[{"data":1,"prerenderedAt":2839},["ShallowReactive",2],{"header-counts":3,"footer-counts":6,"alt-main-autogen":8,"alt-list-autogen":582},{"tools":4,"reviews":5},98,27,{"tools":4,"reviews":5,"playbooks":7,"news":5},23,{"id":9,"title":10,"alternatives":11,"api_compatible":15,"body":19,"category":544,"chinese_friendly":533,"cover":545,"description":546,"domestic":547,"extension":548,"faq":549,"free":550,"github":525,"languages":551,"lastVerified":553,"meta":554,"models":549,"navigation":550,"notSuitable":549,"opensource":550,"path":555,"pillar":556,"platforms":557,"priceTable":549,"pricing":560,"published":561,"relatedPlaybooks":549,"relatedReviews":549,"score":562,"self_host":547,"seo":565,"seoTitle":566,"slug":567,"sources":568,"stem":571,"suitable":549,"tagline":572,"tags":573,"updated":553,"verdict":580,"website":517,"__hash__":581},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Fautogen.md","AutoGen",[12,13,14],"agent\u002Fplatform\u002Fcrewai","agent\u002Fplatform\u002Flangflow","agent\u002Fplatform\u002Fdify",[16,17,18],"OpenAI","Anthropic","Google",{"type":20,"value":21,"toc":528},"minimark",[22,27,36,39,42,101,104,107,111,120,144,149,172,175,212,215,355,358,406,410,442,446,452,470,476,485,488,503,506,511],[23,24,26],"h2",{"id":25},"tldr","TL;DR",[28,29,30,31,35],"p",{},"AutoGen 是微软研究院开源的多 Agent 对话框架（MIT 协议），用 Python 代码定义 Agent 角色、对话流程和工具调用。核心是 ",[32,33,34],"code",{},"ConversableAgent"," + Group Chat 模式——多个 Agent 自动对话协作完成任务，支持代码执行、工具调用、人在回路（human-in-the-loop）。AutoGen 0.4+ 重构为事件驱动架构，性能和扩展性大幅提升。",[28,37,38],{},"适合：AI 研究者、需要精细控制 Agent 协作逻辑的高级开发者、多 Agent 实验项目。不适合：快速原型验证（用 CrewAI）、非技术用户（用 Dify \u002F Flowise）、需要 GUI 的场景、追求 API 稳定性的生产项目。",[23,40,41],{"id":41},"核心能力",[43,44,45,53,59,65,71,77,83,89,95],"ul",{},[46,47,48,52],"li",{},[49,50,51],"strong",{},"多 Agent 对话","：Group Chat 模式，多个 Agent 自动对话协作完成任务",[46,54,55,58],{},[49,56,57],{},"代码执行","：内置 Docker 代码执行器，Agent 可写代码 + 运行 + 调试",[46,60,61,64],{},[49,62,63],{},"工具调用","：自定义函数工具，Agent 自动选择和调用",[46,66,67,70],{},[49,68,69],{},"人在回路","：Human-in-the-loop 模式，关键决策需人工确认",[46,72,73,76],{},[49,74,75],{},"事件驱动架构","：0.4+ 重构为 async 事件驱动，支持分布式 Agent",[46,78,79,82],{},[49,80,81],{},"可定制 Agent","：system message \u002F 工具集 \u002F 终止条件全可自定义",[46,84,85,88],{},[49,86,87],{},"多模型支持","：OpenAI \u002F Claude \u002F Azure \u002F Ollama \u002F Gemini 等",[46,90,91,94],{},[49,92,93],{},"Agent 可组合","：嵌套 Agent、层级 Agent、条件路由",[46,96,97,100],{},[49,98,99],{},"可观测性","：集成 OpenTelemetry \u002F LangSmith 追踪 Agent 行为",[23,102,103],{"id":103},"价格",[28,105,106],{},"完全免费、MIT 开源、商用免费。运行成本仅来自所接入的 LLM API 调用费用。",[23,108,110],{"id":109},"体验与评测资料整理","体验与评测（资料整理）",[112,113,114],"blockquote",{},[28,115,116,117],{},"说明：本节基于官方文档与公开评测整理，非本站独立实测环境，具体数据请以官方为准。\n",[49,118,119],{},"亮点：",[43,121,122,125,128,131,138,141],{},[46,123,124],{},"Group Chat 模式让多 Agent 协作真正\"自动化\"，代码 reviewer + coder + tester 角色分工清晰",[46,126,127],{},"Docker 代码执行器安全隔离，Agent 写的代码在沙箱中运行",[46,129,130],{},"0.4+ 的事件驱动架构性能提升明显，异步并发能力强",[46,132,133,134,137],{},"自定义工具集成灵活，Python 函数加 ",[32,135,136],{},"@user_function"," 装饰器即可",[46,139,140],{},"微软背书，学术认可度高，论文引用多",[46,142,143],{},"人在回路模式适合需要人工把关的高风险场景",[28,145,146],{},[49,147,148],{},"踩坑：",[43,150,151,154,157,160,163,166,169],{},[46,152,153],{},"学习曲线非常陡峭，文档虽全但概念密集，新手容易劝退",[46,155,156],{},"0.2 → 0.4 API 大改，迁移成本高，网上旧教程大量失效",[46,158,159],{},"Agent 对话容易\"跑飞\"——无限循环 \u002F 偏离主题，需仔细设计终止条件",[46,161,162],{},"无 GUI，调试全靠日志和 print，排查多 Agent 对话链路费时",[46,164,165],{},"Token 消耗大——多 Agent 对话轮次多，API 费用叠加明显",[46,167,168],{},"错误处理不够健壮，LLM 返回格式异常时容易崩溃",[46,170,171],{},"社区活跃度不如 LangChain \u002F CrewAI，遇到问题搜索不到答案",[23,173,174],{"id":174},"上手",[176,177,178,184,191,197,203,209],"ol",{},[46,179,180,183],{},[32,181,182],{},"pip install autogen-agentchat autogen-ext","（0.4+ 新包名）",[46,185,186,187,190],{},"配置 LLM：设置 ",[32,188,189],{},"OPENAI_API_KEY"," 环境变量或代码内传入",[46,192,193,194],{},"定义 Agent：",[32,195,196],{},"AssistantAgent(name=\"coder\", system_message=\"...\", model_client=...)",[46,198,199,200],{},"创建 Group Chat：",[32,201,202],{},"RoundRobinGroupChat(agents=[agent1, agent2])",[46,204,205,206],{},"发起任务：",[32,207,208],{},"result = await team.run(task=\"写一个贪吃蛇游戏\")",[46,210,211],{},"进阶：加 Docker 代码执行器 + 自定义工具 + 人在回路",[23,213,214],{"id":214},"对比",[216,217,218,239],"table",{},[219,220,221],"thead",{},[222,223,224,228,230,233,236],"tr",{},[225,226,227],"th",{},"维度",[225,229,10],{},[225,231,232],{},"CrewAI",[225,234,235],{},"LangGraph",[225,237,238],{},"Dify",[240,241,242,258,274,291,306,322,338],"tbody",{},[222,243,244,248,251,253,255],{},[245,246,247],"td",{},"形态",[245,249,250],{},"Python 框架",[245,252,250],{},[245,254,250],{},[245,256,257],{},"可视化平台",[222,259,260,263,266,269,271],{},[245,261,262],{},"上手难度",[245,264,265],{},"高",[245,267,268],{},"中",[245,270,265],{},[245,272,273],{},"低",[222,275,276,279,282,285,288],{},[245,277,278],{},"多 Agent",[245,280,281],{},"✅ Group Chat",[245,283,284],{},"✅ Crew 角色",[245,286,287],{},"✅ 图编排",[245,289,290],{},"✅ 工作流",[222,292,293,295,298,301,303],{},[245,294,57],{},[245,296,297],{},"✅ Docker 沙箱",[245,299,300],{},"需自定义",[245,302,300],{},[245,304,305],{},"沙箱",[222,307,308,311,314,317,319],{},[245,309,310],{},"GUI",[245,312,313],{},"❌",[245,315,316],{},"❌（有 CrewAI Studio）",[245,318,313],{},[245,320,321],{},"✅",[222,323,324,327,330,333,336],{},[245,325,326],{},"API 稳定性",[245,328,329],{},"一般（大改过）",[245,331,332],{},"较好",[245,334,335],{},"好",[245,337,335],{},[222,339,340,343,346,349,352],{},[245,341,342],{},"适合场景",[245,344,345],{},"研究 \u002F 复杂协作",[245,347,348],{},"业务自动化",[245,350,351],{},"精确流程控制",[245,353,354],{},"应用构建",[23,356,357],{"id":357},"避坑",[43,359,360,370,376,382,388,394,400],{},[46,361,362,365,366,369],{},[49,363,364],{},"锁定版本","：0.2 和 0.4 API 不兼容，",[32,367,368],{},"pip install"," 时务必指定版本",[46,371,372,375],{},[49,373,374],{},"设计终止条件","：Group Chat 不设终止条件会无限对话，设 max_turns + 终止关键词",[46,377,378,381],{},[49,379,380],{},"Token 成本控制","：多 Agent 对话 token 消耗是单 Agent 的 3-5 倍，用 GPT-4 级模型注意费用",[46,383,384,387],{},[49,385,386],{},"代码执行器一定要用 Docker","：直接本地执行 Agent 生成的代码有安全风险",[46,389,390,393],{},[49,391,392],{},"别指望第一次跑通","：system message 调试 + 工具定义 + 终止条件需要反复迭代",[46,395,396,399],{},[49,397,398],{},"错误处理要完善","：LLM 返回异常格式时手动 catch + 重试",[46,401,402,405],{},[49,403,404],{},"不要用旧教程","：0.4+ 完全重构，网上大部分 AutoGen 教程是 0.2 版本的",[23,407,409],{"id":408},"适合-不适合","适合 \u002F 不适合",[43,411,412,415,418,421,424,427,430,433,436,439],{},[46,413,414],{},"✅ AI 研究者实验多 Agent 协作模式",[46,416,417],{},"✅ 需要代码级精细控制 Agent 行为",[46,419,420],{},"✅ 需要 Agent 代码执行 + 自动调试",[46,422,423],{},"✅ 人在回路的高风险决策场景",[46,425,426],{},"✅ 学术项目 \u002F 论文复现",[46,428,429],{},"❌ 快速原型验证（用 CrewAI，API 更简洁）",[46,431,432],{},"❌ 非技术用户（用 Dify \u002F Flowise）",[46,434,435],{},"❌ 追求 API 稳定性的生产项目（版本变动大）",[46,437,438],{},"❌ 需要可视化调试（无 GUI，全靠日志）",[46,440,441],{},"❌ 预算敏感场景（多 Agent 对话 token 消耗大）",[23,443,445],{"id":444},"faq","FAQ",[28,447,448,451],{},[49,449,450],{},"Q: AutoGen 和 CrewAI 怎么选？","\nA: AutoGen 更底层、更灵活，适合研究和复杂多 Agent 协作实验，但学习成本高。CrewAI API 更简洁直观，角色 + 任务 + 流程的概念更易理解，适合业务自动化场景。研究选 AutoGen，做产品选 CrewAI。",[28,453,454,457,458,461,462,465,466,469],{},[49,455,456],{},"Q: AutoGen 0.2 和 0.4 有什么区别？","\nA: 0.4 是完全重构版本——从同步改为异步事件驱动架构，包名从 ",[32,459,460],{},"pyautogen"," 改为 ",[32,463,464],{},"autogen-agentchat"," + ",[32,467,468],{},"autogen-ext","，API 全面更新。性能和扩展性大幅提升但旧代码无法直接迁移。新项目直接用 0.4+。",[28,471,472,475],{},[49,473,474],{},"Q: 多 Agent 对话成本高吗？","\nA: 高。多 Agent 每轮对话都消耗 token，一个任务 5-10 轮对话是常态，使用 GPT-4 级模型单个任务可能花费 $0.5-2。建议开发调试用便宜模型（GPT-4o-mini），生产再切高级模型。",[28,477,478,481,482,484],{},[49,479,480],{},"Q: 可以接入本地模型吗？","\nA: 可以。通过 ",[32,483,468],{}," 的 OpenAI 兼容客户端接入 Ollama \u002F vLLM \u002F LM Studio 的本地模型端点。但本地模型能力有限，复杂多 Agent 协作效果可能不如 GPT-4 \u002F Claude。",[23,486,487],{"id":487},"相关阅读",[28,489,490,494,495,494,499],{},[491,492,232],"a",{"href":493},"\u002Fagent\u002Fplatform\u002Fcrewai.html"," · ",[491,496,498],{"href":497},"\u002Fagent\u002Fplatform\u002Fflowise.html","Flowise",[491,500,502],{"href":501},"\u002Fcoding\u002Fapi\u002Flangfuse.html","Langfuse",[23,504,505],{"id":505},"来源",[112,507,508],{},[28,509,510],{},"本文的价格、版本号与性能数据均参考以下官方渠道整理，可能随时间变动，请以官方实时信息为准。",[43,512,513,521],{},[46,514,515],{},[491,516,520],{"href":517,"rel":518},"https:\u002F\u002Fmicrosoft.github.io\u002Fautogen",[519],"nofollow","官网",[46,522,523],{},[491,524,527],{"href":525,"rel":526},"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fautogen",[519],"GitHub",{"title":529,"searchDepth":530,"depth":530,"links":531},"",3,[532,534,535,536,537,538,539,540,541,542,543],{"id":25,"depth":533,"text":26},2,{"id":41,"depth":533,"text":41},{"id":103,"depth":533,"text":103},{"id":109,"depth":533,"text":110},{"id":174,"depth":533,"text":174},{"id":214,"depth":533,"text":214},{"id":357,"depth":533,"text":357},{"id":408,"depth":533,"text":409},{"id":444,"depth":533,"text":445},{"id":487,"depth":533,"text":487},{"id":505,"depth":533,"text":505},"platform","\u002Fimg\u002Ftools\u002Fautogen.webp","AutoGen 真实评测：微软开源的多 Agent 对话框架（MIT 协议），通过代码定义 Agent 角色和协作流程，支持多 Agent 对话、工具调用、代码执行。适合需要精细控制多 Agent 协作逻辑的开发者和研究团队。",false,"md",null,true,[552],"en","2026-07-30",{},"\u002Ftools\u002Fagent\u002Fplatform\u002Fautogen","agent",[558,559],"linux","docker","Free \u002F 开源（MIT）","2026-07-05",{"power":563,"ux":533,"price":564,"cn_support":533,"stability":530},4,5,{"title":10,"description":546},"AutoGen - 微软多 Agent 框架评测与使用 | AIHO","agent\u002Fplatform\u002Fautogen",[569,570],{"title":520,"url":517},{"title":527,"url":525},"tools\u002Fagent\u002Fplatform\u002Fautogen","微软开源多 Agent 对话框架，代码驱动 Agent 协作",[574,575,576,577,578,579],"agent-platform","multi-agent","framework","microsoft","python","opensource","需要代码级精细控制多 Agent 协作逻辑的研究者和高级开发者首选，微软背书 + 代码执行 + Group Chat 模式强大，但学习曲线陡峭、API 稳定性一般、无 GUI，不适合快速原型或非技术用户。","B_cns-NSiPcj5XmWQT-A-_3OiFT01qTSnk0-C6rpCWQ",[583,1094,1750],{"id":584,"title":232,"alternatives":585,"api_compatible":587,"body":589,"category":544,"chinese_friendly":533,"cover":1076,"description":1077,"domestic":547,"extension":548,"faq":549,"free":550,"github":1061,"languages":1078,"lastVerified":553,"meta":1079,"models":549,"navigation":550,"notSuitable":549,"opensource":550,"path":1080,"pillar":556,"platforms":1081,"priceTable":549,"pricing":1082,"published":561,"relatedPlaybooks":549,"relatedReviews":549,"score":1083,"self_host":547,"seo":1084,"seoTitle":1085,"slug":12,"sources":1086,"stem":1089,"suitable":549,"tagline":1090,"tags":1091,"updated":553,"verdict":1092,"website":1055,"__hash__":1093},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Fcrewai.md",[13,567,586],"agent\u002Fplatform\u002Fn8n",[16,17,18,588],"Moonshot Kimi",{"type":20,"value":590,"toc":1063},[591,593,596,599,601,656,658,706,711,713,719,742,746,769,771,808,810,924,926,974,976,1007,1009,1015,1021,1026,1032,1034,1043,1045,1049],[23,592,26],{"id":25},[28,594,595],{},"CrewAI 是开源多 Agent 协作框架（MIT 协议），用 Python 代码定义 Agent 角色（Role）、目标（Goal）、工具（Tools），组合成 Crew 执行任务序列。核心概念清晰——Agent 负责做事、Task 定义做什么、Crew 编排怎么协作。支持顺序 \u002F 层级 \u002F 自定义流程，内置 50+ 工具集成。CrewAI Enterprise 提供云端托管 + 可视化监控。",[28,597,598],{},"适合：需要多 Agent 自动化工作流的开发者、Python 技术栈团队、快速原型验证多 Agent 方案。不适合：非技术用户（用 Dify）、需要极复杂条件分支流程（用 LangGraph）、需要 GUI 可视化编排（用 Flowise）。",[23,600,41],{"id":41},[43,602,603,609,615,621,627,633,639,644,650],{},[46,604,605,608],{},[49,606,607],{},"角色定义","：Agent = Role + Goal + Backstory + Tools，角色人设驱动行为",[46,610,611,614],{},[49,612,613],{},"任务编排","：Task 定义具体任务 + 期望输出 + 分配 Agent",[46,616,617,620],{},[49,618,619],{},"Crew 编排","：顺序执行 \u002F 层级管理 \u002F 自定义流程三种模式",[46,622,623,626],{},[49,624,625],{},"工具集成","：内置 SerperDev \u002F Firecrawl \u002F FileRead \u002F WebScraper 等 50+ 工具",[46,628,629,632],{},[49,630,631],{},"流程控制","：支持任务间依赖、条件路由、输出传递",[46,634,635,638],{},[49,636,637],{},"记忆系统","：Short-term \u002F Long-term \u002F Entity Memory，Agent 可跨任务记忆",[46,640,641,643],{},[49,642,87],{},"：OpenAI \u002F Claude \u002F Gemini \u002F Ollama \u002F 任意 LiteLLM 兼容模型",[46,645,646,649],{},[49,647,648],{},"CrewAI Enterprise","：云端托管 + 可视化 Crew 监控 + 团队协作",[46,651,652,655],{},[49,653,654],{},"输出结构化","：支持 Pydantic 模型定义输出格式",[23,657,103],{"id":103},[216,659,660,672],{},[219,661,662],{},[222,663,664,667,669],{},[225,665,666],{},"方案",[225,668,103],{},[225,670,671],{},"核心功能",[240,673,674,685,695],{},[222,675,676,679,682],{},[245,677,678],{},"开源版",[245,680,681],{},"$0",[245,683,684],{},"完整框架，MIT 协议，本地运行",[222,686,687,689,692],{},[245,688,648],{},[245,690,691],{},"$49\u002F月起",[245,693,694],{},"云端托管 + 可视化监控 + API",[222,696,697,700,703],{},[245,698,699],{},"Enterprise+",[245,701,702],{},"联系销售",[245,704,705],{},"SSO \u002F 私有部署 \u002F 专属支持",[112,707,708],{},[28,709,710],{},"价格信息基于 2026-07 官网，可能调整。",[23,712,110],{"id":109},[112,714,715],{},[28,716,116,717],{},[49,718,119],{},[43,720,721,724,727,730,733,736,739],{},[46,722,723],{},"API 设计非常直观，Agent + Task + Crew 三件套 10 分钟上手",[46,725,726],{},"角色人设（Backstory）确实影响 Agent 行为，写好 backstory 效果提升明显",[46,728,729],{},"层级模式（hierarchical）下 Manager Agent 自动分配任务，适合复杂场景",[46,731,732],{},"内置工具丰富，SerperDev 搜索 + Firecrawl 爬虫开箱即用",[46,734,735],{},"Memory 系统让 Agent 跨任务保持上下文，长流程不丢信息",[46,737,738],{},"Pydantic 结构化输出对后续处理非常友好",[46,740,741],{},"CrewAI Enterprise 的可视化监控能看到每个 Agent 的思考过程",[28,743,744],{},[49,745,148],{},[43,747,748,751,754,757,760,763,766],{},[46,749,750],{},"Agent 偶尔\"不听话\"——偏离角色设定、重复执行、跳过任务",[46,752,753],{},"复杂流程的调试困难，错误信息不够清晰",[46,755,756],{},"Token 消耗不小——多 Agent + 多轮对话 + Memory 存储",[46,758,759],{},"开源版无 GUI，全靠日志调试，Enterprise 版才有可视化",[46,761,762],{},"版本迭代快，API 偶有 breaking changes",[46,764,765],{},"层级模式的 Manager Agent 判断不稳定，有时分配不合理",[46,767,768],{},"中文 system prompt 效果不如英文，建议角色定义用英文",[23,770,174],{"id":174},[176,772,773,779,784,789,795,801],{},[46,774,775,778],{},[32,776,777],{},"pip install crewai crewai-tools","（建议用 uv 管理虚拟环境）",[46,780,186,781,783],{},[32,782,189],{}," 或在代码中指定 model",[46,785,193,786],{},[32,787,788],{},"Agent(role=\"研究员\", goal=\"搜集信息\", tools=[search_tool])",[46,790,791,792],{},"定义 Task：",[32,793,794],{},"Task(description=\"调研XX趋势\", agent=researcher, expected_output=\"报告\")",[46,796,797,798],{},"组建 Crew：",[32,799,800],{},"Crew(agents=[researcher, writer], tasks=[task1, task2], process=Process.sequential)",[46,802,803,804,807],{},"启动：",[32,805,806],{},"result = crew.kickoff()","，查看结果 + 日志",[23,809,214],{"id":214},[216,811,812,826],{},[219,813,814],{},[222,815,816,818,820,822,824],{},[225,817,227],{},[225,819,232],{},[225,821,10],{},[225,823,235],{},[225,825,238],{},[240,827,828,841,858,870,882,897,910],{},[222,829,830,832,834,836,838],{},[245,831,262],{},[245,833,273],{},[245,835,265],{},[245,837,265],{},[245,839,840],{},"极低",[222,842,843,846,849,852,855],{},[245,844,845],{},"API 设计",[245,847,848],{},"优雅直观",[245,850,851],{},"底层灵活",[245,853,854],{},"图模型",[245,856,857],{},"可视化",[222,859,860,862,864,866,868],{},[245,861,278],{},[245,863,284],{},[245,865,281],{},[245,867,287],{},[245,869,290],{},[222,871,872,874,876,878,880],{},[245,873,57],{},[245,875,300],{},[245,877,297],{},[245,879,300],{},[245,881,305],{},[222,883,884,886,889,892,895],{},[245,885,637],{},[245,887,888],{},"✅ 内置",[245,890,891],{},"有限",[245,893,894],{},"需自建",[245,896,321],{},[222,898,899,901,904,906,908],{},[245,900,310],{},[245,902,903],{},"Enterprise 版",[245,905,313],{},[245,907,313],{},[245,909,321],{},[222,911,912,914,916,919,922],{},[245,913,342],{},[245,915,348],{},[245,917,918],{},"研究",[245,920,921],{},"精确流程",[245,923,354],{},[23,925,357],{"id":357},[43,927,928,934,940,946,952,958,968],{},[46,929,930,933],{},[49,931,932],{},"Backstory 认真写","：角色人设直接影响 Agent 行为质量，模糊描述 = 模糊行为",[46,935,936,939],{},[49,937,938],{},"expected_output 必填","：不定义预期输出，Agent 容易跑偏",[46,941,942,945],{},[49,943,944],{},"控制 Agent 数量","：3-5 个 Agent 最佳，超过 8 个协调成本急升",[46,947,948,951],{},[49,949,950],{},"Memory 按需开启","：Long-term Memory 会累积 token 消耗，简单任务关掉",[46,953,954,957],{},[49,955,956],{},"调试用 verbose=True","：开启详细日志看 Agent 思考过程，定位问题",[46,959,960,963,964,967],{},[49,961,962],{},"版本锁定","：",[32,965,966],{},"pip install crewai==x.x.x","，迭代快别用 latest",[46,969,970,973],{},[49,971,972],{},"层级模式慎用","：Manager Agent 不稳定，简单场景用 sequential 更可靠",[23,975,409],{"id":408},[43,977,978,981,984,987,990,993,995,998,1001,1004],{},[46,979,980],{},"✅ Python 开发者快速构建多 Agent 工作流",[46,982,983],{},"✅ 内容生产流水线（调研 → 写作 → 审核）",[46,985,986],{},"✅ 自动化研究 \u002F 数据收集 \u002F 报告生成",[46,988,989],{},"✅ 需要角色分工的协作场景",[46,991,992],{},"✅ 快速原型验证多 Agent 方案",[46,994,432],{},[46,996,997],{},"❌ 需要极复杂条件分支流程（用 LangGraph）",[46,999,1000],{},"❌ 需要精细控制 Agent 对话轮次（用 AutoGen）",[46,1002,1003],{},"❌ 需要免费 GUI 可视化监控（开源版无 GUI）",[46,1005,1006],{},"❌ 预算敏感的高频调用场景（多 Agent token 消耗大）",[23,1008,445],{"id":444},[28,1010,1011,1014],{},[49,1012,1013],{},"Q: CrewAI 和 AutoGen 怎么选？","\nA: CrewAI 上手更快，Agent + Task + Crew 概念直观，适合业务自动化和快速原型。AutoGen 更底层灵活，Group Chat + 代码执行 + 事件驱动适合研究和复杂协作。做产品选 CrewAI，做研究选 AutoGen。",[28,1016,1017,1020],{},[49,1018,1019],{},"Q: 开源版和 Enterprise 版差别大吗？","\nA: 开源版框架功能完整，能跑所有 Agent \u002F Task \u002F Crew。Enterprise 版主要多了云端托管（免运维）、可视化监控（看 Agent 思考过程）、团队协作和 API 服务。如果只是本地跑 Agent 工作流，开源版够用。",[28,1022,1023,1025],{},[49,1024,480],{},"\nA: 可以。CrewAI 基于 LiteLLM，支持 Ollama \u002F vLLM \u002F LM Studio 等本地模型。但本地模型能力有限，角色扮演和工具调用效果可能不如 GPT-4 \u002F Claude。建议开发用便宜模型，生产用高级模型。",[28,1027,1028,1031],{},[49,1029,1030],{},"Q: Agent 总是跑偏怎么办？","\nA: 三步排查：1）检查 Backstory 是否足够具体；2）确认 expected_output 定义清晰；3）开启 verbose=True 看思考过程定位偏移点。复杂任务拆成更小的 Task，每个 Task 目标单一明确。",[23,1033,487],{"id":487},[28,1035,1036,494,1039,494,1041],{},[491,1037,10],{"href":1038},"\u002Fagent\u002Fplatform\u002Fautogen.html",[491,1040,498],{"href":497},[491,1042,502],{"href":501},[23,1044,505],{"id":505},[112,1046,1047],{},[28,1048,510],{},[43,1050,1051,1057],{},[46,1052,1053],{},[491,1054,520],{"href":1055,"rel":1056},"https:\u002F\u002Fcrewai.com",[519],[46,1058,1059],{},[491,1060,527],{"href":1061,"rel":1062},"https:\u002F\u002Fgithub.com\u002FcrewAIInc\u002FcrewAI",[519],{"title":529,"searchDepth":530,"depth":530,"links":1064},[1065,1066,1067,1068,1069,1070,1071,1072,1073,1074,1075],{"id":25,"depth":533,"text":26},{"id":41,"depth":533,"text":41},{"id":103,"depth":533,"text":103},{"id":109,"depth":533,"text":110},{"id":174,"depth":533,"text":174},{"id":214,"depth":533,"text":214},{"id":357,"depth":533,"text":357},{"id":408,"depth":533,"text":409},{"id":444,"depth":533,"text":445},{"id":487,"depth":533,"text":487},{"id":505,"depth":533,"text":505},"\u002Fimg\u002Ftools\u002Fcrewai.webp","CrewAI 真实评测：开源多 Agent 协作框架（MIT 协议），用 Python 代码定义 Agent 角色、任务和协作流程。支持角色分工、任务编排、工具集成，适合需要构建多 Agent 自动化工作流的开发者和企业团队。",[552],{},"\u002Ftools\u002Fagent\u002Fplatform\u002Fcrewai",[558,559],"Free \u002F 开源（MIT）\u002F Enterprise",{"power":563,"ux":530,"price":564,"cn_support":533,"stability":530},{"title":232,"description":1077},"CrewAI - 多 Agent 协作框架评测与使用 | AIHO",[1087,1088],{"title":520,"url":1055},{"title":527,"url":1061},"tools\u002Fagent\u002Fplatform\u002Fcrewai","多 Agent 协作框架，Python 代码定义角色和任务",[574,575,576,578,579],"需要用 Python 快速构建多 Agent 自动化工作流的开发者首选，角色 + 任务 + Crew 的概念直观易学、API 设计优雅，但 Agent 对话可控性和稳定性不如 AutoGen，复杂流程编排需配合 LangGraph。","oi_8bSDkK_n2D-4mhKaXx1OMPYNIsP0E3tESVED9rwo",{"id":1095,"title":1096,"alternatives":1097,"api_compatible":1100,"body":1104,"category":544,"chinese_friendly":530,"cover":1684,"description":1685,"domestic":547,"extension":548,"faq":1686,"free":550,"github":1699,"languages":1700,"lastVerified":1702,"meta":1703,"models":549,"navigation":550,"notSuitable":549,"opensource":550,"path":1704,"pillar":556,"platforms":1705,"priceTable":1709,"pricing":1723,"published":1724,"relatedPlaybooks":1725,"relatedReviews":549,"score":1727,"self_host":550,"seo":1728,"seoTitle":1729,"slug":13,"sources":1730,"stem":1741,"suitable":549,"tagline":1742,"tags":1743,"updated":1734,"verdict":1748,"website":1733,"__hash__":1749},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Flangflow.md","Langflow",[586,1098,1099],"agent\u002Fprotocol\u002Fcomposio","agent\u002Fgeneral\u002Fopenmanus",[16,17,1101,1102,588,1103],"百度文心","阿里通义","DeepSeek",{"type":20,"value":1105,"toc":1672},[1106,1108,1111,1114,1116,1190,1192,1218,1223,1227,1231,1257,1261,1287,1289,1357,1360,1383,1385,1528,1530,1586,1588,1614,1616,1636,1638,1668],[23,1107,26],{"id":25},[28,1109,1110],{},"Langflow 是 2023 开源、2024 被 DataStax 收购的可视化 LangChain 画布。20,000+ GitHub stars、MIT 协议、Python 实现、与 Astra DB 向量库深度集成。差异点：拖拽式画布把 LangChain primitive 映射成节点 + RAG pipeline 原生组件 + 多 agent 工作流 + 节点可下钻到 Python 代码 + 自托管 \u002F Docker \u002F DataStax Astra 云托管 + Pinecone \u002F pgvector \u002F 主流向量库适配 + Visual GUI for building LangChain pipelines。Self-host 免费 \u002F Cloud Free + Paid ~$25\u002F月起。",[28,1112,1113],{},"适合：工程团队要可视化建 LangChain 流 + 不被 SaaS 锁住；合规 \u002F 数据驻留要求自托管；要 Astra DB 一站式 RAG；想用 LangChain 但讨厌纯代码协作。不适合：业务侧 + 非工程师（用 Dify \u002F Coze）；纯无代码偏好（Langflow 仍要懂 LangChain 概念）；轻量场景 + 不需要画布（直接写 LangChain 代码）。",[23,1115,41],{"id":41},[43,1117,1118,1124,1130,1136,1142,1148,1154,1160,1166,1172,1178,1184],{},[46,1119,1120,1123],{},[49,1121,1122],{},"可视化 LangChain 画布","：节点 = primitive，连线 = 数据流",[46,1125,1126,1129],{},[49,1127,1128],{},"RAG pipeline 原生组件","：Chunking \u002F Embedding \u002F VectorStore \u002F Retriever",[46,1131,1132,1135],{},[49,1133,1134],{},"多 agent 工作流","：编排多 agent 协作",[46,1137,1138,1141],{},[49,1139,1140],{},"Python 下钻","：任意节点可写 custom Python",[46,1143,1144,1147],{},[49,1145,1146],{},"向量库适配","：Astra DB \u002F Pinecone \u002F pgvector \u002F Weaviate \u002F Chroma",[46,1149,1150,1153],{},[49,1151,1152],{},"API 部署","：流程一键导出为 REST API",[46,1155,1156,1159],{},[49,1157,1158],{},"Real-time collaboration","：多用户同 project",[46,1161,1162,1165],{},[49,1163,1164],{},"版本控制","：内置 versioning + revert",[46,1167,1168,1171],{},[49,1169,1170],{},"数据可视化","：node output \u002F data flow 可视化调试",[46,1173,1174,1177],{},[49,1175,1176],{},"角色权限","：user auth + RBAC",[46,1179,1180,1183],{},[49,1181,1182],{},"Docker \u002F pip 安装","：5 分钟启动",[46,1185,1186,1189],{},[49,1187,1188],{},"Astra-hosted cloud","：DataStax 托管选项",[23,1191,103],{"id":103},[43,1193,1194,1200,1206,1212],{},[46,1195,1196,1199],{},[49,1197,1198],{},"Self-Host (OSS)","：$0；MIT 完全免费，自付 LLM API + 服务器",[46,1201,1202,1205],{},[49,1203,1204],{},"Cloud Free","：$0；DataStax Astra-hosted 小流量",[46,1207,1208,1211],{},[49,1209,1210],{},"Cloud Paid","：~$25\u002F月起；Astra DB + 更高额度 + 团队协作",[46,1213,1214,1217],{},[49,1215,1216],{},"Enterprise","：联系销售；SSO + audit + 私有部署 + SLA",[112,1219,1220],{},[28,1221,1222],{},"自托管最低成本：$5-10\u002F月 VPS + LLM API token。Cloud Paid $25\u002F月适合不想运维的小团队。",[23,1224,1226],{"id":1225},"实测rag-应用-内部工具搭建","实测（RAG 应用 + 内部工具搭建）",[28,1228,1229],{},[49,1230,119],{},[43,1232,1233,1236,1239,1242,1245,1248,1251,1254],{},[46,1234,1235],{},"画布直观，比纯写 LangChain 协作效率高 5x",[46,1237,1238],{},"节点下钻到 Python 让灵活度不被画布限制",[46,1240,1241],{},"Astra DB 集成省了配 vector store 时间",[46,1243,1244],{},"4.4\u002F5 用户评分（Propicked \u002F Tooliverse）",[46,1246,1247],{},"开源 + 自托管 + 数据驻留满足合规",[46,1249,1250],{},"多 agent 编排比裸 LangChain 调试容易",[46,1252,1253],{},"RAG pipeline 模板一键起 demo",[46,1255,1256],{},"与 DataStax 长期支持降低 abandon ware 风险",[28,1258,1259],{},[49,1260,148],{},[43,1262,1263,1266,1269,1272,1275,1278,1281,1284],{},[46,1264,1265],{},"稳定性偶发：复杂流大流量下 node 偶尔失联（SelectHub 反馈）",[46,1267,1268],{},"第三方 API 依赖：external API 失败时错误处理弱",[46,1270,1271],{},"production readiness 不算 mission-critical（要自加 observability）",[46,1273,1274],{},"LangChain 升级偶尔 break 旧 flow",[46,1276,1277],{},"文档对新组件滞后 1-2 月",[46,1279,1280],{},"中文 UI 不完整，业务侧用户上手陡",[46,1282,1283],{},"大型 flow（100+ 节点）画布卡顿",[46,1285,1286],{},"多人协作偶发同步冲突",[23,1288,174],{"id":174},[1290,1291,1295],"pre",{"className":1292,"code":1293,"language":1294,"meta":529,"style":529},"language-bash shiki shiki-themes github-light github-dark","# pip 安装\npip install langflow\nlangflow run  # http:\u002F\u002Flocalhost:7860\n\n# 或 Docker\ndocker run -p 7860:7860 langflowai\u002Flangflow:latest\n","bash",[32,1296,1297,1306,1319,1330,1335,1340],{"__ignoreMap":529},[1298,1299,1302],"span",{"class":1300,"line":1301},"line",1,[1298,1303,1305],{"class":1304},"sJ8bj","# pip 安装\n",[1298,1307,1308,1312,1316],{"class":1300,"line":533},[1298,1309,1311],{"class":1310},"sScJk","pip",[1298,1313,1315],{"class":1314},"sZZnC"," install",[1298,1317,1318],{"class":1314}," langflow\n",[1298,1320,1321,1324,1327],{"class":1300,"line":530},[1298,1322,1323],{"class":1310},"langflow",[1298,1325,1326],{"class":1314}," run",[1298,1328,1329],{"class":1304},"  # http:\u002F\u002Flocalhost:7860\n",[1298,1331,1332],{"class":1300,"line":563},[1298,1333,1334],{"emptyLinePlaceholder":550},"\n",[1298,1336,1337],{"class":1300,"line":564},[1298,1338,1339],{"class":1304},"# 或 Docker\n",[1298,1341,1343,1345,1347,1351,1354],{"class":1300,"line":1342},6,[1298,1344,559],{"class":1310},[1298,1346,1326],{"class":1314},[1298,1348,1350],{"class":1349},"sj4cs"," -p",[1298,1352,1353],{"class":1314}," 7860:7860",[1298,1355,1356],{"class":1314}," langflowai\u002Flangflow:latest\n",[28,1358,1359],{},"试 RAG 流：",[176,1361,1362,1365,1368,1371,1374,1377,1380],{},[46,1363,1364],{},"新建 flow → 选 Document QA 模板",[46,1366,1367],{},"Document Loader 节点 → 上传 PDF",[46,1369,1370],{},"Splitter → Embedder（OpenAI 或本地）",[46,1372,1373],{},"VectorStore（Astra \u002F Chroma）",[46,1375,1376],{},"Retriever + ChatOpenAI → Chat Output",[46,1378,1379],{},"部署为 API → 拿到 endpoint",[46,1381,1382],{},"复杂场景下钻节点写 Python 自定义",[23,1384,214],{"id":214},[216,1386,1387,1402],{},[219,1388,1389],{},[222,1390,1391,1393,1395,1397,1400],{},[225,1392,227],{},[225,1394,1096],{},[225,1396,238],{},[225,1398,1399],{},"n8n",[225,1401,498],{},[240,1403,1404,1421,1437,1452,1465,1481,1494,1511],{},[222,1405,1406,1409,1412,1415,1418],{},[245,1407,1408],{},"中心",[245,1410,1411],{},"LangChain primitive",[245,1413,1414],{},"LLMOps 全平台",[245,1416,1417],{},"通用 workflow",[245,1419,1420],{},"LangChain（JS）",[222,1422,1423,1426,1429,1432,1435],{},[245,1424,1425],{},"开源",[245,1427,1428],{},"✅ MIT",[245,1430,1431],{},"✅ AGPL",[245,1433,1434],{},"✅ Sustainable",[245,1436,1428],{},[222,1438,1439,1442,1445,1448,1450],{},[245,1440,1441],{},"自托管",[245,1443,1444],{},"✅ pip\u002FDocker",[245,1446,1447],{},"✅ Docker",[245,1449,1447],{},[245,1451,321],{},[222,1453,1454,1456,1459,1461,1463],{},[245,1455,857],{},[245,1457,1458],{},"✅ 旗舰",[245,1460,321],{},[245,1462,321],{},[245,1464,321],{},[222,1466,1467,1470,1473,1476,1479],{},[245,1468,1469],{},"代码下钻",[245,1471,1472],{},"✅ Python",[245,1474,1475],{},"部分",[245,1477,1478],{},"✅ JS",[245,1480,1478],{},[222,1482,1483,1486,1488,1490,1492],{},[245,1484,1485],{},"RAG 内置",[245,1487,321],{},[245,1489,321],{},[245,1491,1475],{},[245,1493,321],{},[222,1495,1496,1499,1502,1505,1508],{},[245,1497,1498],{},"起价（云）",[245,1500,1501],{},"$25\u002F月",[245,1503,1504],{},"$59\u002F月（Team）",[245,1506,1507],{},"自托管 $0",[245,1509,1510],{},"–",[222,1512,1513,1516,1519,1522,1525],{},[245,1514,1515],{},"适合",[245,1517,1518],{},"工程 + LangChain",[245,1520,1521],{},"业务 + LLMOps",[245,1523,1524],{},"通用自动化",[245,1526,1527],{},"JS 生态",[23,1529,357],{"id":357},[43,1531,1532,1538,1544,1550,1556,1562,1568,1574,1580],{},[46,1533,1534,1537],{},[49,1535,1536],{},"自托管推荐 Docker","：pip 版本依赖冲突难调",[46,1539,1540,1543],{},[49,1541,1542],{},"生产加 observability","：langsmith \u002F sentry \u002F prometheus 必装",[46,1545,1546,1549],{},[49,1547,1548],{},"Astra DB cloud free","：起步够用，付费版起步前算清成本",[46,1551,1552,1555],{},[49,1553,1554],{},"复杂 flow 拆模块","：100+ 节点画布卡顿，拆成子 flow",[46,1557,1558,1561],{},[49,1559,1560],{},"LangChain 版本 pin","：Langflow 升级前测试 flow 兼容性",[46,1563,1564,1567],{},[49,1565,1566],{},"第三方 API 加重试","：custom Python 节点写 retry + fallback",[46,1569,1570,1573],{},[49,1571,1572],{},"多人协作 lock","：同时编辑 flow 容易冲突，加 lock 流程",[46,1575,1576,1579],{},[49,1577,1578],{},"中文场景","：UI 英文为主，业务侧用户先培训",[46,1581,1582,1585],{},[49,1583,1584],{},"Self-host vs Cloud","：合规要 self-host，省心要 Cloud",[23,1587,409],{"id":408},[43,1589,1590,1593,1596,1599,1602,1605,1608,1611],{},[46,1591,1592],{},"✅ 工程团队要可视化建 LangChain 流",[46,1594,1595],{},"✅ 合规 \u002F 数据驻留要求自托管",[46,1597,1598],{},"✅ 要 Astra DB 一站式 RAG",[46,1600,1601],{},"✅ Python 团队 + 想画布 + 想下钻代码",[46,1603,1604],{},"❌ 业务侧 + 非工程师（用 Dify \u002F Coze）",[46,1606,1607],{},"❌ 纯无代码偏好",[46,1609,1610],{},"❌ 轻量场景 + 直接写 LangChain 更快",[46,1612,1613],{},"❌ JS 生态优先（用 Flowise）",[23,1615,487],{"id":487},[43,1617,1618,1624,1630],{},[46,1619,1620],{},[491,1621,1623],{"href":1622},"\u002Ftools\u002Fagent\u002Fplatform\u002Fn8n","n8n 评测",[46,1625,1626],{},[491,1627,1629],{"href":1628},"\u002Ftools\u002Fagent\u002Fprotocol\u002Fcomposio","Composio 评测",[46,1631,1632],{},[491,1633,1635],{"href":1634},"\u002Ftools\u002Fagent\u002Fgeneral\u002Fopenmanus","OpenManus 评测",[23,1637,505],{"id":505},[176,1639,1640,1647,1654,1661],{},[46,1641,1642,1643],{},"Langflow 官网 + 定价 ",[491,1644,1645],{"href":1645,"rel":1646},"https:\u002F\u002Fwww.langflow.org\u002Fpricing",[519],[46,1648,1649,1650],{},"AutomationAtlas — Langflow $25\u002Fmo 2026 Comparison ",[491,1651,1652],{"href":1652,"rel":1653},"https:\u002F\u002Fautomationatlas.io\u002Ftools\u002Flangflow",[519],[46,1655,1656,1657],{},"xpay — LangFlow 2026 Visual AI Agent Builder（MIT + DataStax）",[491,1658,1659],{"href":1659,"rel":1660},"https:\u002F\u002Fwww.xpay.sh\u002Fresources\u002Fagentic-frameworks\u002Flangflow",[519],[46,1662,1663,1664],{},"SelectHub — LangFlow Reviews 2026（稳定性 + 生产 readiness）",[491,1665,1666],{"href":1666,"rel":1667},"https:\u002F\u002Fwww.selecthub.com\u002Fp\u002Fai-agent-builder-software\u002Flangflow\u002F",[519],[1669,1670,1671],"style",{},"html pre.shiki code .sJ8bj, html code.shiki .sJ8bj{--shiki-default:#6A737D;--shiki-dark:#6A737D}html pre.shiki code .sScJk, html code.shiki .sScJk{--shiki-default:#6F42C1;--shiki-dark:#B392F0}html pre.shiki code .sZZnC, html code.shiki .sZZnC{--shiki-default:#032F62;--shiki-dark:#9ECBFF}html pre.shiki code .sj4cs, html code.shiki .sj4cs{--shiki-default:#005CC5;--shiki-dark:#79B8FF}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"title":529,"searchDepth":530,"depth":530,"links":1673},[1674,1675,1676,1677,1678,1679,1680,1681,1682,1683],{"id":25,"depth":533,"text":26},{"id":41,"depth":533,"text":41},{"id":103,"depth":533,"text":103},{"id":1225,"depth":533,"text":1226},{"id":174,"depth":533,"text":174},{"id":214,"depth":533,"text":214},{"id":357,"depth":533,"text":357},{"id":408,"depth":533,"text":409},{"id":487,"depth":533,"text":487},{"id":505,"depth":533,"text":505},"\u002Fimg\u002Ftools\u002Flangflow.webp","Langflow 真实评测：2023 开源项目，2024 被 DataStax 收购，与 Astra DB 向量数据库平台深度集成。20,000+ GitHub stars、MIT 协议。差异点：拖拽式可视化画布映射 LangChain 原语 + RAG pipeline 原生组件 + 多 agent 工作流 + Python + 自托管（pip \u002F Docker）+ DataStax Astra 托管云 + Pinecone \u002F pgvector \u002F 主流向量库适配。Cloud 免费档 + 约 $25\u002F月起。",[1687,1690,1693,1696],{"q":1688,"a":1689},"Langflow 和 Dify \u002F n8n \u002F Flowise 怎么选？","Langflow 强『可视化 LangChain 原语 + Python 代码可下钻 + 自托管 + Astra DB 集成』，工程导向。Dify 是 LLMOps 全平台（含 dataset \u002F app \u002F observability），业务侧更友好。n8n 是通用 workflow（非 LangChain 中心），70+ LangChain 节点是后加的。Flowise 是 Langflow 的同类竞品（JS 生态）。工程团队 + Python + 合规自托管 → Langflow；业务 + 完整 LLMOps → Dify；通用自动化 → n8n。",{"q":1691,"a":1692},"为什么被 DataStax 收购？","DataStax 是 Cassandra 商业公司 + Astra DB 向量数据库厂商。收购 Langflow 是为了把『可视化 LangChain builder』和『生产级 vector store』捆成一站式 RAG 解决方案。Langflow 主仓仍是 MIT 开源，但 cloud \u002F 企业版深度依赖 Astra。",{"q":1694,"a":1695},"可视化画布会不会限制灵活度？","Langflow 节点对应 LangChain primitive，可在任意节点下钻到 Python 代码 + 自定义。所以可视化层是『脚手架 + 协作工具』，不是『纯无代码黑盒』。复杂逻辑、custom tool、retriever 都能写代码扩展。",{"q":1697,"a":1698},"production readiness 如何？","用户反馈：原型 + 内部工具非常顺；大流量 \u002F 关键业务要自行加 observability \u002F 错误处理 \u002F 缓存。SelectHub 评测列出『稳定性偶发 + 第三方 API 依赖 + 非完全 production-ready』。生产部署建议 Astra-hosted cloud 或自托管 + 加 sentry \u002F langsmith \u002F prometheus。","https:\u002F\u002Fgithub.com\u002Flangflow-ai\u002Flangflow",[552,1701],"multi","2026-08-02",{},"\u002Ftools\u002Fagent\u002Fplatform\u002Flangflow",[1706,1707,559,1708],"self-host","cloud","web",[1710,1713,1716,1720],{"plan":1198,"price":681,"features":1711,"notes":1712},"MIT + pip\u002FDocker + 全部组件 + Pinecone\u002Fpgvector\u002FAstra 适配","自付 LLM API + 服务器",{"plan":1204,"price":681,"features":1714,"notes":1715},"DataStax 托管 + 小流量","试水",{"plan":1210,"price":1717,"features":1718,"notes":1719},"~$25\u002F月起","Astra DB + 更高额度 + 更多并发 + 团队协作","按使用量阶梯",{"plan":1216,"price":702,"features":1721,"notes":1722},"SSO + audit + 私有部署 + SLA + 数据驻留","合规 \u002F 大客户","Self-host 完全免费 MIT \u002F Cloud Free + 付费起 ~$25·月（Astra-hosted）","2026-06-19",[1726],"onboarding\u002Frag-app-workflow",{"power":563,"ux":564,"price":564,"cn_support":530,"stability":563},{"title":1096,"description":1685},"Langflow 评测 2026：可视化 AI 工作流构建工具，LangChain 低代码平台",[1731,1735,1737,1739],{"name":1732,"url":1733,"accessed":1734},"Langflow 官网","https:\u002F\u002Fwww.langflow.org","2026-06-24",{"name":1736,"url":1652,"accessed":1734},"AutomationAtlas — Langflow $25\u002Fmo 2026 Comparison",{"name":1738,"url":1659,"accessed":1734},"xpay — LangFlow 2026 Visual AI Agent Builder",{"name":1740,"url":1666,"accessed":1734},"SelectHub — LangFlow Reviews 2026","tools\u002Fagent\u002Fplatform\u002Flangflow","DataStax 出品的可视化 LangChain 画布——MIT 开源 + 20k+ stars + 自托管 + Astra DB 云托管",[579,1744,1745,1746,1747,1323],"visual-builder","langchain","rag","datastax","工程团队要可视化建 LangChain 流 + 不被 SaaS 锁住 + 合规要求自托管的最佳选择。要纯无代码 + 业务侧 → Dify \u002F Coze；要纯代码 → 直接写 LangChain \u002F LlamaIndex。","PNaJCu7eJDsH8LhAlO7CJ5JmktlB6hCq0vbBojxlTeM",{"id":1751,"title":238,"alternatives":1752,"api_compatible":1755,"body":1756,"category":544,"chinese_friendly":563,"cover":2782,"description":2783,"domestic":550,"extension":548,"faq":549,"free":550,"github":2048,"languages":2784,"lastVerified":1702,"meta":2787,"models":549,"navigation":550,"notSuitable":549,"opensource":550,"path":2788,"pillar":556,"platforms":2789,"priceTable":2792,"pricing":2809,"published":2810,"relatedPlaybooks":549,"relatedReviews":2811,"score":2816,"self_host":550,"seo":2817,"seoTitle":2818,"slug":14,"sources":2819,"stem":2831,"suitable":549,"tagline":2832,"tags":2833,"updated":1734,"verdict":2837,"website":2725,"__hash__":2838},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Fdify.md",[1753,1754,586,13],"agent\u002Fplatform\u002Fcoze","agent\u002Fplatform\u002Ffastgpt",[16,17,1101,1102,588,1103],{"type":20,"value":1757,"toc":2760},[1758,1760,1785,1790,1793,1798,1801,1869,1872,1876,1879,1893,1911,1915,1918,1929,1933,1941,1952,1956,1959,1962,1966,1975,2033,2040,2044,2052,2105,2108,2128,2132,2135,2193,2199,2203,2295,2298,2327,2330,2464,2482,2520,2523,2595,2597,2600,2620,2623,2652,2654,2716,2718,2749,2757],[23,1759,26],{"id":25},[1761,1762,1767,1773],"div",{"className":1763},[1764,1765,1766],"card","p-5","my-4",[28,1768,1769,1772],{},[49,1770,1771],{},"一句话："," Dify 是开源 LLMOps 平台的事实标准。GitHub 13 万 star、累计 100 万+ 生产 app（据 chatforest.com 2026 评测引用 Dify 官方数据），把\"可视化工作流编排 + RAG 知识库 + Agent + MCP 协议\"打包成一个 Docker Compose 能跑起来的东西。",[28,1774,1775,1776,1779,1780,1784],{},"最大价值是 ",[49,1777,1778],{},"完全开源 + 模型不挑食","——同一个工作流里同时调 OpenAI、Anthropic、Ollama 本地、DeepSeek、Qwen 都行。代价是部署比 ",[491,1781,1783],{"href":1782},"\u002Fagent\u002Fplatform\u002Fcoze.html","Coze"," 折腾，新手得读 1-2 小时文档。",[112,1786,1787],{},[28,1788,1789],{},"来源说明：本文基于 docs.dify.ai 官方文档、langgenius\u002Fdify GitHub 仓库、第三方评测（besthub.dev \u002F chatforest.com \u002F joshuaopolko.com \u002F zhihu 知名专栏）综合归纳。版本号会变，部署要求请以官方最新文档为准。",[23,1791,1792],{"id":1792},"核心特性",[1794,1795,1797],"h3",{"id":1796},"可视化工作流chatflow-workflow","可视化工作流（Chatflow + Workflow）",[28,1799,1800],{},"Dify 把 LLM 应用拆成两种\"应用类型\"：",[216,1802,1803,1815],{},[219,1804,1805],{},[222,1806,1807,1810,1812],{},[225,1808,1809],{},"类型",[225,1811,342],{},[225,1813,1814],{},"编排范式",[240,1816,1817,1830,1843,1856],{},[222,1818,1819,1824,1827],{},[245,1820,1821],{},[49,1822,1823],{},"Chatbot",[245,1825,1826],{},"简单对话机器人",[245,1828,1829],{},"prompt + tools",[222,1831,1832,1837,1840],{},[245,1833,1834],{},[49,1835,1836],{},"Agent",[245,1838,1839],{},"自主多步任务",[245,1841,1842],{},"ReAct \u002F Function Calling",[222,1844,1845,1850,1853],{},[245,1846,1847],{},[49,1848,1849],{},"Chatflow",[245,1851,1852],{},"对话型工作流（多轮 + 分支）",[245,1854,1855],{},"节点 DAG，带聊天上下文",[222,1857,1858,1863,1866],{},[245,1859,1860],{},[49,1861,1862],{},"Workflow",[245,1864,1865],{},"单次输入→输出（API 模式）",[245,1867,1868],{},"节点 DAG，无对话状态",[28,1870,1871],{},"节点类型覆盖：LLM、知识检索、HTTP 请求、代码执行（Python \u002F JS）、条件分支、迭代、变量聚合、参数提取、问题分类——满足\"用拖拽实现可观测的 LLM pipeline\"。",[1794,1873,1875],{"id":1874},"rag-知识库","RAG 知识库",[28,1877,1878],{},"内置完整 RAG 链路：",[176,1880,1881,1884,1887,1890],{},[46,1882,1883],{},"上传文档（PDF \u002F Word \u002F Markdown \u002F 网页）",[46,1885,1886],{},"自动分块 + embedding（可配置分段策略和 embedding 模型）",[46,1888,1889],{},"混合检索（向量 + 全文 + 重排）",[46,1891,1892],{},"引用溯源（回答末尾自动附原文片段）",[28,1894,1895,1896,1901,1902,1905,1906,1910],{},"注意：根据 ",[491,1897,1900],{"href":1898,"rel":1899},"https:\u002F\u002Fzhuanlan.zhihu.com\u002Fp\u002F1887141987838309480",[519],"知乎 LLM 实战笔记 2025-03 对比"," 的实测，Dify ",[49,1903,1904],{},"社区版默认是基础语义检索","，企业版才解锁多路召回 + 重排。RAG 极致精度场景仍推荐 ",[491,1907,1909],{"href":1908},"\u002Fagent\u002Fplatform\u002Ffastgpt.html","FastGPT","（实测准确率高 10+ 个百分点），Dify 胜在工作流而非纯 RAG。",[1794,1912,1914],{"id":1913},"模型生态40-提供商","模型生态：40+ 提供商",[28,1916,1917],{},"Dify 通过插件市场接入主流模型——OpenAI、Anthropic、Google Gemini、Azure、AWS Bedrock、Cohere、xAI、DeepSeek、Qwen、智谱、文心、豆包、月之暗面、Ollama、LM Studio、Replicate、Together AI、OpenRouter……几乎你能数出来的 LLM 提供商都在。",[28,1919,1920,1921,1923,1924,1928],{},"国产模型原生支持（不像 ",[491,1922,1909],{"href":1908}," 需要 ",[491,1925,1927],{"href":1926},"\u002Fcoding\u002Fapi\u002Fone-api.html","OneAPI"," 中转），是 Dify 在国内 toB 场景流行的关键。",[1794,1930,1932],{"id":1931},"mcp-协议支持","MCP 协议支持",[28,1934,1935,1936,1940],{},"Dify 较早接入了 ",[491,1937,1939],{"href":1938},"\u002Fwiki\u002Fmcp.html","MCP（Model Context Protocol）","，工作流可以直接调 MCP Server 暴露的 tools。意味着你可以让 Dify 工作流：",[43,1942,1943,1946,1949],{},[46,1944,1945],{},"通过 MCP 调本地 PostgreSQL \u002F SQLite",[46,1947,1948],{},"通过 MCP 调 GitHub \u002F Slack \u002F Linear",[46,1950,1951],{},"通过 MCP 调自家内部系统（写一个 MCP Server 即可）",[1794,1953,1955],{"id":1954},"api-first","API-first",[28,1957,1958],{},"每个 app 自动暴露 REST API，参数和返回结构自动生成 OpenAPI Schema。集成到自家产品里不需要写包装代码，给前端 \u002F 微信小程序 \u002F 飞书机器人调用都方便。",[23,1960,1961],{"id":1961},"价格与运行成本",[1794,1963,1965],{"id":1964},"云版difyai","云版（dify.ai）",[28,1967,1968,1969,1974],{},"根据 ",[491,1970,1973],{"href":1971,"rel":1972},"https:\u002F\u002Fwww.tooljunction.io\u002Fai-tools\u002Fdify-ai",[519],"tooljunction.io 2026 评测"," 引用的官方定价：",[216,1976,1977,1989],{},[219,1978,1979],{},[222,1980,1981,1984,1986],{},[225,1982,1983],{},"套餐",[225,1985,103],{},[225,1987,1988],{},"主要限制",[240,1990,1991,2002,2013,2024],{},[222,1992,1993,1996,1999],{},[245,1994,1995],{},"Sandbox",[245,1997,1998],{},"免费",[245,2000,2001],{},"200 次模型调用，1 app，5MB 知识库",[222,2003,2004,2007,2010],{},[245,2005,2006],{},"Professional",[245,2008,2009],{},"$59\u002F月起",[245,2011,2012],{},"5000 调用\u002F月，多 app，50MB 知识库",[222,2014,2015,2018,2021],{},[245,2016,2017],{},"Team",[245,2019,2020],{},"$159\u002F月起",[245,2022,2023],{},"团队协作、SSO",[222,2025,2026,2028,2030],{},[245,2027,1216],{},[245,2029,702],{},[245,2031,2032],{},"定制 SLA、私有云",[28,2034,2035,2036,2039],{},"注意：云版价格只是 Dify 平台费，",[49,2037,2038],{},"模型 API 费用另算","（自带 OpenAI \u002F Anthropic key）。",[1794,2041,2043],{"id":2042},"自托管推荐","自托管（推荐）",[28,2045,2046,2051],{},[491,2047,2050],{"href":2048,"rel":2049},"https:\u002F\u002Fgithub.com\u002Flanggenius\u002Fdify",[519],"官方 GitHub 仓库"," 提供 Docker Compose 部署，社区版完全免费可商用：",[1290,2053,2055],{"className":1292,"code":2054,"language":1294,"meta":529,"style":529},"git clone https:\u002F\u002Fgithub.com\u002Flanggenius\u002Fdify.git\ncd dify\u002Fdocker\ncp .env.example .env\ndocker compose up -d\n# 默认 http:\u002F\u002Flocalhost \u002F 端口可在 .env 调整\n",[32,2056,2057,2068,2076,2087,2100],{"__ignoreMap":529},[1298,2058,2059,2062,2065],{"class":1300,"line":1301},[1298,2060,2061],{"class":1310},"git",[1298,2063,2064],{"class":1314}," clone",[1298,2066,2067],{"class":1314}," https:\u002F\u002Fgithub.com\u002Flanggenius\u002Fdify.git\n",[1298,2069,2070,2073],{"class":1300,"line":533},[1298,2071,2072],{"class":1349},"cd",[1298,2074,2075],{"class":1314}," dify\u002Fdocker\n",[1298,2077,2078,2081,2084],{"class":1300,"line":530},[1298,2079,2080],{"class":1310},"cp",[1298,2082,2083],{"class":1314}," .env.example",[1298,2085,2086],{"class":1314}," .env\n",[1298,2088,2089,2091,2094,2097],{"class":1300,"line":563},[1298,2090,559],{"class":1310},[1298,2092,2093],{"class":1314}," compose",[1298,2095,2096],{"class":1314}," up",[1298,2098,2099],{"class":1349}," -d\n",[1298,2101,2102],{"class":1300,"line":564},[1298,2103,2104],{"class":1304},"# 默认 http:\u002F\u002Flocalhost \u002F 端口可在 .env 调整\n",[28,2106,2107],{},"硬件门槛（社区共识，非官方硬性要求）：",[43,2109,2110,2116,2122],{},[46,2111,2112,2115],{},[49,2113,2114],{},"最低","：2 核 4G，纯外接 API 模式",[46,2117,2118,2121],{},[49,2119,2120],{},"推荐","：4 核 8G + 至少 30GB 磁盘（向量数据 + 文件存储）",[46,2123,2124,2127],{},[49,2125,2126],{},"企业","：8 核 16G+，单机日活上千",[1794,2129,2131],{"id":2130},"真实-tco","真实 TCO",[28,2133,2134],{},"按一家中小团队 3 年场景估算（基于上面引用的多份评测交叉对比）：",[216,2136,2137,2149],{},[219,2138,2139],{},[222,2140,2141,2144,2147],{},[225,2142,2143],{},"成本项",[225,2145,2146],{},"云版 Professional",[225,2148,1441],{},[240,2150,2151,2161,2171,2182],{},[222,2152,2153,2156,2159],{},[245,2154,2155],{},"平台费",[245,2157,2158],{},"~$2,100（3 年）",[245,2160,681],{},[222,2162,2163,2166,2168],{},[245,2164,2165],{},"服务器",[245,2167,681],{},[245,2169,2170],{},"~$50\u002F月 × 36 = $1,800",[222,2172,2173,2176,2179],{},[245,2174,2175],{},"模型 API",[245,2177,2178],{},"与下同",[245,2180,2181],{},"与上同",[222,2183,2184,2187,2190],{},[245,2185,2186],{},"运维人力",[245,2188,2189],{},"0",[245,2191,2192],{},"约 0.2 人月",[28,2194,2195,2198],{},[49,2196,2197],{},"结论","：日活 \u003C 100 用云版省心；> 500 或数据敏感场景自托管 ROI 更好。",[23,2200,2202],{"id":2201},"上手-10-分钟","上手 10 分钟",[1290,2204,2206],{"className":1292,"code":2205,"language":1294,"meta":529,"style":529},"# 1. 自托管（社区版）\ngit clone https:\u002F\u002Fgithub.com\u002Flanggenius\u002Fdify.git\ncd dify\u002Fdocker\ncp .env.example .env\ndocker compose up -d\n\n# 2. 浏览器打开 http:\u002F\u002Flocalhost\n#    首次会让你创建 admin 账号\n\n# 3. 进入\"设置 → 模型供应商\"，配置 OpenAI \u002F 国产模型 API key\n\n# 4. 在主界面\"创建空白应用\"，选 Chatflow 或 Workflow\n# 5. 拖入\"开始 → LLM → 结束\"节点试一下基础 prompt\n# 6. 满意了点右上\"发布\"，自动生成 API endpoint\n",[32,2207,2208,2213,2221,2227,2235,2245,2249,2255,2261,2266,2272,2277,2283,2289],{"__ignoreMap":529},[1298,2209,2210],{"class":1300,"line":1301},[1298,2211,2212],{"class":1304},"# 1. 自托管（社区版）\n",[1298,2214,2215,2217,2219],{"class":1300,"line":533},[1298,2216,2061],{"class":1310},[1298,2218,2064],{"class":1314},[1298,2220,2067],{"class":1314},[1298,2222,2223,2225],{"class":1300,"line":530},[1298,2224,2072],{"class":1349},[1298,2226,2075],{"class":1314},[1298,2228,2229,2231,2233],{"class":1300,"line":563},[1298,2230,2080],{"class":1310},[1298,2232,2083],{"class":1314},[1298,2234,2086],{"class":1314},[1298,2236,2237,2239,2241,2243],{"class":1300,"line":564},[1298,2238,559],{"class":1310},[1298,2240,2093],{"class":1314},[1298,2242,2096],{"class":1314},[1298,2244,2099],{"class":1349},[1298,2246,2247],{"class":1300,"line":1342},[1298,2248,1334],{"emptyLinePlaceholder":550},[1298,2250,2252],{"class":1300,"line":2251},7,[1298,2253,2254],{"class":1304},"# 2. 浏览器打开 http:\u002F\u002Flocalhost\n",[1298,2256,2258],{"class":1300,"line":2257},8,[1298,2259,2260],{"class":1304},"#    首次会让你创建 admin 账号\n",[1298,2262,2264],{"class":1300,"line":2263},9,[1298,2265,1334],{"emptyLinePlaceholder":550},[1298,2267,2269],{"class":1300,"line":2268},10,[1298,2270,2271],{"class":1304},"# 3. 进入\"设置 → 模型供应商\"，配置 OpenAI \u002F 国产模型 API key\n",[1298,2273,2275],{"class":1300,"line":2274},11,[1298,2276,1334],{"emptyLinePlaceholder":550},[1298,2278,2280],{"class":1300,"line":2279},12,[1298,2281,2282],{"class":1304},"# 4. 在主界面\"创建空白应用\"，选 Chatflow 或 Workflow\n",[1298,2284,2286],{"class":1300,"line":2285},13,[1298,2287,2288],{"class":1304},"# 5. 拖入\"开始 → LLM → 结束\"节点试一下基础 prompt\n",[1298,2290,2292],{"class":1300,"line":2291},14,[1298,2293,2294],{"class":1304},"# 6. 满意了点右上\"发布\"，自动生成 API endpoint\n",[23,2296,2297],{"id":2297},"国内使用注意事项",[176,2299,2300,2306,2312,2318],{},[46,2301,2302,2305],{},[49,2303,2304],{},"云版 dify.ai 直连国内访问稳定但需要付款","——支持国际信用卡 \u002F Stripe",[46,2307,2308,2311],{},[49,2309,2310],{},"自托管 + 国产模型"," = 完全国内闭环，是 Dify 在国内最大优势",[46,2313,2314,2317],{},[49,2315,2316],{},"Docker 镜像拉取","：国内可能慢，建议配 Docker registry 镜像（阿里云 \u002F 网易）",[46,2319,2320,2323,2324,2326],{},[49,2321,2322],{},"数据合规","：完全自托管时，数据零外泄；某些金融 \u002F 政府客户因此从 ",[491,2325,1783],{"href":1782}," 迁到 Dify",[23,2328,2329],{"id":2329},"与同类怎么选",[216,2331,2332,2353],{},[219,2333,2334],{},[222,2335,2336,2338,2340,2344,2348],{},[225,2337,227],{},[225,2339,238],{},[225,2341,2342],{},[491,2343,1783],{"href":1782},[225,2345,2346],{},[491,2347,1909],{"href":1908},[225,2349,2350],{},[491,2351,1399],{"href":2352},"\u002Fagent\u002Fplatform\u002Fn8n.html",[240,2354,2355,2368,2381,2396,2410,2424,2438,2450],{},[222,2356,2357,2359,2361,2363,2365],{},[245,2358,1425],{},[245,2360,321],{},[245,2362,313],{},[245,2364,321],{},[245,2366,2367],{},"✅（fair-code）",[222,2369,2370,2373,2375,2377,2379],{},[245,2371,2372],{},"私有部署",[245,2374,321],{},[245,2376,313],{},[245,2378,321],{},[245,2380,321],{},[222,2382,2383,2385,2388,2391,2393],{},[245,2384,262],{},[245,2386,2387],{},"★★★☆☆",[245,2389,2390],{},"★★☆☆☆ 最简单",[245,2392,2387],{},[245,2394,2395],{},"★★★★☆",[222,2397,2398,2401,2404,2406,2408],{},[245,2399,2400],{},"工作流编排",[245,2402,2403],{},"★★★★★",[245,2405,2395],{},[245,2407,2387],{},[245,2409,2403],{},[222,2411,2412,2415,2417,2419,2421],{},[245,2413,2414],{},"RAG 精度",[245,2416,2395],{},[245,2418,2387],{},[245,2420,2403],{},[245,2422,2423],{},"★★☆☆☆",[222,2425,2426,2429,2431,2433,2436],{},[245,2427,2428],{},"模型生态",[245,2430,2403],{},[245,2432,2395],{},[245,2434,2435],{},"★★★☆☆（OneAPI 中转）",[245,2437,2395],{},[222,2439,2440,2442,2444,2446,2448],{},[245,2441,1578],{},[245,2443,2395],{},[245,2445,2403],{},[245,2447,2395],{},[245,2449,2387],{},[222,2451,2452,2455,2457,2460,2462],{},[245,2453,2454],{},"字节生态绑定",[245,2456,313],{},[245,2458,2459],{},"✅（飞书\u002F抖音深度集成）",[245,2461,313],{},[245,2463,313],{},[28,2465,2466,2469,2470,2475,2476,2481],{},[49,2467,2468],{},"怎么选","（基于 ",[491,2471,2474],{"href":2472,"rel":2473},"https:\u002F\u002Fwww.besthub.dev\u002Farticles\u002Fcoze-vs-dify-vs-fastgpt-which-ai-agent-platform-fits-your-needs-fa59cf97b798",[519],"BestHub 2025-07"," 和 ",[491,2477,2480],{"href":2478,"rel":2479},"https:\u002F\u002Fwww.cnblogs.com\u002Fuulucias\u002Fp\u002F19449008",[519],"博客园 2026-01"," 两份选型指南综合）：",[43,2483,2484,2490,2498,2505,2512],{},[46,2485,2486,2489],{},[49,2487,2488],{},"数据必须不出内网 + 工作流复杂"," → Dify",[46,2491,2492,2495,2496],{},[49,2493,2494],{},"个人 \u002F 小团队 \u002F 快速原型 + 字节生态"," → ",[491,2497,1783],{"href":1782},[46,2499,2500,2495,2503],{},[49,2501,2502],{},"核心场景就是企业知识库 QA",[491,2504,1909],{"href":1908},[46,2506,2507,2495,2510],{},[49,2508,2509],{},"重点是连接外部 SaaS（Slack \u002F Notion \u002F 数据库）",[491,2511,1399],{"href":2352},[46,2513,2514,2495,2517],{},[49,2515,2516],{},"要画图式表达 LangChain pipeline",[491,2518,1096],{"href":2519},"\u002Fagent\u002Fplatform\u002Flangflow.html",[23,2521,2522],{"id":2522},"避坑清单",[43,2524,2525,2531,2547,2558,2571,2577,2583,2589],{},[46,2526,2527,2530],{},[49,2528,2529],{},"社区版与企业版差距比想象大","：多路召回 \u002F 重排序 \u002F 单点登录 \u002F 审计日志都在企业版。社区版做生产前心里要有数。",[46,2532,2533,963,2539,2542,2543,2546],{},[49,2534,2535,2538],{},[32,2536,2537],{},".env"," 文件改完忘 restart",[32,2540,2541],{},"docker compose down && up -d","，不是 ",[32,2544,2545],{},"restart","——后者不重新加载 env。",[46,2548,2549,963,2552,2557],{},[49,2550,2551],{},"大版本升级会破坏数据库 schema",[491,2553,2556],{"href":2554,"rel":2555},"https:\u002F\u002Fdocs.dify.ai\u002Fzh-hans",[519],"官方升级文档"," 有详细 migration 步骤，跨大版本（如 0.x → 1.x）务必先备份 PostgreSQL 卷。生产环境强烈建议跑 staging 完整验证后再升。",[46,2559,2560,2563,2564,2566,2567,2570],{},[49,2561,2562],{},"RAG 文件大小社区版默认 15MB","：根据上述知乎实测，超过会失败。改 ",[32,2565,2537],{}," 的 ",[32,2568,2569],{},"UPLOAD_FILE_SIZE_LIMIT"," 并重启容器。",[46,2572,2573,2576],{},[49,2574,2575],{},"代码节点的 Sandbox 性能差","：内置代码执行节点跑在隔离容器里启动慢、内存小。生产高频用建议改成 HTTP 节点调外部服务。",[46,2578,2579,2582],{},[49,2580,2581],{},"工作流\"迭代节点\"循环上限","：默认 10 次，复杂 ReAct agent 容易撞天花板，需要在节点设置里调高。",[46,2584,2585,2588],{},[49,2586,2587],{},"Dify Plugin 系统是新东西","：1.0 后引入的 Plugin 体系替代了原来的 Tools\u002FModels 配置方式，老教程可能已过时——以最新官方文档为准。",[46,2590,2591,2594],{},[49,2592,2593],{},"国内 Docker 拉取镜像慢","：先配国内 registry，否则首次 pull 可能要 30+ 分钟。",[23,2596,409],{"id":408},[28,2598,2599],{},"✅ 适合：",[43,2601,2602,2605,2608,2611,2614,2617],{},[46,2603,2604],{},"中大型企业 LLM 中台建设",[46,2606,2607],{},"需要私有化部署（金融 \u002F 医疗 \u002F 政府）",[46,2609,2610],{},"想做\"AI 工作流即产品\"的开发团队",[46,2612,2613],{},"同时需要 RAG + Agent + Workflow 三件套",[46,2615,2616],{},"想用国产模型 + 国际模型混合编排",[46,2618,2619],{},"已经接受 Docker + 一定运维投入",[28,2621,2622],{},"❌ 不适合：",[43,2624,2625,2631,2637,2640,2646],{},[46,2626,2627,2628,2630],{},"纯个人玩家做对话机器人（",[491,2629,1783],{"href":1782}," 更快）",[46,2632,2633,2634,2636],{},"只想做企业知识库 QA（",[491,2635,1909],{"href":1908}," RAG 更专）",[46,2638,2639],{},"团队完全没运维能力（云版还行，自托管会踩坑）",[46,2641,2642,2643,2645],{},"需要深度对接字节飞书 \u002F 抖音（",[491,2644,1783],{"href":1782}," 原生）",[46,2647,2648,2649,2651],{},"工作流核心是连接 100+ SaaS（",[491,2650,1399],{"href":2352}," 节点更全）",[23,2653,487],{"id":487},[43,2655,2656,2668,2686,2705],{},[46,2657,2658,2659,2661,2662,2661,2664,2661,2666],{},"同类对比：",[491,2660,1783],{"href":1782}," \u002F ",[491,2663,1909],{"href":1908},[491,2665,1399],{"href":2352},[491,2667,1096],{"href":2519},[46,2669,2670,2671,2661,2675,2661,2679,2661,2682],{},"概念基础：",[491,2672,2674],{"href":2673},"\u002Fwiki\u002Fai-agent.html","AI Agent",[491,2676,2678],{"href":2677},"\u002Fwiki\u002Frag.html","RAG",[491,2680,2681],{"href":1938},"MCP",[491,2683,2685],{"href":2684},"\u002Fwiki\u002Ffunction-calling.html","Function Calling",[46,2687,2688,2689,2661,2693,2661,2697,2661,2701],{},"模型选型：",[491,2690,2692],{"href":2691},"\u002Fmodels\u002Fgpt-5.html","GPT-5",[491,2694,2696],{"href":2695},"\u002Fmodels\u002Fclaude-sonnet-4.html","Claude Sonnet 4",[491,2698,2700],{"href":2699},"\u002Fmodels\u002Fdeepseek-v3.html","DeepSeek-V3",[491,2702,2704],{"href":2703},"\u002Fmodels\u002Fglm-5.2.html","GLM-5.2",[46,2706,2707,2708,2661,2712],{},"进阶：",[491,2709,2711],{"href":2710},"\u002Fwiki\u002Ffine-tuning-vs-rag.html","Fine-tuning vs RAG",[491,2713,2715],{"href":2714},"\u002Fwiki\u002Fcontext-engineering.html","Context Engineering",[23,2717,505],{"id":505},[43,2719,2720,2727,2733,2739,2746],{},[46,2721,2722,2723],{},"官网：",[491,2724,2725],{"href":2725,"rel":2726},"https:\u002F\u002Fdify.ai",[519],[46,2728,2729,2730],{},"中文文档：",[491,2731,2554],{"href":2554,"rel":2732},[519],[46,2734,2735,2736],{},"GitHub：",[491,2737,2048],{"href":2048,"rel":2738},[519],[46,2740,2741,2742],{},"官方定价：",[491,2743,2744],{"href":2744,"rel":2745},"https:\u002F\u002Fdify.ai\u002Fpricing",[519],[46,2747,2748],{},"第三方评测：tooljunction.io \u002F chatforest.com \u002F besthub.dev \u002F joshuaopolko.com \u002F 知乎 LLM 实战笔记",[28,2750,2751,2752,2756],{},"本卡片由 AIHO 编辑部根据官方公开资料与第三方评测整理。所有事实点均标注来源；如发现版本号 \u002F 价格 \u002F 功能与最新官方信息不一致，请通过 ",[491,2753,2755],{"href":2754},"mailto:hello@aiho.net","反馈邮箱"," 反馈。",[1669,2758,2759],{},"html pre.shiki code .sScJk, html code.shiki .sScJk{--shiki-default:#6F42C1;--shiki-dark:#B392F0}html pre.shiki code .sZZnC, html code.shiki .sZZnC{--shiki-default:#032F62;--shiki-dark:#9ECBFF}html pre.shiki code .sj4cs, html code.shiki .sj4cs{--shiki-default:#005CC5;--shiki-dark:#79B8FF}html pre.shiki code .sJ8bj, html code.shiki .sJ8bj{--shiki-default:#6A737D;--shiki-dark:#6A737D}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"title":529,"searchDepth":530,"depth":530,"links":2761},[2762,2763,2770,2775,2776,2777,2778,2779,2780,2781],{"id":25,"depth":533,"text":26},{"id":1792,"depth":533,"text":1792,"children":2764},[2765,2766,2767,2768,2769],{"id":1796,"depth":530,"text":1797},{"id":1874,"depth":530,"text":1875},{"id":1913,"depth":530,"text":1914},{"id":1931,"depth":530,"text":1932},{"id":1954,"depth":530,"text":1955},{"id":1961,"depth":533,"text":1961,"children":2771},[2772,2773,2774],{"id":1964,"depth":530,"text":1965},{"id":2042,"depth":530,"text":2043},{"id":2130,"depth":530,"text":2131},{"id":2201,"depth":533,"text":2202},{"id":2297,"depth":533,"text":2297},{"id":2329,"depth":533,"text":2329},{"id":2522,"depth":533,"text":2522},{"id":408,"depth":533,"text":409},{"id":487,"depth":533,"text":487},{"id":505,"depth":533,"text":505},"\u002Fimg\u002Ftools\u002Fdify.webp","Dify 2026 真实评测：开源 LLMOps 与 AI Agent 平台，集工作流编排、RAG 知识库、Agent、MCP 和多模型接入于一体。本文对比 Coze、FastGPT、n8n，整理自托管部署、云版价格、适合团队和避坑建议。",[2785,552,2786],"zh","ja",{},"\u002Ftools\u002Fagent\u002Fplatform\u002Fdify",[2790,2791,558,559],"windows","macos",[2793,2797,2801,2805],{"plan":2794,"price":681,"features":2795,"notes":2796},"Self-hosted（开源版）","Docker 一键部署 + 全部核心功能（工作流 \u002F RAG \u002F Agent \u002F MCP）+ 接任意模型 API","私有部署 \u002F 完全免费 \u002F Apache 2.0",{"plan":2798,"price":681,"features":2799,"notes":2800},"Cloud Sandbox（免费云）","官方托管试水档，含基础调用配额","免运维 \u002F 试水 POC",{"plan":2802,"price":2009,"features":2803,"notes":2804},"Cloud Professional","更高调用额度 + 团队协作 + 商用支持","商用云首选",{"plan":2806,"price":2807,"features":2808,"notes":702},"Cloud Team \u002F Enterprise","Custom","更大配额 + SLA + 私有部署支持 + 合规","云版 SaaS（免费档 \u002F Professional $59\u002F月起） + 开源自托管完全免费","2026-06-18",[2812,2813,2814,2815],"coze-deep-review","coze-vs-dify","dify-deep-review","fastgpt-deep-review",{"power":564,"ux":563,"price":564,"cn_support":563,"stability":563},{"title":238,"description":2783},"Dify 评测 2026：开源 LLMOps 与 AI Agent 平台，自托管指南",[2820,2822,2824,2826,2828],{"title":2821,"url":2554},"Dify 官方文档（中文）",{"title":2823,"url":2048},"Dify GitHub",{"title":2825,"url":2744},"Dify 官方定价",{"title":2827,"url":2472},"Coze vs Dify vs FastGPT 选型",{"title":2829,"url":2830},"Dify Self-Hosted Guide 2026","https:\u002F\u002Fjoshuaopolko.com\u002Fdify-self-hosted-guide","tools\u002Fagent\u002Fplatform\u002Fdify","开源 LLMOps 平台，私有部署 Agent 首选",[574,579,1706,1746,2834,2835,2836],"workflow","llmops","mcp","想私有部署、想接全球任意模型，Dify 是答案。比 Coze 工程化、上手陡一点；比 FastGPT 工作流强、RAG 略弱。","q61l3oA5zdTKrp-66KGGh7wde1ZGUvjMulHa4GKaOXE",1785660639233]