[{"data":1,"prerenderedAt":3256},["ShallowReactive",2],{"header-counts":3,"footer-counts":6,"alt-main-manus":8,"alt-list-manus":1045},{"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":16,"category":952,"chinese_friendly":953,"cover":954,"description":955,"domestic":956,"extension":957,"faq":958,"free":959,"github":958,"languages":960,"lastVerified":963,"meta":964,"models":965,"navigation":959,"notSuitable":970,"opensource":956,"path":975,"pillar":976,"platforms":977,"priceTable":982,"pricing":998,"published":999,"relatedPlaybooks":1000,"relatedReviews":1002,"score":1006,"self_host":956,"seo":1008,"seoTitle":1009,"slug":1010,"sources":1011,"stem":1027,"suitable":1028,"tagline":1034,"tags":1035,"updated":1042,"verdict":1043,"website":425,"__hash__":1044},"tools\u002Ftools\u002Fagent\u002Fgeneral\u002Fmanus.md","Manus",[12,13,14],"agent\u002Fgeneral\u002Fgenspark","coding\u002Fagent\u002Fdevin","agent\u002Fplatform\u002Fdify",[],{"type":17,"value":18,"toc":930},"minimark",[19,24,66,72,75,141,144,149,157,162,169,173,181,253,258,262,269,273,281,286,303,306,309,312,319,322,328,389,395,402,407,413,417,464,467,507,510,685,690,707,712,732,735,801,805,808,825,828,849,852,902,905,922],[20,21,23],"h2",{"id":22},"tldr","TL;DR",[25,26,31,51],"div",{"className":27},[28,29,30],"card","p-5","my-4",[32,33,34,38,39,42,43,50],"p",{},[35,36,37],"strong",{},"一句话："," Butterfly Effect（中国创办、新加坡注册）2025-03-06 首发的通用 AI Agent，",[35,40,41],{},"邀请码一度被炒到 ¥5 万-10 万","（",[44,45,49],"a",{"href":46,"rel":47},"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FManus_%28AI_agent%29",[48],"nofollow","据维基百科引用 China Daily 报道","）。多 sub-agent 并行架构——浏览、数据分析、代码执行、写作各自专门子 agent，主 orchestrator 自动路由到 Claude \u002F Qwen \u002F 自研模型，结果合成交付。",[32,52,53,54,42,57,61,62,65],{},"2025-12 传被 ",[35,55,56],{},"Meta 以约 20-30 亿美元收购",[44,58,60],{"href":46,"rel":59},[48],"Reuters \u002F AP 报道","），目前仍独立运营。最大价值是 ",[35,63,64],{},"deep research 类任务的引用密度和质量明显超过 ChatGPT Deep Research","。",[67,68,69],"blockquote",{},[32,70,71],{},"来源说明：本文基于 manus.im 官方页面、Wikipedia \"Manus (AI agent)\" 条目、第三方评测（Pick Right \u002F HyzenPro \u002F Info-Tech Research Group \u002F 36Kr \u002F 世界经济论坛企业页）综合整理。Meta 收购案细节、产品路线图仍在变化中，请以最新官方公告为准。",[20,73,74],{"id":74},"背景与公司",[76,77,78,90,96,102,108,119,129,135],"ul",{},[79,80,81,84,85,89],"li",{},[35,82,83],{},"公司","：Butterfly Effect Pte. Ltd.（",[44,86,88],{"href":46,"rel":87},[48],"蝴蝶效应","），创始人 Xiao Hong（季逸超 Ji Yichao 为 Manus 联合创始人 + 首席科学家）",[79,91,92,95],{},[35,93,94],{},"公司位置","：办公室在北京 + 武汉 + 新加坡，目标市场北美 \u002F 日本 \u002F 韩国（不主打中国大陆）",[79,97,98,101],{},[35,99,100],{},"前作","：Monica，2023 年发布的浏览器扩展 AI 助手",[79,103,104,107],{},[35,105,106],{},"历史融资","：2024 年字节跳动曾出价 ~3000 万美元收购被拒（据 36Kr）",[79,109,110,113,114,118],{},[35,111,112],{},"Manus 启动","：2024-10 立项，灵感来自 ",[44,115,117],{"href":116},"\u002Fcoding\u002Fide\u002Fcursor.html","Cursor","；名字来自 MIT 拉丁校训 \"Mens et Manus\"（手与脑）",[79,120,121,124,125],{},[35,122,123],{},"首发数据","：2025-03-06 邀请制 beta，7 天内 200 万人候补，",[44,126,128],{"href":46,"rel":127},[48],"Demo 视频 20 小时百万播放",[79,130,131,134],{},[35,132,133],{},"营收","：2025-08 ARR 约 9000 万美元 → 2025-12 升至 1.25 亿美元",[79,136,137,140],{},[35,138,139],{},"Meta 收购","：2025-12-29 宣布，估值 20-30 亿美元，目前仍独立运营，但中国大陆已被屏蔽访问 + 关闭中文社交账号",[20,142,143],{"id":143},"核心特性",[145,146,148],"h3",{"id":147},"多-sub-agent-并行架构最大差异化","多 sub-agent 并行架构（最大差异化）",[32,150,151,156],{},[44,152,155],{"href":153,"rel":154},"https:\u002F\u002Fpick-right.com\u002Ftools\u002Fmanus-ai",[48],"Pick Right 2026-04 评测"," 描述的架构：",[67,158,159],{},[32,160,161],{},"\"Where most general-purpose AI products use one model end-to-end, Manus runs multiple specialized sub-agents in parallel: one handles web browsing, one handles data analysis, one handles code execution, one handles synthesis and writing.\"",[32,163,164,165,168],{},"主 orchestrator 给每一步选最合适的模型（Claude \u002F Qwen \u002F 自研），结果合成。这是 Manus 研究输出 ",[35,166,167],{},"引用密度更高、幻觉更少","的工程根源——分工 + 并行让每个 sub-agent 只做自己最擅长的部分。",[145,170,172],{"id":171},"通用-agent-模式","通用 Agent 模式",[32,174,175,180],{},[44,176,179],{"href":177,"rel":178},"https:\u002F\u002Fwww.infotech.com\u002Fresearch\u002Fassessing-manus-the-future-of-agentic-ai",[48],"Info-Tech 评测"," 列出的 GAIA 基准（General AI Assistants）：",[182,183,184,197],"table",{},[185,186,187],"thead",{},[188,189,190,194],"tr",{},[191,192,193],"th",{},"模型",[191,195,196],{},"GAIA 准确率",[198,199,200,213,221,229,237,245],"tbody",{},[188,201,202,208],{},[203,204,205],"td",{},[35,206,207],{},"Manus AI",[203,209,210],{},[35,211,212],{},">65%（SOTA）",[188,214,215,218],{},[203,216,217],{},"H2O.ai (h2oGPTe)",[203,219,220],{},"65%",[188,222,223,226],{},[203,224,225],{},"Google Langfun",[203,227,228],{},"49%",[188,230,231,234],{},[203,232,233],{},"Microsoft o1",[203,235,236],{},"38%",[188,238,239,242],{},[203,240,241],{},"OpenAI GPT-4o",[203,243,244],{},"32%",[188,246,247,250],{},[203,248,249],{},"OpenAI GPT-4 + Plugins",[203,251,252],{},"15-30%",[67,254,255],{},[32,256,257],{},"注：基准数据需第三方验证；Manus 官方公布数据，请审慎参考。",[145,259,261],{"id":260},"后台执行-长任务","后台执行 + 长任务",[32,263,264,265,268],{},"最特别的体验：任务下达后",[35,266,267],{},"可以关掉浏览器","，Manus 在云端继续跑（小时级），完成时通知。这种\"开着任务下班\"的模式是它 viral 的关键。",[145,270,272],{"id":271},"web-app-builder","Web App Builder",[32,274,275,276,280],{},"直接生成完整网站和应用，内置数据库 + Stripe 支付 + SEO。但 ",[44,277,279],{"href":153,"rel":278},[48],"Pick Right 评测"," 直白警告：",[67,282,283],{},[32,284,285],{},"\"Promising but buggy enough that I wouldn't ship to production from it yet.\"",[32,287,288,289,293,294,293,298,302],{},"复杂场景下出 bug 多，目前不建议生产部署，",[44,290,292],{"href":291},"\u002Fcoding\u002Fbuilder\u002Fbolt-new.html","Bolt.new"," \u002F ",[44,295,297],{"href":296},"\u002Fcoding\u002Fbuilder\u002Flovable.html","Lovable",[44,299,301],{"href":300},"\u002Fcoding\u002Fbuilder\u002Fv0.html","v0"," 仍是 production app building 的更稳选项。",[145,304,305],{"id":305},"桌面应用",[32,307,308],{},"提供 desktop app，能读本地文件 + 集成你的机器，不仅限于浏览器内任务。",[145,310,311],{"id":311},"多模型路由",[32,313,314,315,318],{},"Claude \u002F Qwen \u002F Manus 自研模型，",[35,316,317],{},"按任务步骤自动选","——这是 Manus 跟单一模型 agent 的根本差异。",[20,320,321],{"id":321},"价格与运行成本",[32,323,324,327],{},[44,325,279],{"href":153,"rel":326},[48]," 公开档位：",[182,329,330,343],{},[185,331,332],{},[188,333,334,337,340],{},[191,335,336],{},"档位",[191,338,339],{},"价格",[191,341,342],{},"关键点",[198,344,345,356,367,378],{},[188,346,347,350,353],{},[203,348,349],{},"Free",[203,351,352],{},"$0",[203,354,355],{},"每日有限 credits，够 1 个高强度任务\u002F天",[188,357,358,361,364],{},[203,359,360],{},"Pro",[203,362,363],{},"$20\u002F月",[203,365,366],{},"大多数付费用户落点；中等 credit + 多模型",[188,368,369,372,375],{},[203,370,371],{},"Plus",[203,373,374],{},"$50\u002F月",[203,376,377],{},"更高 credit + 优先队列，10+ 任务\u002F周",[188,379,380,383,386],{},[203,381,382],{},"Pro+ \u002F Team",[203,384,385],{},"最高 $200\u002F月",[203,387,388],{},"最大 credit + 团队空间（功能仍有限）",[32,390,391,394],{},[35,392,393],{},"credit 经济学","：每个动作消耗 credits（浏览、代码运行、模型调用都计费）。",[32,396,397,401],{},[44,398,400],{"href":153,"rel":399},[48],"Pick Right 真实使用反馈","：",[67,403,404],{},[32,405,406],{},"\"Credits run out faster than the pricing page suggests. Heavy users routinely buy credit packs on top of subscriptions.\"",[32,408,409,412],{},[35,410,411],{},"预算建议","：先用 Free 跑 3-5 个真实任务评估消耗速度，再决定档位。",[20,414,416],{"id":415},"上手-5-分钟","上手 5 分钟",[418,419,420,428,431,449,452,458,461],"ol",{},[79,421,422,423],{},"打开 ",[44,424,427],{"href":425,"rel":426},"https:\u002F\u002Fmanus.im",[48],"manus.im",[79,429,430],{},"账号注册（Google \u002F Apple OAuth 最快，国内手机号注册受限）",[79,432,433,434,437,438],{},"给一个完整任务描述（",[35,435,436],{},"关键","：不要碎片化指令，给全场景）\n",[439,440,445],"pre",{"className":441,"code":443,"language":444},[442],"language-text","\"调研欧洲前 10 大 EV 充电网络（覆盖率、价格、可靠性、充电速度），\n产出带引用的 Markdown 对比表\"\n","text",[446,447,443],"code",{"__ignoreMap":448},"",[79,450,451],{},"选模型路由（Auto 推荐）",[79,453,454,455,457],{},"提交后",[35,456,267],{},"——任务在云端跑",[79,459,460],{},"完成后邮件 \u002F 站内通知",[79,462,463],{},"看结果 \u002F 下载交付物（Markdown \u002F Excel \u002F Word \u002F 网页）",[20,465,466],{"id":466},"国内使用注意事项",[418,468,469,479,485,491],{},[79,470,471,401,474,478],{},[35,472,473],{},"大陆访问被屏蔽",[44,475,477],{"href":46,"rel":476},[48],"Wikipedia 引用 36Kr 报道","，Butterfly Effect 已关闭中文社交账号、阻断中国大陆访问，原 Alibaba Qwen 合作版\"中文版 Manus\"已搁置",[79,480,481,484],{},[35,482,483],{},"访问需稳定代理","：日韩 \u002F 美国节点",[79,486,487,490],{},[35,488,489],{},"账号 \u002F 支付","：海外信用卡（Visa \u002F MasterCard），第三方代付方案有限",[79,492,493,496,497,501,502,506],{},[35,494,495],{},"替代路径","：国内可考虑 ",[44,498,500],{"href":499},"\u002Fagent\u002Fgeneral\u002Fgenspark.html","Genspark"," \u002F Devv \u002F ",[44,503,505],{"href":504},"\u002Fagent\u002Fgeneral\u002Fflowith.html","元宝 Yuanbao"," \u002F 秘塔 Metaso 等",[20,508,509],{"id":509},"与同类怎么选",[182,511,512,537],{},[185,513,514],{},[188,515,516,519,521,527,531,534],{},[191,517,518],{},"维度",[191,520,10],{},[191,522,523],{},[44,524,526],{"href":525},"\u002Fcoding\u002Fagent\u002Fdevin.html","Devin",[191,528,529],{},[44,530,500],{"href":499},[191,532,533],{},"ChatGPT Deep Research",[191,535,536],{},"GenAgent",[198,538,539,559,579,598,613,630,647,665],{},[188,540,541,544,547,550,553,556],{},[203,542,543],{},"核心定位",[203,545,546],{},"通用 Agent",[203,548,549],{},"AI 程序员",[203,551,552],{},"AI 搜索 + Agent",[203,554,555],{},"LLM Deep Research",[203,557,558],{},"通用",[188,560,561,564,567,570,573,576],{},[203,562,563],{},"架构",[203,565,566],{},"多 sub-agent 并行",[203,568,569],{},"单 Agent + 沙盒",[203,571,572],{},"多模型",[203,574,575],{},"单模型",[203,577,578],{},"—",[188,580,581,584,587,590,593,596],{},[203,582,583],{},"长任务",[203,585,586],{},"★★★★★ 小时级",[203,588,589],{},"★★★★★",[203,591,592],{},"★★★☆☆",[203,594,595],{},"★★★★☆",[203,597,592],{},[188,599,600,603,605,607,609,611],{},[203,601,602],{},"引用密度",[203,604,589],{},[203,606,592],{},[203,608,595],{},[203,610,592],{},[203,612,592],{},[188,614,615,618,621,624,626,628],{},[203,616,617],{},"App Builder",[203,619,620],{},"⚠️ 有但 buggy",[203,622,623],{},"❌",[203,625,623],{},[203,627,623],{},[203,629,623],{},[188,631,632,635,638,641,643,645],{},[203,633,634],{},"中文",[203,636,637],{},"⚠️ 大陆屏蔽",[203,639,640],{},"⚠️",[203,642,589],{},[203,644,640],{},[203,646,595],{},[188,648,649,651,654,657,660,663],{},[203,650,339],{},[203,652,653],{},"$20-$200",[203,655,656],{},"$500\u002F月",[203,658,659],{},"$24.99\u002F月",[203,661,662],{},"随 ChatGPT Plus",[203,664,578],{},[188,666,667,670,673,676,679,682],{},[203,668,669],{},"适合场景",[203,671,672],{},"research \u002F 数据分析",[203,674,675],{},"写代码 \u002F 修 bug",[203,677,678],{},"信息检索",[203,680,681],{},"单次深度研究",[203,683,684],{},"综合",[32,686,687,401],{},[35,688,689],{},"选 Manus 如果你",[76,691,692,698,701,704],{},[79,693,694,695],{},"重视 research \u002F 数据分析任务的 ",[35,696,697],{},"引用密度和结构化输出",[79,699,700],{},"想试\"开任务下班、明早看结果\"的工作流",[79,702,703],{},"海外 \u002F 能解决账号网络问题",[79,705,706],{},"预算 $20-$50\u002F月，重度用户",[32,708,709,401],{},[35,710,711],{},"别选 Manus 如果你",[76,713,714,723,726,729],{},[79,715,716,717,293,719,722],{},"国内裸用（",[44,718,500],{"href":499},[44,720,721],{"href":504},"Yuanbao"," 更顺）",[79,724,725],{},"想生产部署 App（Web App Builder bug 多）",[79,727,728],{},"团队协作场景（功能不完善）",[79,730,731],{},"预算 \u003C $20\u002F月（Free 档够评估，付费档不一定划算）",[20,733,734],{"id":734},"避坑清单",[76,736,737,743,753,759,765,777,783,789],{},[79,738,739,742],{},[35,740,741],{},"大陆访问已被官方屏蔽","：2025 年起阻断中国大陆访问，原\"中文版 Manus\"项目搁置",[79,744,745,401,748,752],{},[35,746,747],{},"credit 烧得比官方页面暗示的快",[44,749,751],{"href":153,"rel":750},[48],"Pick Right 2026 评测"," 真实反馈，重度用户经常额外买 credit pack",[79,754,755,758],{},[35,756,757],{},"任务一启动无法控预算","：开始跑后只能 cancel 或等结束，credits 会一直消耗",[79,760,761,764],{},[35,762,763],{},"Web App Builder 别上生产","：demo 漂亮，复杂场景翻车，Bolt.new \u002F Lovable \u002F v0 仍是生产部署更稳选项",[79,766,767,770,771,776],{},[35,768,769],{},"每次任务从零开始","：没有持久 workspace，不像 ",[44,772,775],{"href":773,"rel":774},"https:\u002F\u002Fclaude.com",[48],"Claude Projects"," \u002F ChatGPT Custom GPTs 能跨会话记忆",[79,778,779,782],{},[35,780,781],{},"没有 HubSpot \u002F Salesforce \u002F Notion \u002F Slack 集成","：拿到结果后要自己手动搬到工具栈",[79,784,785,788],{},[35,786,787],{},"团队协作能力差","：单用户产品为主，无共享 workspace \u002F 评论 \u002F 审计",[79,790,791,794,795,800],{},[35,792,793],{},"Meta 收购的不确定性","：Wikipedia 引用 ",[44,796,799],{"href":797,"rel":798},"https:\u002F\u002Fwww.nytimes.com\u002F2026\u002F03\u002F17\u002Ftechnology\u002Fchina-scrutiny-meta-manus.html",[48],"纽时 2026-03"," 报道，中国审查 Meta 收购案，长期路线图待观察",[20,802,804],{"id":803},"适合-不适合","适合 \u002F 不适合",[32,806,807],{},"✅ 适合：",[76,809,810,813,816,819,822],{},[79,811,812],{},"委托多步骤 research（\"调研 X，产出 Markdown 对比表\"）",[79,814,815],{},"CSV \u002F Excel 重的数据分析任务",[79,817,818],{},"长任务 + 不想盯着看（小时级）",[79,820,821],{},"单兵作战的咨询顾问 \u002F 分析师 \u002F 创业者",[79,823,824],{},"对\"通用 Agent 体感天花板\"感兴趣的测试者",[32,826,827],{},"❌ 不适合：",[76,829,830,833,836,839,842],{},[79,831,832],{},"生产级应用部署",[79,834,835],{},"团队协作工作流",[79,837,838],{},"预算极敏感（free 档非常受限）",[79,840,841],{},"大陆稳定访问需求",[79,843,844,845,848],{},"需要深度持久上下文（用 ",[44,846,775],{"href":773,"rel":847},[48]," \u002F ChatGPT Custom GPT 等）",[20,850,851],{"id":851},"相关阅读",[76,853,854,865,876,891],{},[79,855,856,857,293,859,861,862,864],{},"同类对比：",[44,858,526],{"href":525},[44,860,500],{"href":499}," \u002F Elicit \u002F ",[44,863,721],{"href":504}," \u002F Metaso",[79,866,867,868,293,872,875],{},"概念：",[44,869,871],{"href":870},"\u002Fwiki\u002Fai-agent.html","AI Agent",[44,873,874],{"href":870},"Multi-Agent"," \u002F Computer Use \u002F Deep Research",[79,877,878,879,293,883,293,887],{},"模型：",[44,880,882],{"href":881},"\u002Fmodels\u002Fclaude-sonnet-4.html","Claude Sonnet 4",[44,884,886],{"href":885},"\u002Fmodels\u002Fclaude-opus-4.html","Claude Opus 4",[44,888,890],{"href":889},"\u002Fmodels\u002Fqwen-3.html","Qwen3",[79,892,893,894,293,898],{},"进阶：",[44,895,897],{"href":896},"\u002Fwiki\u002Fcontext-engineering.html","Context Engineering",[44,899,901],{"href":900},"\u002Fwiki\u002Fprompt-engineering.html","Prompt Engineering",[20,903,904],{"id":904},"来源",[76,906,907,913,916,919],{},[79,908,909,910],{},"官网：",[44,911,425],{"href":425,"rel":912},[48],[79,914,915],{},"Wikipedia：\"Manus (AI agent)\" 条目",[79,917,918],{},"第三方评测：pick-right.com \u002F hyzenpro.com \u002F infotech.com \u002F weforum.org",[79,920,921],{},"媒体报道：Reuters \u002F AP \u002F China Daily \u002F 36Kr \u002F NYT",[32,923,924,925,929],{},"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；如发现价格 \u002F 功能 \u002F 公司状态与最新官方信息不一致，请通过 ",[44,926,928],{"href":927},"mailto:hello@aiho.net","反馈邮箱"," 反馈。",{"title":448,"searchDepth":931,"depth":931,"links":932},3,[933,935,936,944,945,946,947,948,949,950,951],{"id":22,"depth":934,"text":23},2,{"id":74,"depth":934,"text":74},{"id":143,"depth":934,"text":143,"children":937},[938,939,940,941,942,943],{"id":147,"depth":931,"text":148},{"id":171,"depth":931,"text":172},{"id":260,"depth":931,"text":261},{"id":271,"depth":931,"text":272},{"id":305,"depth":931,"text":305},{"id":311,"depth":931,"text":311},{"id":321,"depth":934,"text":321},{"id":415,"depth":934,"text":416},{"id":466,"depth":934,"text":466},{"id":509,"depth":934,"text":509},{"id":734,"depth":934,"text":734},{"id":803,"depth":934,"text":804},{"id":851,"depth":934,"text":851},{"id":904,"depth":934,"text":904},"general",4,"\u002Fimg\u002Ftools\u002Fmanus.webp","Manus 真实评测：Butterfly Effect（蝴蝶效应）出品的通用 AI Agent，多 sub-agent 并行架构 + 多模型路由（Claude \u002F Qwen \u002F 自研）。2025-03 首发即引爆，2025-12 传被 Meta 以约 20 亿美元收购。AIHO 编辑部基于官方文档与多份评测整理。",false,"md",null,true,[961,962],"zh","en","2026-08-02",{},[966,967,968,969],"claude-sonnet-4","claude-opus-4","qwen-max","manus-internal",[971,972,973,974],"生产级 Web App 部署（Web App Builder 仍有 bug）","团队协作场景（功能未完善）","需要持久工作空间 \u002F 跨会话上下文","对成本极敏感（credits 烧得快）","\u002Ftools\u002Fagent\u002Fgeneral\u002Fmanus","agent",[978,979,980,981],"web","windows","macos","linux",[983,986,990,993],{"plan":349,"price":352,"limit":984,"cn_pay":578,"note":985},"每日有限 credits，足够每天 1 个 demanding 任务","试水\u002F评估",{"plan":360,"price":363,"limit":987,"cn_pay":988,"note":989},"更高 credit + 更长任务时长 + 多模型路由","⚠️ 需海外卡","个人主力档",{"plan":371,"price":374,"limit":991,"cn_pay":640,"note":992},"更大 credit 池 + 优先队列","10+ 任务\u002F周",{"plan":994,"price":995,"limit":996,"cn_pay":640,"note":997},"Pro+\u002FTeam","$200\u002F月","最大 credit + 团队工作空间（功能受限）","团队功能仍有限","Free \u002F Pro $20\u002Fmo \u002F Plus $50\u002Fmo \u002F Pro+ Team 最高 $200\u002Fmo","2026-06-18",[1001],"onboarding\u002Fmanus-getting-started",[1003,1004,1005],"manus-deep-review","autoglm-deep-review","openmanus-deep-review",{"power":1007,"ux":1007,"price":931,"cn_support":953,"stability":931},5,{"title":10,"description":955},"Manus AI Agent 评测 2026：通用 AI 智能体，自主完成任务，对比 Devin","agent\u002Fgeneral\u002Fmanus",[1012,1014,1017,1019,1021,1024],{"title":1013,"url":425},"Manus 官网",{"title":1015,"url":1016},"Manus 维基百科","https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FManus_(AI_agent)",{"title":1018,"url":153},"Manus AI Review 2026 (Pick Right)",{"title":1020,"url":177},"Info-Tech Manus Assessment",{"title":1022,"url":1023},"WEF Butterfly Effect Page","https:\u002F\u002Fwww.weforum.org\u002Forganizations\u002Fbutterfly-effect",{"title":1025,"url":1026},"HyzenPro 2026 Review","https:\u002F\u002Fhyzenpro.com\u002Fblog\u002Fmanus-ai-review","tools\u002Fagent\u002Fgeneral\u002Fmanus",[1029,1030,1031,1032,1033],"需要委托多步骤研究 \u002F 报告（带引用密度的 deep research）","数据分析（CSV \u002F Excel 重的任务）","需要","调研类工作（市场调研、竞品对比、文献综述）","对 Computer Use 类 Agent 实践有兴趣的人","通用 Agent 体感天花板，自主完成复杂多步骤任务",[1036,1037,1038,1039,1040,1041],"general-agent","autonomous","computer-use","multi-agent","butterfly-effect","monica","2026-06-24","通用 Agent 体感天花板代表。多 sub-agent 并行 + 自动模型路由让 research \u002F 数据分析输出明显比 ChatGPT Deep Research 引用密度更高。慢、credit 烧得快、Web App Builder 仍有 bug，但任务跑通时确实让人惊艳。","wcviURLmHN3k98k-_qnPkJ09CuZAWesyVH7TwM3COfM",[1046,1628,2169],{"id":1047,"title":500,"alternatives":1048,"api_compatible":958,"body":1052,"category":952,"chinese_friendly":953,"cover":1568,"description":1569,"domestic":959,"extension":957,"faq":1570,"free":959,"github":958,"languages":1583,"lastVerified":963,"meta":1585,"models":958,"navigation":959,"notSuitable":958,"opensource":956,"path":1586,"pillar":976,"platforms":1587,"priceTable":1590,"pricing":1601,"published":1602,"relatedPlaybooks":1603,"relatedReviews":1605,"score":1606,"self_host":956,"seo":1607,"seoTitle":1608,"slug":12,"sources":1609,"stem":1618,"suitable":958,"tagline":1619,"tags":1620,"updated":1042,"verdict":1625,"website":1626,"__hash__":1627},"tools\u002Ftools\u002Fagent\u002Fgeneral\u002Fgenspark.md",[1049,1050,1051],"agent\u002Fgeneral\u002Fflowith","agent\u002Fdesktop\u002Fclaude-desktop","agent\u002Fgeneral\u002Falice",{"type":17,"value":1053,"toc":1556},[1054,1056,1059,1062,1065,1139,1141,1164,1169,1173,1178,1201,1206,1232,1235,1255,1258,1415,1418,1474,1476,1502,1504,1524,1526],[20,1055,23],{"id":22},[32,1057,1058],{},"Genspark 是 2024 年创立于 Palo Alto 的 AI Super Agent 公司，CEO Eric Jing（前 Microsoft Bing \u002F 小冰），团队来自 Microsoft + Google。2026-04 ARR 突破 $250M（12 个月内）。差异点：『多 Agent 架构』路由不同 agent + 模型组合，而非单 LLM + Super Agent 自治执行覆盖研究 \u002F 内容 \u002F 调用 \u002F 建站 \u002F 视频 \u002F 数据分析 \u002F 真实电话 + Sparkpage 可视化输出。Free 100-200 daily credits \u002F Plus $25 月度 12k credits \u002F Pro $249 月度 125k credits。",[32,1060,1061],{},"适合：要一站式 Super Agent 完成完整工作流（研究 → 文档 → 演示 → 网站）；重度内容创作者 + 营销 \u002F 销售（Call For Me 跟进）；中级专业用户（Plus $25 \u002F 月划算）。不适合：纯研究问答（Perplexity \u002F ChatGPT 更便宜）；预算极紧（credit 消耗快）；需要透明引用 \u002F 学术严谨（Sparkpage 引用不如 Perplexity 清晰）；要画布多线程（用 Flowith）。",[20,1063,1064],{"id":1064},"核心能力",[76,1066,1067,1073,1079,1085,1091,1097,1103,1109,1115,1121,1127,1133],{},[79,1068,1069,1072],{},[35,1070,1071],{},"Super Agent","：自主规划 + 多步执行 + 工具调用",[79,1074,1075,1078],{},[35,1076,1077],{},"多 Agent 架构","：搜索 \u002F 研究 \u002F 内容 \u002F 调用 \u002F 数据分析 agent 协同",[79,1080,1081,1084],{},[35,1082,1083],{},"Sparkpage","：可视化研究输出（含图表 + 引用 + 可分享链接）",[79,1086,1087,1090],{},[35,1088,1089],{},"AI Slides","：10-15 张幻灯片含图表",[79,1092,1093,1096],{},[35,1094,1095],{},"AI Sites","：完整 landing page 一键生成",[79,1098,1099,1102],{},[35,1100,1101],{},"AI Video","：30-60s 短视频生成",[79,1104,1105,1108],{},[35,1106,1107],{},"AI Pods","：10 分钟 AI podcast 生成",[79,1110,1111,1114],{},[35,1112,1113],{},"Call For Me","：调用真实电话完成预订 \u002F 跟进",[79,1116,1117,1120],{},[35,1118,1119],{},"AI Sheets","：CSV 数据分析 + 自动图表",[79,1122,1123,1126],{},[35,1124,1125],{},"Plus 不限聊天","：o3-Pro \u002F Claude \u002F Gemini 顶级模型 chat 无限",[79,1128,1129,1132],{},[35,1130,1131],{},"iOS \u002F Android App","：移动端可用",[79,1134,1135,1138],{},[35,1136,1137],{},"企业 SSO \u002F API","（Pro+）：集成内部系统",[20,1140,339],{"id":339},[76,1142,1143,1148,1153,1158],{},[79,1144,1145,1147],{},[35,1146,349],{},"：100-200 daily credits（~3-8 简单任务）",[79,1149,1150,1152],{},[35,1151,371],{},"：$25\u002F月（年付 $240，~$20\u002F月）；12,000 月度 credits + 顶级模型不限 chat",[79,1154,1155,1157],{},[35,1156,360],{},"：$249\u002F月（年付 $2388，~$199\u002F月）；125,000 credits + 优先 + 早期 + 高端 agent",[79,1159,1160,1163],{},[35,1161,1162],{},"Extra credits","：仅 Pro 可买，$10 \u002F 5,000 credits",[67,1165,1166],{},[32,1167,1168],{},"Plus 一月做 10-15 个严肃任务（slides \u002F 研究 \u002F sites）；重度做视频 \u002F 通话 \u002F 多 sites 建议 Pro。",[20,1170,1172],{"id":1171},"实测营销-销售-内容生产","实测（营销 \u002F 销售 \u002F 内容生产）",[32,1174,1175],{},[35,1176,1177],{},"亮点：",[76,1179,1180,1183,1186,1189,1192,1195,1198],{},[79,1181,1182],{},"一句话『做一份 B2B SaaS 竞品分析 + 5 张幻灯片 + landing page』全自动跑通",[79,1184,1185],{},"AI Sheets 处理 CSV 数据 + 自动出图 + 写入 Sparkpage 一步到位",[79,1187,1188],{},"Call For Me 跟进客户 \u002F 餐厅预订实际能用，时效感最强",[79,1190,1191],{},"Plus 顶级模型不限 chat 性价比超高，比 ChatGPT Plus $20 + Claude Pro $20 + Perplexity Pro $20 全订便宜",[79,1193,1194],{},"多 Agent 在复杂任务上确实优于单 LLM 的 ChatGPT",[79,1196,1197],{},"Sparkpage 输出可直接分享 + 嵌入",[79,1199,1200],{},"ARR $250M 增长说明产品市场契合度（PMF）非常好",[32,1202,1203],{},[35,1204,1205],{},"踩坑：",[76,1207,1208,1211,1214,1217,1220,1223,1226,1229],{},[79,1209,1210],{},"credit 消耗不透明，复杂任务前难估算成本",[79,1212,1213],{},"用户反馈 customer support + refund 有问题",[79,1215,1216],{},"生成内容偶尔『AI 味』重，要人工编辑润色",[79,1218,1219],{},"引用 \u002F 来源不如 Perplexity 严谨，学术场景慎用",[79,1221,1222],{},"AI Sites \u002F Video 编辑灵活度低，定制要走外部工具",[79,1224,1225],{},"地域限制：印度 \u002F 巴西 \u002F 巴基斯坦 \u002F 尼日利亚 \u002F 印尼等部分国家 Plus \u002F Pro 不可用",[79,1227,1228],{},"Call For Me 隐私 \u002F 合规要小心（GDPR \u002F 加州 CCPA \u002F 录音同意）",[79,1230,1231],{},"Pro $249 月价格陡，中端断档",[20,1233,1234],{"id":1234},"上手",[418,1236,1237,1240,1243,1246,1249,1252],{},[79,1238,1239],{},"genspark.ai 注册 → Free 拿 100-200 daily credits",[79,1241,1242],{},"Super Agent 试『一句话目标』：『做一份 AI Agent 2026 竞品分析 Sparkpage + 5 张 Slides』",[79,1244,1245],{},"等 5-10 分钟看 Sparkpage 输出 → 编辑 → 分享链接",[79,1247,1248],{},"AI Sheets 上传 CSV → 自动出图 → 集成进 Sparkpage",[79,1250,1251],{},"Call For Me 试一次小餐厅预订（自有号码 + 同意）",[79,1253,1254],{},"Plus $25 \u002F 月体验完整 → 决定 Pro",[20,1256,1257],{"id":1257},"对比",[182,1259,1260,1277],{},[185,1261,1262],{},[188,1263,1264,1266,1268,1271,1274],{},[191,1265,518],{},[191,1267,500],{},[191,1269,1270],{},"Flowith",[191,1272,1273],{},"Perplexity Pro",[191,1275,1276],{},"ChatGPT Plus",[198,1278,1279,1295,1310,1324,1338,1354,1369,1384,1399],{},[188,1280,1281,1283,1286,1289,1292],{},[203,1282,1071],{},[203,1284,1285],{},"✅ 多 Agent",[203,1287,1288],{},"✅ Neo 自治",[203,1290,1291],{},"–",[203,1293,1294],{},"Operator",[188,1296,1297,1300,1303,1306,1308],{},[203,1298,1299],{},"可视化输出",[203,1301,1302],{},"✅ Sparkpage",[203,1304,1305],{},"✅ 画布",[203,1307,1291],{},[203,1309,1291],{},[188,1311,1312,1315,1318,1320,1322],{},[203,1313,1314],{},"真实电话调用",[203,1316,1317],{},"✅ Call For Me",[203,1319,1291],{},[203,1321,1291],{},[203,1323,1291],{},[188,1325,1326,1329,1332,1334,1336],{},[203,1327,1328],{},"一键建站",[203,1330,1331],{},"✅ AI Sites",[203,1333,1291],{},[203,1335,1291],{},[203,1337,1291],{},[188,1339,1340,1343,1346,1349,1351],{},[203,1341,1342],{},"视频 \u002F 音频生成",[203,1344,1345],{},"✅ Video \u002F Pods",[203,1347,1348],{},"部分",[203,1350,1291],{},[203,1352,1353],{},"Sora（独立）",[188,1355,1356,1359,1362,1364,1367],{},[203,1357,1358],{},"引用透明",[203,1360,1361],{},"中",[203,1363,1291],{},[203,1365,1366],{},"✅ 顶级",[203,1368,1348],{},[188,1370,1371,1374,1377,1380,1382],{},[203,1372,1373],{},"起价",[203,1375,1376],{},"$25\u002F月（Plus）",[203,1378,1379],{},"$19.90\u002F月",[203,1381,363],{},[203,1383,363],{},[188,1385,1386,1389,1392,1395,1397],{},[203,1387,1388],{},"最贵",[203,1390,1391],{},"$249\u002F月",[203,1393,1394],{},"$499.90\u002F月",[203,1396,363],{},[203,1398,995],{},[188,1400,1401,1404,1407,1410,1413],{},[203,1402,1403],{},"适合",[203,1405,1406],{},"一站式 Super Agent",[203,1408,1409],{},"画布深度",[203,1411,1412],{},"搜索引用",[203,1414,558],{},[20,1416,1417],{"id":1417},"避坑",[76,1419,1420,1426,1432,1438,1444,1450,1456,1462,1468],{},[79,1421,1422,1425],{},[35,1423,1424],{},"Plus 先用一个月","：评估实际 credit 消耗再决定升 Pro",[79,1427,1428,1431],{},[35,1429,1430],{},"复杂任务前估算 credit","：AI Video \u002F Pods \u002F Sites 烧得快",[79,1433,1434,1437],{},[35,1435,1436],{},"Sparkpage 引用要核实","：学术场景不要直接用，把链接打开看原文",[79,1439,1440,1443],{},[35,1441,1442],{},"Call For Me 合规","：欧盟 \u002F 加州录音同意法规要遵守",[79,1445,1446,1449],{},[35,1447,1448],{},"生成内容人工润色","：AI 味需要花 10-20% 时间打磨",[79,1451,1452,1455],{},[35,1453,1454],{},"地域限制","：在受限国家用 VPN + 海外卡，企业合规要走法务",[79,1457,1458,1461],{},[35,1459,1460],{},"AI Sites 不可深度编辑","：要灵活定制走 Webflow \u002F Framer + AI 工具",[79,1463,1464,1467],{},[35,1465,1466],{},"Pro 不要直接上","：Plus 重度可能仍够用，$200 差价省下来",[79,1469,1470,1473],{},[35,1471,1472],{},"Extra credits 性价比低","：$10 \u002F 5k credits 不如直接升 Pro",[20,1475,804],{"id":803},[76,1477,1478,1481,1484,1487,1490,1493,1496,1499],{},[79,1479,1480],{},"✅ 营销 \u002F 销售 \u002F 内容创作者 + 要全栈输出",[79,1482,1483],{},"✅ Plus $25 重度多模型不限 chat 用户",[79,1485,1486],{},"✅ 要真实电话调用 \u002F 一键建站 \u002F 短视频",[79,1488,1489],{},"✅ 中级专业用户（自由职业 \u002F 小团队）",[79,1491,1492],{},"❌ 纯研究问答（Perplexity 便宜 + 引用强）",[79,1494,1495],{},"❌ 学术严谨场景",[79,1497,1498],{},"❌ 预算极紧",[79,1500,1501],{},"❌ 要画布 + 多线程深度（用 Flowith）",[20,1503,851],{"id":851},[76,1505,1506,1512,1518],{},[79,1507,1508],{},[44,1509,1511],{"href":1510},"\u002Ftools\u002Fagent\u002Fgeneral\u002Fflowith","Flowith 评测",[79,1513,1514],{},[44,1515,1517],{"href":1516},"\u002Ftools\u002Fagent\u002Fdesktop\u002Fclaude-desktop","Claude Desktop 评测",[79,1519,1520],{},[44,1521,1523],{"href":1522},"\u002Ftools\u002Fagent\u002Fgeneral\u002Falice","Alice 评测",[20,1525,904],{"id":904},[418,1527,1528,1535,1542,1549],{},[79,1529,1530,1531],{},"Genspark Pricing 官方 ",[44,1532,1533],{"href":1533,"rel":1534},"https:\u002F\u002Fgenspark.ai\u002Fpricing",[48],[79,1536,1537,1538],{},"WebCraft — Genspark 2026 Review（Plus \u002F Pro \u002F credit 消耗）",[44,1539,1540],{"href":1540,"rel":1541},"https:\u002F\u002Fwebscraft.org\u002Fblog\u002Fgenspark-ai-oglyad-superagent-yakiy-avtonomno-stvoryuye-sayti-prezentatsiyi?lang=en",[48],[79,1543,1544,1545],{},"Rimo — Genspark 2026 + ARR $250M 数据 ",[44,1546,1547],{"href":1547,"rel":1548},"https:\u002F\u002Frimo.app\u002Fen\u002Fblogs\u002Fgenspark-ai_en-US",[48],[79,1550,1551,1552],{},"Lindy — Genspark Features 2026 Tested ",[44,1553,1554],{"href":1554,"rel":1555},"https:\u002F\u002Fwww.lindy.ai\u002Fblog\u002Fgenspark-ai-features",[48],{"title":448,"searchDepth":931,"depth":931,"links":1557},[1558,1559,1560,1561,1562,1563,1564,1565,1566,1567],{"id":22,"depth":934,"text":23},{"id":1064,"depth":934,"text":1064},{"id":339,"depth":934,"text":339},{"id":1171,"depth":934,"text":1172},{"id":1234,"depth":934,"text":1234},{"id":1257,"depth":934,"text":1257},{"id":1417,"depth":934,"text":1417},{"id":803,"depth":934,"text":804},{"id":851,"depth":934,"text":851},{"id":904,"depth":934,"text":904},"\u002Fimg\u002Ftools\u002Fgenspark.webp","Genspark 真实评测：2024 年创立于 Palo Alto，CEO Eric Jing（前 Microsoft Bing \u002F 小冰），团队来自 Microsoft + Google。差异点：『多 Agent 架构』而非单 LLM + Super Agent 自治执行（研究 \u002F 内容 \u002F 调用 \u002F 建站 \u002F 视频 \u002F 数据分析）+ Sparkpage 可视化输出 + Call For Me 真实电话调度。2026-04 ARR $250M（12 个月内达成）。Free 100-200 daily credits \u002F Plus $25 月度 12k credits \u002F Pro $249 月度 125k credits。",[1571,1574,1577,1580],{"q":1572,"a":1573},"credit 消耗如何？","Simple Sparkpage \u002F 研究 30-80 credits；AI Slides（10-15 张 + 图表）250-450；AI Sites（完整 landing）600-1200；AI Video（30-60s）800-2000；Call For Me（2-4 分钟）400-900；AI Pods（10 分钟 podcast）1000-1800。Plus 12k credits 大约一个月做 10-15 个严肃任务（不重度跑视频 \u002F 通话）。",{"q":1575,"a":1576},"Call For Me 真的能打电话？","对，调用真实电话 API（背后用 Bland AI \u002F Vapi 类供应商），帮你预订餐厅 \u002F 跟进客户 \u002F 查信息。需要明确隐私 + 合规：欧盟 \u002F 加州的呼叫录音 \u002F 同意法规要遵守，敏感场景慎用。",{"q":1578,"a":1579},"多 Agent 架构和单 LLM 区别？","Genspark 不依赖单一 LLM，而是按任务类型路由到不同 agent + 模型组合（搜索 agent \u002F 研究 agent \u002F 内容 agent \u002F 调用 agent 等）。优势：每类任务用最适合的工具 + 大型搜索 + 验证流水线；劣势：黑盒程度高 + credit 消耗模型让用量不可预测。",{"q":1581,"a":1582},"和 Flowith \u002F ChatGPT \u002F Perplexity 怎么选？","Genspark 强在『全栈 Super Agent + 真实调用（电话 \u002F 搜索 \u002F 建站）+ Sparkpage 可视化』。Flowith 强在『画布 + 长任务自治 + Knowledge Garden』。Perplexity 强在『搜索 + 引用透明 + 价格便宜』。ChatGPT 强在『通用 + 生态成熟』。一站式 Super Agent → Genspark；画布多线程 → Flowith；搜索问答 → Perplexity。",[962,961,1584],"multi",{},"\u002Ftools\u002Fagent\u002Fgeneral\u002Fgenspark",[978,1588,1589],"ios","android",[1591,1594,1598],{"plan":349,"price":352,"features":1592,"notes":1593},"100-200 daily credits + 基础模型 + Sparkpage \u002F 研究","试水",{"plan":371,"price":1595,"features":1596,"notes":1597},"$25\u002F月","12,000 月度 credits + 顶级模型不限聊天（o3-Pro \u002F Claude \u002F Gemini）+ Slides \u002F 研究","年付 $240",{"plan":360,"price":1391,"features":1599,"notes":1600},"125,000 月度 credits + 优先速度 + 早期功能 + AI Sites \u002F Video \u002F Call","年付 $2388","Free (100-200 daily) \u002F Plus $25·月 (12k credits) \u002F Pro $249·月 (125k credits) \u002F 年付有折扣","2026-06-19",[1604],"onboarding\u002Fsuper-agent-workflow",[1004,1003,1005],{"power":1007,"ux":953,"price":931,"cn_support":931,"stability":953},{"title":500,"description":1569},"Genspark 评测 2026：AI Super Agent，多 Agent 架构 + Sparkpage",[1610,1612,1614,1616],{"name":1611,"url":1533,"accessed":1042},"Genspark 官网 + Pricing",{"name":1613,"url":1540,"accessed":1042},"WebCraft — Genspark 2026 Review (Plus \u002F Pro \u002F Use Cases)",{"name":1615,"url":1547,"accessed":1042},"Rimo — Genspark 2026 + ARR $250M",{"name":1617,"url":1554,"accessed":1042},"Lindy — Genspark Features Tested 2026","tools\u002Fagent\u002Fgeneral\u002Fgenspark","Palo Alto 出品的 AI Super Agent——多 Agent 架构 + Sparkpage \u002F Slides \u002F Sites \u002F Video \u002F Call For Me 全栈",[1621,1039,1622,1623,1624],"super-agent","sparkpage","call-agent","genspark","Super Agent 类目里目前最完整的一站式产品——研究 \u002F 内容 \u002F 建站 \u002F 视频 \u002F 电话调度全包。Plus $25 \u002F 月在重度场景非常划算。要纯研究问答用 Perplexity \u002F ChatGPT 更便宜。","https:\u002F\u002Fgenspark.ai","hOtbYh5Ukxv1GZh0uG1a-ZODYPTQcbcpORQESsipc0Q",{"id":1629,"title":526,"alternatives":1630,"api_compatible":1633,"body":1636,"category":976,"chinese_friendly":934,"cover":2112,"description":2113,"domestic":956,"extension":957,"faq":2114,"free":956,"github":958,"languages":2127,"lastVerified":963,"meta":2128,"models":958,"navigation":959,"notSuitable":958,"opensource":956,"path":2129,"pillar":2130,"platforms":2131,"priceTable":2134,"pricing":2144,"published":1602,"relatedPlaybooks":2145,"relatedReviews":2147,"score":2149,"self_host":956,"seo":2150,"seoTitle":2151,"slug":13,"sources":2152,"stem":2161,"suitable":958,"tagline":2162,"tags":2163,"updated":1042,"verdict":2166,"website":2167,"__hash__":2168},"tools\u002Ftools\u002Fcoding\u002Fagent\u002Fdevin.md",[1631,1010,1632],"coding\u002Fagent\u002Fopenhands","coding\u002Fcli\u002Fclaude-code",[1634,1635],"OpenAI","Anthropic",{"type":17,"value":1637,"toc":2100},[1638,1640,1643,1646,1649,1652,1676,1679,1683,1688,1762,1767,1821,1824,1827,1832,1846,1851,1868,1870,1887,1890,1991,1993,2025,2027,2044,2046,2068,2070],[20,1639,23],{"id":22},[32,1641,1642],{},"Devin 是 Cognition Labs 推出的「自主 AI 软件工程师」。2024-03 首发时 $500\u002F月 + 引爆全网，2025-04 Devin 2.0 把入门价砍到 $20\u002F月（按 ACU 计费），2026 年又通过收购 Windsurf 扩张到桌面端。它的核心定位不是 copilot，而是「异步 agent」——把任务扔进去，它在云沙箱里独立 plan → code → test → PR，你第二天来看结果。",[32,1644,1645],{},"适合：有明确边界的工程任务（重构、bug fix、新 feature），不需要全程盯着的团队。不适合：交互式探索、紧急 hotfix、中文项目、超预算敏感的个人。",[20,1647,1648],{"id":1648},"它是什么",[32,1650,1651],{},"Devin 不在你电脑上跑——它有自己的云端 IDE + shell + 浏览器，等于一台 7×24 在线的远程开发机。给它一个 issue（GitHub \u002F Linear \u002F Slack 都行），它会：",[418,1653,1654,1661,1667,1670,1673],{},[79,1655,1656,1657,1660],{},"进 sandbox、",[446,1658,1659],{},"git clone"," 仓库、看 README、构建 Devin Wiki",[79,1662,1663,1666],{},[35,1664,1665],{},"Interactive Planning","：跟你确认实现方案",[79,1668,1669],{},"写代码、跑测试、调浏览器看结果",[79,1671,1672],{},"失败自己排查、改、再试（带 citations 引用代码行）",[79,1674,1675],{},"提 PR、回 Slack「搞定」",[32,1677,1678],{},"主打 Cloud + Async + 多并发：Core 最多 10 个 Devin 同时跑不同任务，Team 起无限并发。",[20,1680,1682],{"id":1681},"价格2026-06-综合官方-第三方追踪","价格（2026-06 综合官方 + 第三方追踪）",[32,1684,1685],{},[35,1686,1687],{},"Cloud Devin（按 ACU 计费）",[182,1689,1690,1709],{},[185,1691,1692],{},[188,1693,1694,1697,1700,1703,1706],{},[191,1695,1696],{},"Plan",[191,1698,1699],{},"月费",[191,1701,1702],{},"ACU 价",[191,1704,1705],{},"并发",[191,1707,1708],{},"关键能力",[198,1710,1711,1730,1747],{},[188,1712,1713,1718,1721,1724,1727],{},[203,1714,1715],{},[35,1716,1717],{},"Core",[203,1719,1720],{},"$20 起步（PAYG）",[203,1722,1723],{},"$2.25\u002FACU",[203,1725,1726],{},"10",[203,1728,1729],{},"Devin IDE、Ask Devin、Wiki",[188,1731,1732,1735,1738,1741,1744],{},[203,1733,1734],{},"Team",[203,1736,1737],{},"$500 含 250 ACU",[203,1739,1740],{},"$2\u002FACU（含内）",[203,1742,1743],{},"∞",[203,1745,1746],{},"API、专属 Slack 频道、early access",[188,1748,1749,1752,1755,1757,1759],{},[203,1750,1751],{},"Enterprise",[203,1753,1754],{},"Custom",[203,1756,1754],{},[203,1758,1743],{},[203,1760,1761],{},"VPC、SSO、Custom Devins",[32,1763,1764],{},[35,1765,1766],{},"Devin Desktop（2026-06 Windsurf 改名后）",[182,1768,1769,1780],{},[185,1770,1771],{},[188,1772,1773,1775,1777],{},[191,1774,1696],{},[191,1776,339],{},[191,1778,1779],{},"适用",[198,1781,1782,1791,1800,1810],{},[188,1783,1784,1786,1788],{},[203,1785,349],{},[203,1787,352],{},[203,1789,1790],{},"轻量配额",[188,1792,1793,1795,1797],{},[203,1794,360],{},[203,1796,363],{},[203,1798,1799],{},"个人开发者",[188,1801,1802,1805,1807],{},[203,1803,1804],{},"Max",[203,1806,995],{},[203,1808,1809],{},"重度用户",[188,1811,1812,1815,1818],{},[203,1813,1814],{},"Teams",[203,1816,1817],{},"$80\u002F月 + $40\u002F席",[203,1819,1820],{},"团队",[32,1822,1823],{},"ACU 经济账：1 ACU = 15 分钟 Devin 主动工作。一个中等任务（小 feature\u002F简单 bug）约消耗 0.5–1.5 ACU ≈ $1.13–$3.38。一个复杂 task（跨文件重构 + 测试）2–4 ACU ≈ $5–$10。",[20,1825,1826],{"id":1826},"实测",[32,1828,1829,401],{},[35,1830,1831],{},"好的地方",[76,1833,1834,1837,1840,1843],{},[79,1835,1836],{},"单次任务交付质量稳定——比早期 demo 时代靠谱多了",[79,1838,1839],{},"Devin Wiki 自动生成的代码库地图很有用",[79,1841,1842],{},"多并发：开 5 个 Devin 同时干 5 个独立 bug，效率拉满",[79,1844,1845],{},"Interactive Planning 让你能纠偏，不会写到一半发现方向错",[32,1847,1848,401],{},[35,1849,1850],{},"踩坑",[76,1852,1853,1856,1859,1862,1865],{},[79,1854,1855],{},"ACU 烧得快——一次 Devin 跑歪 + 自我修复循环可能 2-3 ACU 没了",[79,1857,1858],{},"中文几乎不能用——Cognition 没做本地化，prompt 中文也回英文",[79,1860,1861],{},"国内开发者用：网络 + 价格 + 中文三重劝退，建议看 OpenHands \u002F Manus",[79,1863,1864],{},"复杂上下文（10w 行老代码）依然容易迷失，需要人辅助拆任务",[79,1866,1867],{},"Free trial 没有，最低要先付 $20 才能试",[20,1869,1234],{"id":1234},[418,1871,1872,1875,1878,1881,1884],{},[79,1873,1874],{},"devin.ai 注册（GitHub OAuth），充值 $20 拿到 9 ACU",[79,1876,1877],{},"连接 GitHub repo + 选触发方式（Linear\u002FSlack\u002FWeb）",[79,1879,1880],{},"给一个 issue + 验收标准（越具体越省 ACU）",[79,1882,1883],{},"Interactive Planning：审一遍方案",[79,1885,1886],{},"Devin 自动跑、PR 自动开、等 review",[20,1888,1889],{"id":1889},"对比一览",[182,1891,1892,1908],{},[185,1893,1894],{},[188,1895,1896,1898,1900,1903,1905],{},[191,1897,518],{},[191,1899,526],{},[191,1901,1902],{},"OpenHands",[191,1904,10],{},[191,1906,1907],{},"Claude Code",[198,1909,1910,1927,1944,1960,1975],{},[188,1911,1912,1915,1918,1921,1924],{},[203,1913,1914],{},"模式",[203,1916,1917],{},"异步 Cloud Agent",[203,1919,1920],{},"自托管 \u002F Cloud",[203,1922,1923],{},"异步 Cloud",[203,1925,1926],{},"同步 CLI",[188,1928,1929,1932,1935,1938,1941],{},[203,1930,1931],{},"起步价",[203,1933,1934],{},"$20 + ACU",[203,1936,1937],{},"免费（开源）",[203,1939,1940],{},"$39\u002F月",[203,1942,1943],{},"Pro $20\u002F月",[188,1945,1946,1948,1951,1954,1957],{},[203,1947,1705],{},[203,1949,1950],{},"10–∞",[203,1952,1953],{},"自托管无限",[203,1955,1956],{},"5+",[203,1958,1959],{},"1",[188,1961,1962,1964,1967,1970,1973],{},[203,1963,634],{},[203,1965,1966],{},"弱",[203,1968,1969],{},"强（社区维护）",[203,1971,1972],{},"强（国产）",[203,1974,1361],{},[188,1976,1977,1980,1983,1986,1988],{},[203,1978,1979],{},"沙箱",[203,1981,1982],{},"内置云端",[203,1984,1985],{},"Docker \u002F Cloud",[203,1987,1982],{},[203,1989,1990],{},"本地",[20,1992,1417],{"id":1417},[76,1994,1995,2001,2007,2013,2019],{},[79,1996,1997,2000],{},[35,1998,1999],{},"不要不写验收标准就丢任务","：Devin 会尽职跑，但跑到哪里算「完成」很主观，写明「跑通这个测试」「PR 必须含 changelog」",[79,2002,2003,2006],{},[35,2004,2005],{},"新项目 \u002F 老项目都要先 Wiki","：让它先生成 Devin Wiki，后续任务上下文准很多",[79,2008,2009,2012],{},[35,2010,2011],{},"ACU 监控要开","：仪表盘里设单任务 ACU 上限，避免 runaway",[79,2014,2015,2018],{},[35,2016,2017],{},"PR 必须 review","：Devin 偶尔会引入安全问题（hardcode key、删测试）",[79,2020,2021,2024],{},[35,2022,2023],{},"国内团队三思","：网络、定价、语言成本叠加后，单位产出 ROI 不一定胜过 Cursor + Claude",[20,2026,804],{"id":803},[76,2028,2029,2032,2035,2038,2041],{},[79,2030,2031],{},"✅ 海外团队、英文项目、批量化的中小 task",[79,2033,2034],{},"✅ 想做「夜跑 agent」流水线的工程团队",[79,2036,2037],{},"❌ 中文项目、国内个人开发者",[79,2039,2040],{},"❌ 预算敏感（同样钱可以买 4 个 Cursor 席位）",[79,2042,2043],{},"❌ 需要实时协作 \u002F 紧急 hotfix",[20,2045,851],{"id":851},[76,2047,2048,2055,2061],{},[79,2049,2050,2054],{},[44,2051,2053],{"href":2052},"\u002Ftools\u002Fcoding\u002Fagent\u002Fopenhands","OpenHands 评测","（Devin 的开源平替）",[79,2056,2057,2060],{},[44,2058,2059],{"href":975},"Manus 评测","（国产 cloud agent）",[79,2062,2063,2067],{},[44,2064,2066],{"href":2065},"\u002Ftools\u002Fcoding\u002Fcli\u002Fclaude-code","Claude Code 评测","（同步替代方案）",[20,2069,904],{"id":904},[418,2071,2072,2079,2086,2093],{},[79,2073,2074,2075],{},"SiliconAngle — Devin 2.0 with lower starting price（2025-04-03）",[44,2076,2077],{"href":2077,"rel":2078},"https:\u002F\u002Fsiliconangle.com\u002F2025\u002F04\u002F03\u002Fcognition-ai-launches-revamped-coding-assistant-devin-2-0-much-lower-starting-price",[48],[79,2080,2081,2082],{},"CostBench — Devin AI Pricing 2026（2026-05-04）",[44,2083,2084],{"href":2084,"rel":2085},"https:\u002F\u002Fcostbench.com\u002Fsoftware\u002Fai-coding-assistants\u002Fdevin-ai\u002F",[48],[79,2087,2088,2089],{},"eesel — Cognition AI Pricing Explained ",[44,2090,2091],{"href":2091,"rel":2092},"https:\u002F\u002Fwww.eesel.ai\u002Fblog\u002Fcognition-ai-pricing",[48],[79,2094,2095,2096],{},"Vibe Coding — Devin Review 2026 \u002F Windsurf rebrand ",[44,2097,2098],{"href":2098,"rel":2099},"https:\u002F\u002Fvibecoding.app\u002Fblog\u002Fdevin-review",[48],{"title":448,"searchDepth":931,"depth":931,"links":2101},[2102,2103,2104,2105,2106,2107,2108,2109,2110,2111],{"id":22,"depth":934,"text":23},{"id":1648,"depth":934,"text":1648},{"id":1681,"depth":934,"text":1682},{"id":1826,"depth":934,"text":1826},{"id":1234,"depth":934,"text":1234},{"id":1889,"depth":934,"text":1889},{"id":1417,"depth":934,"text":1417},{"id":803,"depth":934,"text":804},{"id":851,"depth":934,"text":851},{"id":904,"depth":934,"text":904},"\u002Fimg\u002Ftools\u002Fdevin.webp","Devin 真实评测：Cognition Labs 的自主编程 Agent。Core $20\u002F月 + $2.25\u002FACU（1 ACU ≈ 15 分钟），Team $500\u002F月含 250 ACUs。2026 收购 Windsurf 后产品线扩张到 Desktop Free\u002FPro\u002FMax。",[2115,2118,2121,2124],{"q":2116,"a":2117},"1 ACU 等于多少？","1 ACU ≈ Devin 主动工作 15 分钟。Core 单价 $2.25\u002FACU，Team 单价 $2\u002FACU。换算下来 Devin 主动小时成本约 $8–$9，确实接近 Upwork 入门级海外开发者，但能 7×24 工作。",{"q":2119,"a":2120},"Devin 和 Cursor \u002F Claude Code 区别？","Cursor \u002F Claude Code 是『陪你写』的 copilot，Devin 是『代你写』的 async agent——你给任务、它在云沙箱里独立完成、返回 PR。前者实时同步，后者异步交付。",{"q":2122,"a":2123},"2026 年 Devin \u002F Windsurf 关系？","Cognition 收购了 Windsurf 资产（前 Codeium），2026-06 后 Devin Desktop 取代 Windsurf IDE，新增 Free \u002F Pro $20\u002FMax $200 桌面端订阅，Cloud 版（Core\u002FTeam\u002FEnterprise）保留 ACU 计费。",{"q":2125,"a":2126},"Devin 能跑国内项目吗？","技术上可以——它在云沙箱跑 Linux 命令、git、浏览器。但默认走美国节点，访问国内 git\u002Fnpm 镜像慢，且不支持中文 prompt（会回英文）。国内团队建议看 OpenHands 自托管或 Manus。",[962],{},"\u002Ftools\u002Fcoding\u002Fagent\u002Fdevin","coding",[2132,2133],"cloud","desktop",[2135,2139,2142],{"plan":2136,"price":2137,"seats":1959,"notes":2138},"Core (PAYG)","$20\u002F月起","Pay-as-you-go，$2.25\u002FACU，最多 10 并发会话",{"plan":1734,"price":656,"seats":2140,"notes":2141},"≥1","含 250 ACUs（$2\u002FACU 实际单价）、无限并发、Slack 专属频道",{"plan":1751,"price":1754,"seats":578,"notes":2143},"VPC 部署、SAML\u002FOIDC SSO、Custom Devins","Core $20·月 起按 ACU \u002F Team $500·月含 250 ACU \u002F Enterprise 定制",[2146],"review\u002Fai-pr-review-pipeline",[2148],"augment-deep-review",{"power":1007,"ux":953,"price":934,"cn_support":934,"stability":953},{"title":526,"description":2113},"Devin AI 编程 Agent 评测 2026：Cognition 出品，自动驾驶代码生成",[2153,2155,2157,2159],{"name":2154,"url":2077,"accessed":1042},"SiliconAngle — Devin 2.0 launch",{"name":2156,"url":2084,"accessed":1042},"CostBench — Devin AI Pricing 2026",{"name":2158,"url":2091,"accessed":1042},"eesel — Cognition AI Pricing",{"name":2160,"url":2098,"accessed":1042},"Vibe Coding — Devin Review 2026 (Windsurf rebrand)","tools\u002Fcoding\u002Fagent\u002Fdevin","Cognition 出品的自主 AI 软件工程师，ACU 计费，2026 已并入 Windsurf 阵营",[976,1037,2132,2164,2165],"ACU","async","AI 自主编程的开山旗手，2026 价格从 $500 起跳降到 $20 起步。ACU 计费让成本可预算化，但单 ACU $2.25 仍是同类最贵，适合『把任务扔进去、明天看 PR』的异步开发模式。","https:\u002F\u002Fdevin.ai","MA81RLifvrit8tLQc807RgMZSU7N3U4Sbogf1MrCYJ4",{"id":2170,"title":2171,"alternatives":2172,"api_compatible":2177,"body":2182,"category":3199,"chinese_friendly":953,"cover":3200,"description":3201,"domestic":959,"extension":957,"faq":958,"free":959,"github":2466,"languages":3202,"lastVerified":963,"meta":3204,"models":958,"navigation":959,"notSuitable":958,"opensource":959,"path":3205,"pillar":976,"platforms":3206,"priceTable":3207,"pricing":3223,"published":999,"relatedPlaybooks":958,"relatedReviews":3224,"score":3229,"self_host":959,"seo":3230,"seoTitle":3231,"slug":14,"sources":3232,"stem":3244,"suitable":958,"tagline":3245,"tags":3246,"updated":1042,"verdict":3254,"website":3144,"__hash__":3255},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Fdify.md","Dify",[2173,2174,2175,2176],"agent\u002Fplatform\u002Fcoze","agent\u002Fplatform\u002Ffastgpt","agent\u002Fplatform\u002Fn8n","agent\u002Fplatform\u002Flangflow",[1634,1635,2178,2179,2180,2181],"百度文心","阿里通义","Moonshot Kimi","DeepSeek",{"type":17,"value":2183,"toc":3177},[2184,2186,2206,2211,2213,2217,2220,2288,2291,2295,2298,2312,2330,2334,2337,2348,2352,2360,2371,2375,2378,2380,2384,2393,2451,2458,2462,2470,2533,2536,2556,2560,2563,2622,2628,2632,2726,2728,2757,2759,2895,2913,2952,2954,3026,3028,3030,3050,3052,3081,3083,3136,3138,3168,3173],[20,2185,23],{"id":22},[25,2187,2189,2194],{"className":2188},[28,29,30],[32,2190,2191,2193],{},[35,2192,37],{}," 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。",[32,2195,2196,2197,2200,2201,2205],{},"最大价值是 ",[35,2198,2199],{},"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 ",[44,2202,2204],{"href":2203},"\u002Fagent\u002Fplatform\u002Fcoze.html","Coze"," 折腾，新手得读 1-2 小时文档。",[67,2207,2208],{},[32,2209,2210],{},"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。",[20,2212,143],{"id":143},[145,2214,2216],{"id":2215},"可视化工作流chatflow-workflow","可视化工作流（Chatflow + Workflow）",[32,2218,2219],{},"Dify 把 LLM 应用拆成两种\"应用类型\"：",[182,2221,2222,2234],{},[185,2223,2224],{},[188,2225,2226,2229,2231],{},[191,2227,2228],{},"类型",[191,2230,669],{},[191,2232,2233],{},"编排范式",[198,2235,2236,2249,2262,2275],{},[188,2237,2238,2243,2246],{},[203,2239,2240],{},[35,2241,2242],{},"Chatbot",[203,2244,2245],{},"简单对话机器人",[203,2247,2248],{},"prompt + tools",[188,2250,2251,2256,2259],{},[203,2252,2253],{},[35,2254,2255],{},"Agent",[203,2257,2258],{},"自主多步任务",[203,2260,2261],{},"ReAct \u002F Function Calling",[188,2263,2264,2269,2272],{},[203,2265,2266],{},[35,2267,2268],{},"Chatflow",[203,2270,2271],{},"对话型工作流（多轮 + 分支）",[203,2273,2274],{},"节点 DAG，带聊天上下文",[188,2276,2277,2282,2285],{},[203,2278,2279],{},[35,2280,2281],{},"Workflow",[203,2283,2284],{},"单次输入→输出（API 模式）",[203,2286,2287],{},"节点 DAG，无对话状态",[32,2289,2290],{},"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\"。",[145,2292,2294],{"id":2293},"rag-知识库","RAG 知识库",[32,2296,2297],{},"内置完整 RAG 链路：",[418,2299,2300,2303,2306,2309],{},[79,2301,2302],{},"上传文档（PDF \u002F Word \u002F Markdown \u002F 网页）",[79,2304,2305],{},"自动分块 + embedding（可配置分段策略和 embedding 模型）",[79,2307,2308],{},"混合检索（向量 + 全文 + 重排）",[79,2310,2311],{},"引用溯源（回答末尾自动附原文片段）",[32,2313,2314,2315,2320,2321,2324,2325,2329],{},"注意：根据 ",[44,2316,2319],{"href":2317,"rel":2318},"https:\u002F\u002Fzhuanlan.zhihu.com\u002Fp\u002F1887141987838309480",[48],"知乎 LLM 实战笔记 2025-03 对比"," 的实测，Dify ",[35,2322,2323],{},"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 ",[44,2326,2328],{"href":2327},"\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。",[145,2331,2333],{"id":2332},"模型生态40-提供商","模型生态：40+ 提供商",[32,2335,2336],{},"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。",[32,2338,2339,2340,2342,2343,2347],{},"国产模型原生支持（不像 ",[44,2341,2328],{"href":2327}," 需要 ",[44,2344,2346],{"href":2345},"\u002Fcoding\u002Fapi\u002Fone-api.html","OneAPI"," 中转），是 Dify 在国内 toB 场景流行的关键。",[145,2349,2351],{"id":2350},"mcp-协议支持","MCP 协议支持",[32,2353,2354,2355,2359],{},"Dify 较早接入了 ",[44,2356,2358],{"href":2357},"\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：",[76,2361,2362,2365,2368],{},[79,2363,2364],{},"通过 MCP 调本地 PostgreSQL \u002F SQLite",[79,2366,2367],{},"通过 MCP 调 GitHub \u002F Slack \u002F Linear",[79,2369,2370],{},"通过 MCP 调自家内部系统（写一个 MCP Server 即可）",[145,2372,2374],{"id":2373},"api-first","API-first",[32,2376,2377],{},"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。",[20,2379,321],{"id":321},[145,2381,2383],{"id":2382},"云版difyai","云版（dify.ai）",[32,2385,2386,2387,2392],{},"根据 ",[44,2388,2391],{"href":2389,"rel":2390},"https:\u002F\u002Fwww.tooljunction.io\u002Fai-tools\u002Fdify-ai",[48],"tooljunction.io 2026 评测"," 引用的官方定价：",[182,2394,2395,2407],{},[185,2396,2397],{},[188,2398,2399,2402,2404],{},[191,2400,2401],{},"套餐",[191,2403,339],{},[191,2405,2406],{},"主要限制",[198,2408,2409,2420,2431,2441],{},[188,2410,2411,2414,2417],{},[203,2412,2413],{},"Sandbox",[203,2415,2416],{},"免费",[203,2418,2419],{},"200 次模型调用，1 app，5MB 知识库",[188,2421,2422,2425,2428],{},[203,2423,2424],{},"Professional",[203,2426,2427],{},"$59\u002F月起",[203,2429,2430],{},"5000 调用\u002F月，多 app，50MB 知识库",[188,2432,2433,2435,2438],{},[203,2434,1734],{},[203,2436,2437],{},"$159\u002F月起",[203,2439,2440],{},"团队协作、SSO",[188,2442,2443,2445,2448],{},[203,2444,1751],{},[203,2446,2447],{},"联系销售",[203,2449,2450],{},"定制 SLA、私有云",[32,2452,2453,2454,2457],{},"注意：云版价格只是 Dify 平台费，",[35,2455,2456],{},"模型 API 费用另算","（自带 OpenAI \u002F Anthropic key）。",[145,2459,2461],{"id":2460},"自托管推荐","自托管（推荐）",[32,2463,2464,2469],{},[44,2465,2468],{"href":2466,"rel":2467},"https:\u002F\u002Fgithub.com\u002Flanggenius\u002Fdify",[48],"官方 GitHub 仓库"," 提供 Docker Compose 部署，社区版完全免费可商用：",[439,2471,2475],{"className":2472,"code":2473,"language":2474,"meta":448,"style":448},"language-bash shiki shiki-themes github-light github-dark","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","bash",[446,2476,2477,2493,2502,2513,2527],{"__ignoreMap":448},[2478,2479,2482,2486,2490],"span",{"class":2480,"line":2481},"line",1,[2478,2483,2485],{"class":2484},"sScJk","git",[2478,2487,2489],{"class":2488},"sZZnC"," clone",[2478,2491,2492],{"class":2488}," https:\u002F\u002Fgithub.com\u002Flanggenius\u002Fdify.git\n",[2478,2494,2495,2499],{"class":2480,"line":934},[2478,2496,2498],{"class":2497},"sj4cs","cd",[2478,2500,2501],{"class":2488}," dify\u002Fdocker\n",[2478,2503,2504,2507,2510],{"class":2480,"line":931},[2478,2505,2506],{"class":2484},"cp",[2478,2508,2509],{"class":2488}," .env.example",[2478,2511,2512],{"class":2488}," .env\n",[2478,2514,2515,2518,2521,2524],{"class":2480,"line":953},[2478,2516,2517],{"class":2484},"docker",[2478,2519,2520],{"class":2488}," compose",[2478,2522,2523],{"class":2488}," up",[2478,2525,2526],{"class":2497}," -d\n",[2478,2528,2529],{"class":2480,"line":1007},[2478,2530,2532],{"class":2531},"sJ8bj","# 默认 http:\u002F\u002Flocalhost \u002F 端口可在 .env 调整\n",[32,2534,2535],{},"硬件门槛（社区共识，非官方硬性要求）：",[76,2537,2538,2544,2550],{},[79,2539,2540,2543],{},[35,2541,2542],{},"最低","：2 核 4G，纯外接 API 模式",[79,2545,2546,2549],{},[35,2547,2548],{},"推荐","：4 核 8G + 至少 30GB 磁盘（向量数据 + 文件存储）",[79,2551,2552,2555],{},[35,2553,2554],{},"企业","：8 核 16G+，单机日活上千",[145,2557,2559],{"id":2558},"真实-tco","真实 TCO",[32,2561,2562],{},"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）：",[182,2564,2565,2578],{},[185,2566,2567],{},[188,2568,2569,2572,2575],{},[191,2570,2571],{},"成本项",[191,2573,2574],{},"云版 Professional",[191,2576,2577],{},"自托管",[198,2579,2580,2590,2600,2611],{},[188,2581,2582,2585,2588],{},[203,2583,2584],{},"平台费",[203,2586,2587],{},"~$2,100（3 年）",[203,2589,352],{},[188,2591,2592,2595,2597],{},[203,2593,2594],{},"服务器",[203,2596,352],{},[203,2598,2599],{},"~$50\u002F月 × 36 = $1,800",[188,2601,2602,2605,2608],{},[203,2603,2604],{},"模型 API",[203,2606,2607],{},"与下同",[203,2609,2610],{},"与上同",[188,2612,2613,2616,2619],{},[203,2614,2615],{},"运维人力",[203,2617,2618],{},"0",[203,2620,2621],{},"约 0.2 人月",[32,2623,2624,2627],{},[35,2625,2626],{},"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。",[20,2629,2631],{"id":2630},"上手-10-分钟","上手 10 分钟",[439,2633,2635],{"className":2472,"code":2634,"language":2474,"meta":448,"style":448},"# 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",[446,2636,2637,2642,2650,2656,2664,2674,2680,2686,2692,2697,2703,2708,2714,2720],{"__ignoreMap":448},[2478,2638,2639],{"class":2480,"line":2481},[2478,2640,2641],{"class":2531},"# 1. 自托管（社区版）\n",[2478,2643,2644,2646,2648],{"class":2480,"line":934},[2478,2645,2485],{"class":2484},[2478,2647,2489],{"class":2488},[2478,2649,2492],{"class":2488},[2478,2651,2652,2654],{"class":2480,"line":931},[2478,2653,2498],{"class":2497},[2478,2655,2501],{"class":2488},[2478,2657,2658,2660,2662],{"class":2480,"line":953},[2478,2659,2506],{"class":2484},[2478,2661,2509],{"class":2488},[2478,2663,2512],{"class":2488},[2478,2665,2666,2668,2670,2672],{"class":2480,"line":1007},[2478,2667,2517],{"class":2484},[2478,2669,2520],{"class":2488},[2478,2671,2523],{"class":2488},[2478,2673,2526],{"class":2497},[2478,2675,2677],{"class":2480,"line":2676},6,[2478,2678,2679],{"emptyLinePlaceholder":959},"\n",[2478,2681,2683],{"class":2480,"line":2682},7,[2478,2684,2685],{"class":2531},"# 2. 浏览器打开 http:\u002F\u002Flocalhost\n",[2478,2687,2689],{"class":2480,"line":2688},8,[2478,2690,2691],{"class":2531},"#    首次会让你创建 admin 账号\n",[2478,2693,2695],{"class":2480,"line":2694},9,[2478,2696,2679],{"emptyLinePlaceholder":959},[2478,2698,2700],{"class":2480,"line":2699},10,[2478,2701,2702],{"class":2531},"# 3. 进入\"设置 → 模型供应商\"，配置 OpenAI \u002F 国产模型 API key\n",[2478,2704,2706],{"class":2480,"line":2705},11,[2478,2707,2679],{"emptyLinePlaceholder":959},[2478,2709,2711],{"class":2480,"line":2710},12,[2478,2712,2713],{"class":2531},"# 4. 在主界面\"创建空白应用\"，选 Chatflow 或 Workflow\n",[2478,2715,2717],{"class":2480,"line":2716},13,[2478,2718,2719],{"class":2531},"# 5. 拖入\"开始 → LLM → 结束\"节点试一下基础 prompt\n",[2478,2721,2723],{"class":2480,"line":2722},14,[2478,2724,2725],{"class":2531},"# 6. 满意了点右上\"发布\"，自动生成 API endpoint\n",[20,2727,466],{"id":466},[418,2729,2730,2736,2742,2748],{},[79,2731,2732,2735],{},[35,2733,2734],{},"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",[79,2737,2738,2741],{},[35,2739,2740],{},"自托管 + 国产模型"," = 完全国内闭环，是 Dify 在国内最大优势",[79,2743,2744,2747],{},[35,2745,2746],{},"Docker 镜像拉取","：国内可能慢，建议配 Docker registry 镜像（阿里云 \u002F 网易）",[79,2749,2750,2753,2754,2756],{},[35,2751,2752],{},"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 ",[44,2755,2204],{"href":2203}," 迁到 Dify",[20,2758,509],{"id":509},[182,2760,2761,2783],{},[185,2762,2763],{},[188,2764,2765,2767,2769,2773,2777],{},[191,2766,518],{},[191,2768,2171],{},[191,2770,2771],{},[44,2772,2204],{"href":2203},[191,2774,2775],{},[44,2776,2328],{"href":2327},[191,2778,2779],{},[44,2780,2782],{"href":2781},"\u002Fagent\u002Fplatform\u002Fn8n.html","n8n",[198,2784,2785,2800,2813,2827,2840,2854,2868,2881],{},[188,2786,2787,2790,2793,2795,2797],{},[203,2788,2789],{},"开源",[203,2791,2792],{},"✅",[203,2794,623],{},[203,2796,2792],{},[203,2798,2799],{},"✅（fair-code）",[188,2801,2802,2805,2807,2809,2811],{},[203,2803,2804],{},"私有部署",[203,2806,2792],{},[203,2808,623],{},[203,2810,2792],{},[203,2812,2792],{},[188,2814,2815,2818,2820,2823,2825],{},[203,2816,2817],{},"上手难度",[203,2819,592],{},[203,2821,2822],{},"★★☆☆☆ 最简单",[203,2824,592],{},[203,2826,595],{},[188,2828,2829,2832,2834,2836,2838],{},[203,2830,2831],{},"工作流编排",[203,2833,589],{},[203,2835,595],{},[203,2837,592],{},[203,2839,589],{},[188,2841,2842,2845,2847,2849,2851],{},[203,2843,2844],{},"RAG 精度",[203,2846,595],{},[203,2848,592],{},[203,2850,589],{},[203,2852,2853],{},"★★☆☆☆",[188,2855,2856,2859,2861,2863,2866],{},[203,2857,2858],{},"模型生态",[203,2860,589],{},[203,2862,595],{},[203,2864,2865],{},"★★★☆☆（OneAPI 中转）",[203,2867,595],{},[188,2869,2870,2873,2875,2877,2879],{},[203,2871,2872],{},"中文场景",[203,2874,595],{},[203,2876,589],{},[203,2878,595],{},[203,2880,592],{},[188,2882,2883,2886,2888,2891,2893],{},[203,2884,2885],{},"字节生态绑定",[203,2887,623],{},[203,2889,2890],{},"✅（飞书\u002F抖音深度集成）",[203,2892,623],{},[203,2894,623],{},[32,2896,2897,2900,2901,2906,2907,2912],{},[35,2898,2899],{},"怎么选","（基于 ",[44,2902,2905],{"href":2903,"rel":2904},"https:\u002F\u002Fwww.besthub.dev\u002Farticles\u002Fcoze-vs-dify-vs-fastgpt-which-ai-agent-platform-fits-your-needs-fa59cf97b798",[48],"BestHub 2025-07"," 和 ",[44,2908,2911],{"href":2909,"rel":2910},"https:\u002F\u002Fwww.cnblogs.com\u002Fuulucias\u002Fp\u002F19449008",[48],"博客园 2026-01"," 两份选型指南综合）：",[76,2914,2915,2921,2929,2936,2943],{},[79,2916,2917,2920],{},[35,2918,2919],{},"数据必须不出内网 + 工作流复杂"," → Dify",[79,2922,2923,2926,2927],{},[35,2924,2925],{},"个人 \u002F 小团队 \u002F 快速原型 + 字节生态"," → ",[44,2928,2204],{"href":2203},[79,2930,2931,2926,2934],{},[35,2932,2933],{},"核心场景就是企业知识库 QA",[44,2935,2328],{"href":2327},[79,2937,2938,2926,2941],{},[35,2939,2940],{},"重点是连接外部 SaaS（Slack \u002F Notion \u002F 数据库）",[44,2942,2782],{"href":2781},[79,2944,2945,2926,2948],{},[35,2946,2947],{},"要画图式表达 LangChain pipeline",[44,2949,2951],{"href":2950},"\u002Fagent\u002Fplatform\u002Flangflow.html","Langflow",[20,2953,734],{"id":734},[76,2955,2956,2962,2978,2989,3002,3008,3014,3020],{},[79,2957,2958,2961],{},[35,2959,2960],{},"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。",[79,2963,2964,401,2970,2973,2974,2977],{},[35,2965,2966,2969],{},[446,2967,2968],{},".env"," 文件改完忘 restart",[446,2971,2972],{},"docker compose down && up -d","，不是 ",[446,2975,2976],{},"restart","——后者不重新加载 env。",[79,2979,2980,401,2983,2988],{},[35,2981,2982],{},"大版本升级会破坏数据库 schema",[44,2984,2987],{"href":2985,"rel":2986},"https:\u002F\u002Fdocs.dify.ai\u002Fzh-hans",[48],"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。",[79,2990,2991,2994,2995,2997,2998,3001],{},[35,2992,2993],{},"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 ",[446,2996,2968],{}," 的 ",[446,2999,3000],{},"UPLOAD_FILE_SIZE_LIMIT"," 并重启容器。",[79,3003,3004,3007],{},[35,3005,3006],{},"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。",[79,3009,3010,3013],{},[35,3011,3012],{},"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。",[79,3015,3016,3019],{},[35,3017,3018],{},"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。",[79,3021,3022,3025],{},[35,3023,3024],{},"国内 Docker 拉取镜像慢","：先配国内 registry，否则首次 pull 可能要 30+ 分钟。",[20,3027,804],{"id":803},[32,3029,807],{},[76,3031,3032,3035,3038,3041,3044,3047],{},[79,3033,3034],{},"中大型企业 LLM 中台建设",[79,3036,3037],{},"需要私有化部署（金融 \u002F 医疗 \u002F 政府）",[79,3039,3040],{},"想做\"AI 工作流即产品\"的开发团队",[79,3042,3043],{},"同时需要 RAG + Agent + Workflow 三件套",[79,3045,3046],{},"想用国产模型 + 国际模型混合编排",[79,3048,3049],{},"已经接受 Docker + 一定运维投入",[32,3051,827],{},[76,3053,3054,3060,3066,3069,3075],{},[79,3055,3056,3057,3059],{},"纯个人玩家做对话机器人（",[44,3058,2204],{"href":2203}," 更快）",[79,3061,3062,3063,3065],{},"只想做企业知识库 QA（",[44,3064,2328],{"href":2327}," RAG 更专）",[79,3067,3068],{},"团队完全没运维能力（云版还行，自托管会踩坑）",[79,3070,3071,3072,3074],{},"需要深度对接字节飞书 \u002F 抖音（",[44,3073,2204],{"href":2203}," 原生）",[79,3076,3077,3078,3080],{},"工作流核心是连接 100+ SaaS（",[44,3079,2782],{"href":2781}," 节点更全）",[20,3082,851],{"id":851},[76,3084,3085,3095,3111,3128],{},[79,3086,856,3087,293,3089,293,3091,293,3093],{},[44,3088,2204],{"href":2203},[44,3090,2328],{"href":2327},[44,3092,2782],{"href":2781},[44,3094,2951],{"href":2950},[79,3096,3097,3098,293,3100,293,3104,293,3107],{},"概念基础：",[44,3099,871],{"href":870},[44,3101,3103],{"href":3102},"\u002Fwiki\u002Frag.html","RAG",[44,3105,3106],{"href":2357},"MCP",[44,3108,3110],{"href":3109},"\u002Fwiki\u002Ffunction-calling.html","Function Calling",[79,3112,3113,3114,293,3118,293,3120,293,3124],{},"模型选型：",[44,3115,3117],{"href":3116},"\u002Fmodels\u002Fgpt-5.html","GPT-5",[44,3119,882],{"href":881},[44,3121,3123],{"href":3122},"\u002Fmodels\u002Fdeepseek-v3.html","DeepSeek-V3",[44,3125,3127],{"href":3126},"\u002Fmodels\u002Fglm-5.2.html","GLM-5.2",[79,3129,893,3130,293,3134],{},[44,3131,3133],{"href":3132},"\u002Fwiki\u002Ffine-tuning-vs-rag.html","Fine-tuning vs RAG",[44,3135,897],{"href":896},[20,3137,904],{"id":904},[76,3139,3140,3146,3152,3158,3165],{},[79,3141,909,3142],{},[44,3143,3144],{"href":3144,"rel":3145},"https:\u002F\u002Fdify.ai",[48],[79,3147,3148,3149],{},"中文文档：",[44,3150,2985],{"href":2985,"rel":3151},[48],[79,3153,3154,3155],{},"GitHub：",[44,3156,2466],{"href":2466,"rel":3157},[48],[79,3159,3160,3161],{},"官方定价：",[44,3162,3163],{"href":3163,"rel":3164},"https:\u002F\u002Fdify.ai\u002Fpricing",[48],[79,3166,3167],{},"第三方评测：tooljunction.io \u002F chatforest.com \u002F besthub.dev \u002F joshuaopolko.com \u002F 知乎 LLM 实战笔记",[32,3169,3170,3171,929],{},"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 ",[44,3172,928],{"href":927},[3174,3175,3176],"style",{},"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":448,"searchDepth":931,"depth":931,"links":3178},[3179,3180,3187,3192,3193,3194,3195,3196,3197,3198],{"id":22,"depth":934,"text":23},{"id":143,"depth":934,"text":143,"children":3181},[3182,3183,3184,3185,3186],{"id":2215,"depth":931,"text":2216},{"id":2293,"depth":931,"text":2294},{"id":2332,"depth":931,"text":2333},{"id":2350,"depth":931,"text":2351},{"id":2373,"depth":931,"text":2374},{"id":321,"depth":934,"text":321,"children":3188},[3189,3190,3191],{"id":2382,"depth":931,"text":2383},{"id":2460,"depth":931,"text":2461},{"id":2558,"depth":931,"text":2559},{"id":2630,"depth":934,"text":2631},{"id":466,"depth":934,"text":466},{"id":509,"depth":934,"text":509},{"id":734,"depth":934,"text":734},{"id":803,"depth":934,"text":804},{"id":851,"depth":934,"text":851},{"id":904,"depth":934,"text":904},"platform","\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。",[961,962,3203],"ja",{},"\u002Ftools\u002Fagent\u002Fplatform\u002Fdify",[979,980,981,2517],[3208,3212,3216,3220],{"plan":3209,"price":352,"features":3210,"notes":3211},"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":3213,"price":352,"features":3214,"notes":3215},"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":3217,"price":2427,"features":3218,"notes":3219},"Cloud Professional","更高调用额度 + 团队协作 + 商用支持","商用云首选",{"plan":3221,"price":1754,"features":3222,"notes":2447},"Cloud Team \u002F Enterprise","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",[3225,3226,3227,3228],"coze-deep-review","coze-vs-dify","dify-deep-review","fastgpt-deep-review",{"power":1007,"ux":953,"price":1007,"cn_support":953,"stability":953},{"title":2171,"description":3201},"Dify 评测 2026：开源 LLMOps 与 AI Agent 平台，自托管指南",[3233,3235,3237,3239,3241],{"title":3234,"url":2985},"Dify 官方文档（中文）",{"title":3236,"url":2466},"Dify GitHub",{"title":3238,"url":3163},"Dify 官方定价",{"title":3240,"url":2903},"Coze vs Dify vs FastGPT 选型",{"title":3242,"url":3243},"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",[3247,3248,3249,3250,3251,3252,3253],"agent-platform","opensource","self-host","rag","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",1785660639073]