \"zod\";\nimport { myTool1, myTool2 } from \"@\u002Ftools\";\n\nconst MyAgentInput = z.object({\n  message: z.string(),\n});\n\nconst MyAgentOutput = z.object({\n  message: z.string(),\n});\n\nexport const myToolbox = icepick.toolbox({\n  tools: [myTool1, myTool2],\n});\n\nexport const myAgent = icepick.agent({\n  name: \"my-agent\",\n  executionTimeout: \"15m\",\n  inputSchema: MyAgentInput,\n  outputSchema: MyAgentOutput,\n  description: \"Description of what this agent does\",\n  fn: async (input, ctx) => {\n    const result = await myToolbox.pickAndRun({\n      prompt: input.message,\n    });\n\n    switch (result.name) {\n      case \"myTool1\":\n        return {\n          message: `Result: ${result.output}`,\n        };\n      case \"myTool2\":\n        return {\n          message: `Another result: ${result.output}`,\n        };\n      default:\n        return myToolbox.assertExhaustive(result);\n    }\n  },\n});\n```\n\n*运行示例代理：*\n```bash\nicepick run my-agent\n```","电商平台的客服团队需要处理大量用户咨询，但现有系统无法应对高并发和故障恢复，导致服务不稳定。  \n\n### 没有 icepick 时  \n- 任务队列容易崩溃，故障时需手动重启并重传历史数据  \n- 高峰期响应延迟超过30秒，用户投诉率上升20%  \n- 每次新增功能需修改底层调度逻辑，开发效率低  \n- 服务器资源利用率不足，高峰期频繁扩容导致成本激增  \n- 故障恢复后无法自动续传未完成的任务，需人工介入  \n\n### 使用 icepick 后  \n- 任务自动持久化，故障后能从上次状态恢复，恢复时间缩短至5秒  \n- 分布式集群自动负载均衡，高峰期响应延迟稳定在1秒内  \n- 新功能直接通过代码定义，无需修改调度逻辑，开发效率提升40%  \n- 动态扩容策略智能调节资源，高峰期成本降低60%  \n- 支持并发执行和重试机制，任务失败率从15%降至3%  \n\n核心价值在于通过零成本抽象实现高可用、可扩展的AI代理系统，让团队专注于业务逻辑而非基础设施维护。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fhatchet-dev_icepick_abffe8fb.png","hatchet-dev","Hatchet","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fhatchet-dev_a5931e30.png","",null,"support@hatchet.run","https:\u002F\u002Fhatchet.run","https:\u002F\u002Fgithub.com\u002Fhatchet-dev",[84,88,92,96],{"name":85,"color":86,"percentage":87},"TypeScript","#3178c6",72.3,{"name":89,"color":90,"percentage":91},"Handlebars","#f7931e",25.8,{"name":93,"color":94,"percentage":95},"JavaScript","#f1e05a",1.8,{"name":97,"color":98,"percentage":99},"Dockerfile","#384d54",0.1,562,27,"2026-04-04T14:19:00","MIT","Linux, macOS, Windows","未说明",{"notes":107,"python":105,"dependencies":108},"需安装Node.js和npm，项目基于TypeScript构建，依赖Hatchet任务队列服务",[],[13,14,26,15],[111,112,113,114,115,116,117,118,119],"agentic-ai","ai","bun","llm","no-frameworks","nodejs","orchestration","scalability","typescript",4,"2026-03-27T02:49:30.150509","2026-04-06T06:52:13.476885",[124],{"id":125,"question_zh":126,"answer_zh":127,"source_url":128},4964,"项目是否还在维护？","icepick 项目目前没有主动开发新功能，但其基于 hatchet 的核心 SDK 仍具有参考价值。建议关注 hatchet 项目（https:\u002F\u002Fgithub.com\u002Fhatchet-dev\u002Fhatchet）以获取最新进展。","https:\u002F\u002Fgithub.com\u002Fhatchet-dev\u002Ficepick\u002Fissues\u002F8",[]]