[{"data":1,"prerenderedAt":1715},["ShallowReactive",2],{"header-counts":3,"footer-counts":6,"compare-dify-vs-ragflow":8,"compare-a-dify":9,"compare-b-ragflow":1170},{"tools":4,"reviews":5},98,27,{"tools":4,"reviews":5,"playbooks":7,"news":5},22,null,{"id":10,"title":11,"alternatives":12,"api_compatible":8,"body":17,"category":1101,"chinese_friendly":385,"cover":1102,"description":1103,"domestic":1104,"extension":1105,"faq":8,"free":1104,"github":331,"languages":1106,"lastVerified":8,"meta":1110,"models":8,"navigation":552,"notSuitable":8,"opensource":552,"path":1111,"pillar":1112,"platforms":1113,"priceTable":1117,"pricing":1134,"published":1135,"relatedPlaybooks":8,"relatedReviews":1136,"score":1141,"self_host":552,"seo":1142,"seoTitle":1143,"slug":1144,"sources":1145,"stem":1157,"suitable":8,"tagline":1158,"tags":1159,"updated":1167,"verdict":1168,"website":1043,"__hash__":1169},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Fdify.md","Dify",[13,14,15,16],"agent\u002Fplatform\u002Fcoze","agent\u002Fplatform\u002Ffastgpt","agent\u002Fplatform\u002Fn8n","agent\u002Fplatform\u002Flangflow",{"type":18,"value":19,"toc":1079},"minimark",[20,25,53,59,62,67,70,145,148,152,155,171,190,194,197,208,212,220,232,236,239,242,246,255,316,323,327,335,405,408,428,432,435,495,501,505,600,603,632,635,776,794,833,836,909,913,916,936,939,968,971,1033,1036,1067,1075],[21,22,24],"h2",{"id":23},"tldr","TL;DR",[26,27,32,40],"div",{"className":28},[29,30,31],"card","p-5","my-4",[33,34,35,39],"p",{},[36,37,38],"strong",{},"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。",[33,41,42,43,46,47,52],{},"最大价值是 ",[36,44,45],{},"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 ",[48,49,51],"a",{"href":50},"\u002Fagent\u002Fplatform\u002Fcoze.html","Coze"," 折腾，新手得读 1-2 小时文档。",[54,55,56],"blockquote",{},[33,57,58],{},"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。",[21,60,61],{"id":61},"核心特性",[63,64,66],"h3",{"id":65},"可视化工作流chatflow-workflow","可视化工作流（Chatflow + Workflow）",[33,68,69],{},"Dify 把 LLM 应用拆成两种\"应用类型\"：",[71,72,73,89],"table",{},[74,75,76],"thead",{},[77,78,79,83,86],"tr",{},[80,81,82],"th",{},"类型",[80,84,85],{},"适合场景",[80,87,88],{},"编排范式",[90,91,92,106,119,132],"tbody",{},[77,93,94,100,103],{},[95,96,97],"td",{},[36,98,99],{},"Chatbot",[95,101,102],{},"简单对话机器人",[95,104,105],{},"prompt + tools",[77,107,108,113,116],{},[95,109,110],{},[36,111,112],{},"Agent",[95,114,115],{},"自主多步任务",[95,117,118],{},"ReAct \u002F Function Calling",[77,120,121,126,129],{},[95,122,123],{},[36,124,125],{},"Chatflow",[95,127,128],{},"对话型工作流（多轮 + 分支）",[95,130,131],{},"节点 DAG，带聊天上下文",[77,133,134,139,142],{},[95,135,136],{},[36,137,138],{},"Workflow",[95,140,141],{},"单次输入→输出（API 模式）",[95,143,144],{},"节点 DAG，无对话状态",[33,146,147],{},"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\"。",[63,149,151],{"id":150},"rag-知识库","RAG 知识库",[33,153,154],{},"内置完整 RAG 链路：",[156,157,158,162,165,168],"ol",{},[159,160,161],"li",{},"上传文档（PDF \u002F Word \u002F Markdown \u002F 网页）",[159,163,164],{},"自动分块 + embedding（可配置分段策略和 embedding 模型）",[159,166,167],{},"混合检索（向量 + 全文 + 重排）",[159,169,170],{},"引用溯源（回答末尾自动附原文片段）",[33,172,173,174,180,181,184,185,189],{},"注意：根据 ",[48,175,179],{"href":176,"rel":177},"https:\u002F\u002Fzhuanlan.zhihu.com\u002Fp\u002F1887141987838309480",[178],"nofollow","知乎 LLM 实战笔记 2025-03 对比"," 的实测，Dify ",[36,182,183],{},"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 ",[48,186,188],{"href":187},"\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。",[63,191,193],{"id":192},"模型生态40-提供商","模型生态：40+ 提供商",[33,195,196],{},"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。",[33,198,199,200,202,203,207],{},"国产模型原生支持（不像 ",[48,201,188],{"href":187}," 需要 ",[48,204,206],{"href":205},"\u002Fcoding\u002Fapi\u002Fone-api.html","OneAPI"," 中转），是 Dify 在国内 toB 场景流行的关键。",[63,209,211],{"id":210},"mcp-协议支持","MCP 协议支持",[33,213,214,215,219],{},"Dify 较早接入了 ",[48,216,218],{"href":217},"\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：",[221,222,223,226,229],"ul",{},[159,224,225],{},"通过 MCP 调本地 PostgreSQL \u002F SQLite",[159,227,228],{},"通过 MCP 调 GitHub \u002F Slack \u002F Linear",[159,230,231],{},"通过 MCP 调自家内部系统（写一个 MCP Server 即可）",[63,233,235],{"id":234},"api-first","API-first",[33,237,238],{},"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。",[21,240,241],{"id":241},"价格与运行成本",[63,243,245],{"id":244},"云版difyai","云版（dify.ai）",[33,247,248,249,254],{},"根据 ",[48,250,253],{"href":251,"rel":252},"https:\u002F\u002Fwww.tooljunction.io\u002Fai-tools\u002Fdify-ai",[178],"tooljunction.io 2026 评测"," 引用的官方定价：",[71,256,257,270],{},[74,258,259],{},[77,260,261,264,267],{},[80,262,263],{},"套餐",[80,265,266],{},"价格",[80,268,269],{},"主要限制",[90,271,272,283,294,305],{},[77,273,274,277,280],{},[95,275,276],{},"Sandbox",[95,278,279],{},"免费",[95,281,282],{},"200 次模型调用，1 app，5MB 知识库",[77,284,285,288,291],{},[95,286,287],{},"Professional",[95,289,290],{},"$59\u002F月起",[95,292,293],{},"5000 调用\u002F月，多 app，50MB 知识库",[77,295,296,299,302],{},[95,297,298],{},"Team",[95,300,301],{},"$159\u002F月起",[95,303,304],{},"团队协作、SSO",[77,306,307,310,313],{},[95,308,309],{},"Enterprise",[95,311,312],{},"联系销售",[95,314,315],{},"定制 SLA、私有云",[33,317,318,319,322],{},"注意：云版价格只是 Dify 平台费，",[36,320,321],{},"模型 API 费用另算","（自带 OpenAI \u002F Anthropic key）。",[63,324,326],{"id":325},"自托管推荐","自托管（推荐）",[33,328,329,334],{},[48,330,333],{"href":331,"rel":332},"https:\u002F\u002Fgithub.com\u002Flanggenius\u002Fdify",[178],"官方 GitHub 仓库"," 提供 Docker Compose 部署，社区版完全免费可商用：",[336,337,342],"pre",{"className":338,"code":339,"language":340,"meta":341,"style":341},"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","",[343,344,345,361,371,383,398],"code",{"__ignoreMap":341},[346,347,350,354,358],"span",{"class":348,"line":349},"line",1,[346,351,353],{"class":352},"sScJk","git",[346,355,357],{"class":356},"sZZnC"," clone",[346,359,360],{"class":356}," https:\u002F\u002Fgithub.com\u002Flanggenius\u002Fdify.git\n",[346,362,364,368],{"class":348,"line":363},2,[346,365,367],{"class":366},"sj4cs","cd",[346,369,370],{"class":356}," dify\u002Fdocker\n",[346,372,374,377,380],{"class":348,"line":373},3,[346,375,376],{"class":352},"cp",[346,378,379],{"class":356}," .env.example",[346,381,382],{"class":356}," .env\n",[346,384,386,389,392,395],{"class":348,"line":385},4,[346,387,388],{"class":352},"docker",[346,390,391],{"class":356}," compose",[346,393,394],{"class":356}," up",[346,396,397],{"class":366}," -d\n",[346,399,401],{"class":348,"line":400},5,[346,402,404],{"class":403},"sJ8bj","# 默认 http:\u002F\u002Flocalhost \u002F 端口可在 .env 调整\n",[33,406,407],{},"硬件门槛（社区共识，非官方硬性要求）：",[221,409,410,416,422],{},[159,411,412,415],{},[36,413,414],{},"最低","：2 核 4G，纯外接 API 模式",[159,417,418,421],{},[36,419,420],{},"推荐","：4 核 8G + 至少 30GB 磁盘（向量数据 + 文件存储）",[159,423,424,427],{},[36,425,426],{},"企业","：8 核 16G+，单机日活上千",[63,429,431],{"id":430},"真实-tco","真实 TCO",[33,433,434],{},"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）：",[71,436,437,450],{},[74,438,439],{},[77,440,441,444,447],{},[80,442,443],{},"成本项",[80,445,446],{},"云版 Professional",[80,448,449],{},"自托管",[90,451,452,463,473,484],{},[77,453,454,457,460],{},[95,455,456],{},"平台费",[95,458,459],{},"~$2,100（3 年）",[95,461,462],{},"$0",[77,464,465,468,470],{},[95,466,467],{},"服务器",[95,469,462],{},[95,471,472],{},"~$50\u002F月 × 36 = $1,800",[77,474,475,478,481],{},[95,476,477],{},"模型 API",[95,479,480],{},"与下同",[95,482,483],{},"与上同",[77,485,486,489,492],{},[95,487,488],{},"运维人力",[95,490,491],{},"0",[95,493,494],{},"约 0.2 人月",[33,496,497,500],{},[36,498,499],{},"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。",[21,502,504],{"id":503},"上手-10-分钟","上手 10 分钟",[336,506,508],{"className":338,"code":507,"language":340,"meta":341,"style":341},"# 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",[343,509,510,515,523,529,537,547,554,560,566,571,577,582,588,594],{"__ignoreMap":341},[346,511,512],{"class":348,"line":349},[346,513,514],{"class":403},"# 1. 自托管（社区版）\n",[346,516,517,519,521],{"class":348,"line":363},[346,518,353],{"class":352},[346,520,357],{"class":356},[346,522,360],{"class":356},[346,524,525,527],{"class":348,"line":373},[346,526,367],{"class":366},[346,528,370],{"class":356},[346,530,531,533,535],{"class":348,"line":385},[346,532,376],{"class":352},[346,534,379],{"class":356},[346,536,382],{"class":356},[346,538,539,541,543,545],{"class":348,"line":400},[346,540,388],{"class":352},[346,542,391],{"class":356},[346,544,394],{"class":356},[346,546,397],{"class":366},[346,548,550],{"class":348,"line":549},6,[346,551,553],{"emptyLinePlaceholder":552},true,"\n",[346,555,557],{"class":348,"line":556},7,[346,558,559],{"class":403},"# 2. 浏览器打开 http:\u002F\u002Flocalhost\n",[346,561,563],{"class":348,"line":562},8,[346,564,565],{"class":403},"#    首次会让你创建 admin 账号\n",[346,567,569],{"class":348,"line":568},9,[346,570,553],{"emptyLinePlaceholder":552},[346,572,574],{"class":348,"line":573},10,[346,575,576],{"class":403},"# 3. 进入\"设置 → 模型供应商\"，配置 OpenAI \u002F 国产模型 API key\n",[346,578,580],{"class":348,"line":579},11,[346,581,553],{"emptyLinePlaceholder":552},[346,583,585],{"class":348,"line":584},12,[346,586,587],{"class":403},"# 4. 在主界面\"创建空白应用\"，选 Chatflow 或 Workflow\n",[346,589,591],{"class":348,"line":590},13,[346,592,593],{"class":403},"# 5. 拖入\"开始 → LLM → 结束\"节点试一下基础 prompt\n",[346,595,597],{"class":348,"line":596},14,[346,598,599],{"class":403},"# 6. 满意了点右上\"发布\"，自动生成 API endpoint\n",[21,601,602],{"id":602},"国内使用注意事项",[156,604,605,611,617,623],{},[159,606,607,610],{},[36,608,609],{},"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",[159,612,613,616],{},[36,614,615],{},"自托管 + 国产模型"," = 完全国内闭环，是 Dify 在国内最大优势",[159,618,619,622],{},[36,620,621],{},"Docker 镜像拉取","：国内可能慢，建议配 Docker registry 镜像（阿里云 \u002F 网易）",[159,624,625,628,629,631],{},[36,626,627],{},"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 ",[48,630,51],{"href":50}," 迁到 Dify",[21,633,634],{"id":634},"与同类怎么选",[71,636,637,660],{},[74,638,639],{},[77,640,641,644,646,650,654],{},[80,642,643],{},"维度",[80,645,11],{},[80,647,648],{},[48,649,51],{"href":50},[80,651,652],{},[48,653,188],{"href":187},[80,655,656],{},[48,657,659],{"href":658},"\u002Fagent\u002Fplatform\u002Fn8n.html","n8n",[90,661,662,678,691,707,721,735,749,762],{},[77,663,664,667,670,673,675],{},[95,665,666],{},"开源",[95,668,669],{},"✅",[95,671,672],{},"❌",[95,674,669],{},[95,676,677],{},"✅（fair-code）",[77,679,680,683,685,687,689],{},[95,681,682],{},"私有部署",[95,684,669],{},[95,686,672],{},[95,688,669],{},[95,690,669],{},[77,692,693,696,699,702,704],{},[95,694,695],{},"上手难度",[95,697,698],{},"★★★☆☆",[95,700,701],{},"★★☆☆☆ 最简单",[95,703,698],{},[95,705,706],{},"★★★★☆",[77,708,709,712,715,717,719],{},[95,710,711],{},"工作流编排",[95,713,714],{},"★★★★★",[95,716,706],{},[95,718,698],{},[95,720,714],{},[77,722,723,726,728,730,732],{},[95,724,725],{},"RAG 精度",[95,727,706],{},[95,729,698],{},[95,731,714],{},[95,733,734],{},"★★☆☆☆",[77,736,737,740,742,744,747],{},[95,738,739],{},"模型生态",[95,741,714],{},[95,743,706],{},[95,745,746],{},"★★★☆☆（OneAPI 中转）",[95,748,706],{},[77,750,751,754,756,758,760],{},[95,752,753],{},"中文场景",[95,755,706],{},[95,757,714],{},[95,759,706],{},[95,761,698],{},[77,763,764,767,769,772,774],{},[95,765,766],{},"字节生态绑定",[95,768,672],{},[95,770,771],{},"✅（飞书\u002F抖音深度集成）",[95,773,672],{},[95,775,672],{},[33,777,778,781,782,787,788,793],{},[36,779,780],{},"怎么选","（基于 ",[48,783,786],{"href":784,"rel":785},"https:\u002F\u002Fwww.besthub.dev\u002Farticles\u002Fcoze-vs-dify-vs-fastgpt-which-ai-agent-platform-fits-your-needs-fa59cf97b798",[178],"BestHub 2025-07"," 和 ",[48,789,792],{"href":790,"rel":791},"https:\u002F\u002Fwww.cnblogs.com\u002Fuulucias\u002Fp\u002F19449008",[178],"博客园 2026-01"," 两份选型指南综合）：",[221,795,796,802,810,817,824],{},[159,797,798,801],{},[36,799,800],{},"数据必须不出内网 + 工作流复杂"," → Dify",[159,803,804,807,808],{},[36,805,806],{},"个人 \u002F 小团队 \u002F 快速原型 + 字节生态"," → ",[48,809,51],{"href":50},[159,811,812,807,815],{},[36,813,814],{},"核心场景就是企业知识库 QA",[48,816,188],{"href":187},[159,818,819,807,822],{},[36,820,821],{},"重点是连接外部 SaaS（Slack \u002F Notion \u002F 数据库）",[48,823,659],{"href":658},[159,825,826,807,829],{},[36,827,828],{},"要画图式表达 LangChain pipeline",[48,830,832],{"href":831},"\u002Fagent\u002Fplatform\u002Flangflow.html","Langflow",[21,834,835],{"id":835},"避坑清单",[221,837,838,844,861,872,885,891,897,903],{},[159,839,840,843],{},[36,841,842],{},"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。",[159,845,846,852,853,856,857,860],{},[36,847,848,851],{},[343,849,850],{},".env"," 文件改完忘 restart","：",[343,854,855],{},"docker compose down && up -d","，不是 ",[343,858,859],{},"restart","——后者不重新加载 env。",[159,862,863,852,866,871],{},[36,864,865],{},"大版本升级会破坏数据库 schema",[48,867,870],{"href":868,"rel":869},"https:\u002F\u002Fdocs.dify.ai\u002Fzh-hans",[178],"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。",[159,873,874,877,878,880,881,884],{},[36,875,876],{},"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 ",[343,879,850],{}," 的 ",[343,882,883],{},"UPLOAD_FILE_SIZE_LIMIT"," 并重启容器。",[159,886,887,890],{},[36,888,889],{},"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。",[159,892,893,896],{},[36,894,895],{},"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。",[159,898,899,902],{},[36,900,901],{},"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。",[159,904,905,908],{},[36,906,907],{},"国内 Docker 拉取镜像慢","：先配国内 registry，否则首次 pull 可能要 30+ 分钟。",[21,910,912],{"id":911},"适合-不适合","适合 \u002F 不适合",[33,914,915],{},"✅ 适合：",[221,917,918,921,924,927,930,933],{},[159,919,920],{},"中大型企业 LLM 中台建设",[159,922,923],{},"需要私有化部署（金融 \u002F 医疗 \u002F 政府）",[159,925,926],{},"想做\"AI 工作流即产品\"的开发团队",[159,928,929],{},"同时需要 RAG + Agent + Workflow 三件套",[159,931,932],{},"想用国产模型 + 国际模型混合编排",[159,934,935],{},"已经接受 Docker + 一定运维投入",[33,937,938],{},"❌ 不适合：",[221,940,941,947,953,956,962],{},[159,942,943,944,946],{},"纯个人玩家做对话机器人（",[48,945,51],{"href":50}," 更快）",[159,948,949,950,952],{},"只想做企业知识库 QA（",[48,951,188],{"href":187}," RAG 更专）",[159,954,955],{},"团队完全没运维能力（云版还行，自托管会踩坑）",[159,957,958,959,961],{},"需要深度对接字节飞书 \u002F 抖音（",[48,960,51],{"href":50}," 原生）",[159,963,964,965,967],{},"工作流核心是连接 100+ SaaS（",[48,966,659],{"href":658}," 节点更全）",[21,969,970],{"id":970},"相关阅读",[221,972,973,985,1003,1022],{},[159,974,975,976,978,979,978,981,978,983],{},"同类对比：",[48,977,51],{"href":50}," \u002F ",[48,980,188],{"href":187},[48,982,659],{"href":658},[48,984,832],{"href":831},[159,986,987,988,978,992,978,996,978,999],{},"概念基础：",[48,989,991],{"href":990},"\u002Fwiki\u002Fai-agent.html","AI Agent",[48,993,995],{"href":994},"\u002Fwiki\u002Frag.html","RAG",[48,997,998],{"href":217},"MCP",[48,1000,1002],{"href":1001},"\u002Fwiki\u002Ffunction-calling.html","Function Calling",[159,1004,1005,1006,978,1010,978,1014,978,1018],{},"模型选型：",[48,1007,1009],{"href":1008},"\u002Fmodels\u002Fgpt-5.html","GPT-5",[48,1011,1013],{"href":1012},"\u002Fmodels\u002Fclaude-sonnet-4.html","Claude Sonnet 4",[48,1015,1017],{"href":1016},"\u002Fmodels\u002Fdeepseek-v3.html","DeepSeek-V3",[48,1019,1021],{"href":1020},"\u002Fmodels\u002Fglm-5.2.html","GLM-5.2",[159,1023,1024,1025,978,1029],{},"进阶：",[48,1026,1028],{"href":1027},"\u002Fwiki\u002Ffine-tuning-vs-rag.html","Fine-tuning vs RAG",[48,1030,1032],{"href":1031},"\u002Fwiki\u002Fcontext-engineering.html","Context Engineering",[21,1034,1035],{"id":1035},"来源",[221,1037,1038,1045,1051,1057,1064],{},[159,1039,1040,1041],{},"官网：",[48,1042,1043],{"href":1043,"rel":1044},"https:\u002F\u002Fdify.ai",[178],[159,1046,1047,1048],{},"中文文档：",[48,1049,868],{"href":868,"rel":1050},[178],[159,1052,1053,1054],{},"GitHub：",[48,1055,331],{"href":331,"rel":1056},[178],[159,1058,1059,1060],{},"官方定价：",[48,1061,1062],{"href":1062,"rel":1063},"https:\u002F\u002Fdify.ai\u002Fpricing",[178],[159,1065,1066],{},"第三方评测：tooljunction.io \u002F chatforest.com \u002F besthub.dev \u002F joshuaopolko.com \u002F 知乎 LLM 实战笔记",[33,1068,1069,1070,1074],{},"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 ",[48,1071,1073],{"href":1072},"mailto:hello@aiho.net","反馈邮箱"," 反馈。",[1076,1077,1078],"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":341,"searchDepth":373,"depth":373,"links":1080},[1081,1082,1089,1094,1095,1096,1097,1098,1099,1100],{"id":23,"depth":363,"text":24},{"id":61,"depth":363,"text":61,"children":1083},[1084,1085,1086,1087,1088],{"id":65,"depth":373,"text":66},{"id":150,"depth":373,"text":151},{"id":192,"depth":373,"text":193},{"id":210,"depth":373,"text":211},{"id":234,"depth":373,"text":235},{"id":241,"depth":363,"text":241,"children":1090},[1091,1092,1093],{"id":244,"depth":373,"text":245},{"id":325,"depth":373,"text":326},{"id":430,"depth":373,"text":431},{"id":503,"depth":363,"text":504},{"id":602,"depth":363,"text":602},{"id":634,"depth":363,"text":634},{"id":835,"depth":363,"text":835},{"id":911,"depth":363,"text":912},{"id":970,"depth":363,"text":970},{"id":1035,"depth":363,"text":1035},"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。",false,"md",[1107,1108,1109],"zh","en","ja",{},"\u002Ftools\u002Fagent\u002Fplatform\u002Fdify","agent",[1114,1115,1116,388],"windows","macos","linux",[1118,1122,1126,1130],{"plan":1119,"price":462,"features":1120,"notes":1121},"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":1123,"price":462,"features":1124,"notes":1125},"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":1127,"price":290,"features":1128,"notes":1129},"Cloud Professional","更高调用额度 + 团队协作 + 商用支持","商用云首选",{"plan":1131,"price":1132,"features":1133,"notes":312},"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",[1137,1138,1139,1140],"coze-deep-review","coze-vs-dify","dify-deep-review","fastgpt-deep-review",{"power":400,"ux":385,"price":400,"cn_support":385,"stability":385},{"title":11,"description":1103},"Dify 评测 2026：开源 LLMOps 与 AI Agent 平台，自托管指南","agent\u002Fplatform\u002Fdify",[1146,1148,1150,1152,1154],{"title":1147,"url":868},"Dify 官方文档（中文）",{"title":1149,"url":331},"Dify GitHub",{"title":1151,"url":1062},"Dify 官方定价",{"title":1153,"url":784},"Coze vs Dify vs FastGPT 选型",{"title":1155,"url":1156},"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",[1160,1161,1162,1163,1164,1165,1166],"agent-platform","opensource","self-host","rag","workflow","llmops","mcp","2026-06-24","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。","p5aiXfjt5rD0m3qxj903DVwZMIONVWKdagLa7niYhcE",{"id":1171,"title":1172,"alternatives":1173,"api_compatible":8,"body":1175,"category":1101,"chinese_friendly":385,"cover":1693,"description":1694,"domestic":1104,"extension":1105,"faq":8,"free":1104,"github":1677,"languages":1695,"lastVerified":1696,"meta":1697,"models":8,"navigation":552,"notSuitable":8,"opensource":552,"path":1698,"pillar":1112,"platforms":1699,"priceTable":8,"pricing":1700,"published":1701,"relatedPlaybooks":8,"relatedReviews":8,"score":1702,"self_host":1104,"seo":1703,"seoTitle":1704,"slug":1705,"sources":1706,"stem":1709,"suitable":8,"tagline":1710,"tags":1711,"updated":1696,"verdict":1713,"website":1670,"__hash__":1714},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Fragflow.md","RAGFlow",[1144,14,1174],"agent\u002Fplatform\u002Fanythingllm",{"type":18,"value":1176,"toc":1680},[1177,1179,1182,1185,1188,1244,1246,1292,1297,1301,1309,1329,1334,1357,1360,1390,1393,1528,1531,1579,1581,1613,1617,1623,1629,1635,1641,1643,1657,1659,1664],[21,1178,24],{"id":23},[33,1180,1181],{},"RAGFlow 是 InfiniFlow（中国团队）出品的开源 RAG 引擎（Apache 2.0），核心卖点是深度文档解析 + 高召回率切片 + 引用溯源。支持 PDF \u002F Word \u002F Excel \u002F PPT \u002F 图片 \u002F 扫描件，内置 OCR + 版面分析 + 表格识别，切片质量远超通用 RAG 方案。Docker 自托管 + Cloud 云端，中文支持好（文档 \u002F UI \u002F 社区）。",[33,1183,1184],{},"适合：需要精准文档问答的企业知识库、复杂文档（表格 \u002F 图文 \u002F 扫描件）场景、中文 RAG 需求、对召回率要求高的业务。不适合：需要复杂 Agent 编排（用 Dify）、资源有限的小服务器、需要精美 UI 的 C 端产品。",[21,1186,1187],{"id":1187},"核心能力",[221,1189,1190,1196,1202,1208,1214,1220,1226,1232,1238],{},[159,1191,1192,1195],{},[36,1193,1194],{},"深度文档解析","：PDF \u002F Word \u002F Excel \u002F PPT \u002F 图片 \u002F 扫描件，版面分析 + 表格识别",[159,1197,1198,1201],{},[36,1199,1200],{},"OCR 引擎","：内置 PaddleOCR \u002F DeepDOC，支持中英文扫描件识别",[159,1203,1204,1207],{},[36,1205,1206],{},"智能切片","：基于版面分析的语义切片，保留段落 \u002F 表格 \u002F 标题结构",[159,1209,1210,1213],{},[36,1211,1212],{},"高召回率","：混合检索（全文 + 向量）+ 重排序（Rerank），召回精度高",[159,1215,1216,1219],{},[36,1217,1218],{},"引用溯源","：回答标注来源文档 + 页码 + 原文片段，可验证",[159,1221,1222,1225],{},[36,1223,1224],{},"多模型接入","：OpenAI \u002F Claude \u002F Ollama \u002F 通义千问 \u002F 智谱 \u002F 月之暗面",[159,1227,1228,1231],{},[36,1229,1230],{},"多向量数据库","：Elasticsearch \u002F Infinity（自研）\u002F Chroma",[159,1233,1234,1237],{},[36,1235,1236],{},"知识库管理","：多知识库 + 文档分类 + 解析状态监控",[159,1239,1240,1243],{},[36,1241,1242],{},"API 接口","：完整 REST API + SDK，可集成到外部系统",[21,1245,266],{"id":266},[71,1247,1248,1260],{},[74,1249,1250],{},[77,1251,1252,1255,1257],{},[80,1253,1254],{},"方案",[80,1256,266],{},[80,1258,1259],{},"核心功能",[90,1261,1262,1272,1283],{},[77,1263,1264,1267,1269],{},[95,1265,1266],{},"开源版",[95,1268,462],{},[95,1270,1271],{},"完整功能，Apache 2.0，自托管",[77,1273,1274,1277,1280],{},[95,1275,1276],{},"Cloud",[95,1278,1279],{},"按量付费",[95,1281,1282],{},"托管服务，免运维",[77,1284,1285,1287,1289],{},[95,1286,309],{},[95,1288,312],{},[95,1290,1291],{},"私有部署 + 技术支持 + 定制",[54,1293,1294],{},[33,1295,1296],{},"价格信息基于 2026-07 官网，可能调整。",[21,1298,1300],{"id":1299},"体验与评测资料整理","体验与评测（资料整理）",[54,1302,1303],{},[33,1304,1305,1306],{},"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",[36,1307,1308],{},"亮点：",[221,1310,1311,1314,1317,1320,1323,1326],{},[159,1312,1313],{},"文档解析质量在开源 RAG 中最强——复杂表格、多栏排版、图文混排都能正确识别",[159,1315,1316],{},"扫描件 OCR 效果好，中文印刷体识别准确率高",[159,1318,1319],{},"引用溯源到页码 + 原文片段，回答可信度高",[159,1321,1322],{},"混合检索 + Rerank 召回精度明显优于纯向量检索",[159,1324,1325],{},"中国团队出品，中文文档和社区支持好，Issue 响应快",[159,1327,1328],{},"支持通义千问 \u002F 智谱 \u002F 月之暗面等国产模型，国内场景适配好",[33,1330,1331],{},[36,1332,1333],{},"踩坑：",[221,1335,1336,1339,1342,1345,1348,1351,1354],{},[159,1337,1338],{},"资源消耗大——Elasticsearch + Redis + MinIO + RAGFlow 本身，至少 16GB 内存",[159,1340,1341],{},"部署较重，Docker Compose 起来 5+ 容器，配置复杂",[159,1343,1344],{},"大文件解析慢——100 页 PDF 解析 + 切片可能 5-10 分钟",[159,1346,1347],{},"UI 仍有粗糙处，文档管理界面交互不够流畅",[159,1349,1350],{},"Agent 能力弱——RAG 问答是强项，复杂工具调用 \u002F 多步推理不如 Dify",[159,1352,1353],{},"解析失败的重试机制不完善，偶尔卡在 parsing 状态",[159,1355,1356],{},"版本迭代快，升级需注意数据迁移",[21,1358,1359],{"id":1359},"上手",[156,1361,1362,1365,1371,1377,1384,1387],{},[159,1363,1364],{},"系统准备：确保 16GB+ 内存 + Docker + Docker Compose",[159,1366,1367,1368],{},"克隆仓库：",[343,1369,1370],{},"git clone https:\u002F\u002Fgithub.com\u002Finfiniflow\u002Fragflow.git",[159,1372,1373,1374],{},"启动服务：",[343,1375,1376],{},"cd ragflow\u002Fdocker && docker compose up -d",[159,1378,1379,1380,1383],{},"访问 ",[343,1381,1382],{},"http:\u002F\u002Flocalhost:80","，注册管理员账号",[159,1385,1386],{},"配置模型：Settings → Model Providers 添加 LLM + Embedding + Rerank",[159,1388,1389],{},"创建知识库 → 上传文档 → 等待解析完成 → 开始问答",[21,1391,1392],{"id":1392},"对比",[71,1394,1395,1410],{},[74,1396,1397],{},[77,1398,1399,1401,1403,1405,1407],{},[80,1400,643],{},[80,1402,1172],{},[80,1404,11],{},[80,1406,188],{},[80,1408,1409],{},"AnythingLLM",[90,1411,1412,1429,1442,1457,1471,1484,1498,1512],{},[77,1413,1414,1417,1420,1423,1426],{},[95,1415,1416],{},"文档解析",[95,1418,1419],{},"✅ 最强",[95,1421,1422],{},"中",[95,1424,1425],{},"强",[95,1427,1428],{},"弱",[77,1430,1431,1434,1436,1438,1440],{},[95,1432,1433],{},"表格识别",[95,1435,669],{},[95,1437,672],{},[95,1439,669],{},[95,1441,672],{},[77,1443,1444,1447,1450,1452,1455],{},[95,1445,1446],{},"OCR",[95,1448,1449],{},"✅ 内置",[95,1451,672],{},[95,1453,1454],{},"需配置",[95,1456,1454],{},[77,1458,1459,1462,1465,1467,1469],{},[95,1460,1461],{},"召回精度",[95,1463,1464],{},"高",[95,1466,1464],{},[95,1468,1464],{},[95,1470,1422],{},[77,1472,1473,1475,1478,1480,1482],{},[95,1474,1218],{},[95,1476,1477],{},"✅ 页码+片段",[95,1479,669],{},[95,1481,669],{},[95,1483,669],{},[77,1485,1486,1489,1491,1493,1495],{},[95,1487,1488],{},"Agent 能力",[95,1490,1428],{},[95,1492,1425],{},[95,1494,1422],{},[95,1496,1497],{},"基础",[77,1499,1500,1503,1505,1507,1509],{},[95,1501,1502],{},"资源消耗",[95,1504,1464],{},[95,1506,1422],{},[95,1508,1422],{},[95,1510,1511],{},"低",[77,1513,1514,1517,1520,1523,1525],{},[95,1515,1516],{},"中文支持",[95,1518,1519],{},"✅ 优秀",[95,1521,1522],{},"好",[95,1524,1522],{},[95,1526,1527],{},"一般",[21,1529,1530],{"id":1530},"避坑",[221,1532,1533,1539,1549,1555,1561,1567,1573],{},[159,1534,1535,1538],{},[36,1536,1537],{},"资源一定要够","：低于 16GB 内存别部署，ES + Redis + MinIO 都吃内存",[159,1540,1541,1544,1545,1548],{},[36,1542,1543],{},"Elasticsearch 配置","：默认 JVM 堆偏小，大知识库调 ",[343,1546,1547],{},"ES_JAVA_OPTS"," 到 4-8GB",[159,1550,1551,1554],{},[36,1552,1553],{},"大文件拆分上传","：超过 100 页的 PDF 拆成小文件，解析更稳定",[159,1556,1557,1560],{},[36,1558,1559],{},"解析失败检查格式","：加密 PDF \u002F 损坏文件会卡住，上传前检查",[159,1562,1563,1566],{},[36,1564,1565],{},"Rerank 模型别省","：召回精度提升的关键，用 bge-reranker 或 Cohere Rerank",[159,1568,1569,1572],{},[36,1570,1571],{},"不要当 Agent 平台用","：RAG 问答是核心，复杂工具调用上 Dify",[159,1574,1575,1578],{},[36,1576,1577],{},"定期备份","：ES 数据 + MinIO 文件，升级前完整快照",[21,1580,912],{"id":911},[221,1582,1583,1586,1589,1592,1595,1598,1601,1604,1607,1610],{},[159,1584,1585],{},"✅ 需要精准文档问答的企业知识库",[159,1587,1588],{},"✅ 复杂文档（表格 \u002F 图文 \u002F 扫描件）RAG 场景",[159,1590,1591],{},"✅ 中文 RAG 需求（国产模型 + 中文 OCR）",[159,1593,1594],{},"✅ 对召回率和引用溯源要求高的业务",[159,1596,1597],{},"✅ 有运维能力的团队私有化部署",[159,1599,1600],{},"❌ 需要复杂 Agent 编排（用 Dify）",[159,1602,1603],{},"❌ 资源有限的小服务器（至少 16GB 内存）",[159,1605,1606],{},"❌ 需要精美 C 端 UI 的产品",[159,1608,1609],{},"❌ 无运维能力的团队（用 Cloud 版或 FastGPT）",[159,1611,1612],{},"❌ 纯英文简单文档场景（AnythingLLM 更轻量）",[21,1614,1616],{"id":1615},"faq","FAQ",[33,1618,1619,1622],{},[36,1620,1621],{},"Q: RAGFlow 和 Dify 怎么选？","\nA: RAGFlow 专注 RAG——文档解析 + 检索精度 + 引用溯源是核心强项，适合文档密集型知识库。Dify 是完整 AI 应用平台——工作流 + Agent + RAG + API 管理，功能更全。纯文档问答选 RAGFlow，构建 AI 应用选 Dify，两者也可配合使用。",[33,1624,1625,1628],{},[36,1626,1627],{},"Q: 部署需要什么配置？","\nA: 最低 16GB 内存 + 4 核 CPU + 50GB 磁盘。生产环境建议 32GB 内存 + 8 核 + SSD。Elasticsearch 是内存大户，知识库文档量大时 ES JVM 堆需 8GB+。如果资源有限，考虑用 Infinity（RAGFlow 自研向量库）替代 ES。",[33,1630,1631,1634],{},[36,1632,1633],{},"Q: 支持中文 OCR 吗？","\nA: 支持。内置 PaddleOCR + DeepDOC 引擎，中文印刷体识别准确率高。手写体效果一般，复杂背景的扫描件建议预处理（去噪 \u002F 矫正）后再上传。OCR 默认开启，可在解析模板中配置。",[33,1636,1637,1640],{},[36,1638,1639],{},"Q: 和 FastGPT 比 RAG 精度如何？","\nA: 两者 RAG 精度都属第一梯队。RAGFlow 的优势在文档解析——复杂表格、多栏版面、图文混排的识别更准确，切片质量更高。FastGPT 的优势在工作流编排和知识库管理 UI 更成熟。文档解析要求高选 RAGFlow，流程管理要求高选 FastGPT。",[21,1642,970],{"id":970},[33,1644,1645,1648,1649,1648,1653],{},[48,1646,1409],{"href":1647},"\u002Fagent\u002Fplatform\u002Fanythingllm.html"," · ",[48,1650,1652],{"href":1651},"\u002Fagent\u002Fplatform\u002Fflowise.html","Flowise",[48,1654,1656],{"href":1655},"\u002Fcoding\u002Fapi\u002Flangfuse.html","Langfuse",[21,1658,1035],{"id":1035},[54,1660,1661],{},[33,1662,1663],{},"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。",[221,1665,1666,1673],{},[159,1667,1668],{},[48,1669,1672],{"href":1670,"rel":1671},"https:\u002F\u002Fragflow.io",[178],"官网",[159,1674,1675],{},[48,1676,1679],{"href":1677,"rel":1678},"https:\u002F\u002Fgithub.com\u002Finfiniflow\u002Fragflow",[178],"GitHub",{"title":341,"searchDepth":373,"depth":373,"links":1681},[1682,1683,1684,1685,1686,1687,1688,1689,1690,1691,1692],{"id":23,"depth":363,"text":24},{"id":1187,"depth":363,"text":1187},{"id":266,"depth":363,"text":266},{"id":1299,"depth":363,"text":1300},{"id":1359,"depth":363,"text":1359},{"id":1392,"depth":363,"text":1392},{"id":1530,"depth":363,"text":1530},{"id":911,"depth":363,"text":912},{"id":1615,"depth":363,"text":1616},{"id":970,"depth":363,"text":970},{"id":1035,"depth":363,"text":1035},"\u002Fimg\u002Ftools\u002Fragflow.webp","RAGFlow 真实评测：InfiniFlow 出品的开源 RAG 引擎（Apache 2.0 协议），深度文档解析（PDF\u002FWord\u002FExcel\u002F图片）+ 高召回率切片 + 引用溯源。支持 Docker 自托管，适合需要精准文档问答和知识库检索的企业场景。",[1108,1107],"2026-07-30",{},"\u002Ftools\u002Fagent\u002Fplatform\u002Fragflow",[1116,388],"Free \u002F 开源（Apache 2.0）\u002F Cloud","2026-07-05",{"power":385,"ux":373,"price":400,"cn_support":385,"stability":373},{"title":1172,"description":1694},"RAGFlow - 开源 RAG 引擎评测与部署 | AIHO","agent\u002Fplatform\u002Fragflow",[1707,1708],{"title":1672,"url":1670},{"title":1679,"url":1677},"tools\u002Fagent\u002Fplatform\u002Fragflow","开源 RAG 引擎，深度文档解析 + 高召回率",[1160,1161,1163,1712,1162],"document-parsing","需要精准文档解析和高召回率 RAG 的企业场景首选，深度文档解析（复杂表格\u002F版面\u002FOCR）+ 引用溯源能力在开源 RAG 引擎中最强，中国团队出品中文支持好，但部署资源要求高、Agent 能力弱、UI 仍需打磨。","bLpjRG4MMrBFYyeFSz4XzLMwVql0SyGD3sB3Tr5g38g",1785428441137]