[{"data":1,"prerenderedAt":1715},["ShallowReactive",2],{"header-counts":3,"footer-counts":6,"compare-ragflow-vs-dify":8,"compare-a-ragflow":9,"compare-b-dify":606},{"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":16,"category":569,"chinese_friendly":570,"cover":571,"description":572,"domestic":573,"extension":574,"faq":8,"free":573,"github":550,"languages":575,"lastVerified":578,"meta":579,"models":8,"navigation":580,"notSuitable":8,"opensource":580,"path":581,"pillar":582,"platforms":583,"priceTable":8,"pricing":586,"published":587,"relatedPlaybooks":8,"relatedReviews":8,"score":588,"self_host":573,"seo":590,"seoTitle":591,"slug":592,"sources":593,"stem":596,"suitable":8,"tagline":597,"tags":598,"updated":578,"verdict":604,"website":542,"__hash__":605},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Fragflow.md","RAGFlow",[13,14,15],"agent\u002Fplatform\u002Fdify","agent\u002Fplatform\u002Ffastgpt","agent\u002Fplatform\u002Fanythingllm",{"type":17,"value":18,"toc":553},"minimark",[19,24,28,31,34,93,96,151,157,161,169,189,194,217,220,252,255,395,398,446,450,482,486,492,498,504,510,513,528,531,536],[20,21,23],"h2",{"id":22},"tldr","TL;DR",[25,26,27],"p",{},"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）。",[25,29,30],{},"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。",[20,32,33],{"id":33},"核心能力",[35,36,37,45,51,57,63,69,75,81,87],"ul",{},[38,39,40,44],"li",{},[41,42,43],"strong",{},"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",[38,46,47,50],{},[41,48,49],{},"OCR 引擎","：内置 PaddleOCR \u002F DeepDOC，支持中英文扫描件识别",[38,52,53,56],{},[41,54,55],{},"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",[38,58,59,62],{},[41,60,61],{},"高召回率","：混合检索（全文 + 向量）+ 重排序（Rerank），召回精度高",[38,64,65,68],{},[41,66,67],{},"引用溯源","：回答标注来源文档 + 页码 + 原文片段，可验证",[38,70,71,74],{},[41,72,73],{},"多模型接入","：OpenAI \u002F Claude \u002F Ollama \u002F 通义千问 \u002F 智谱 \u002F 月之暗面",[38,76,77,80],{},[41,78,79],{},"多向量数据库","：Elasticsearch \u002F Infinity（自研）\u002F Chroma",[38,82,83,86],{},[41,84,85],{},"知识库管理","：多知识库 + 文档分类 + 解析状态监控",[38,88,89,92],{},[41,90,91],{},"API 接口","：完整 REST API + SDK，可集成到外部系统",[20,94,95],{"id":95},"价格",[97,98,99,114],"table",{},[100,101,102],"thead",{},[103,104,105,109,111],"tr",{},[106,107,108],"th",{},"方案",[106,110,95],{},[106,112,113],{},"核心功能",[115,116,117,129,140],"tbody",{},[103,118,119,123,126],{},[120,121,122],"td",{},"开源版",[120,124,125],{},"$0",[120,127,128],{},"完整功能，Apache 2.0，自托管",[103,130,131,134,137],{},[120,132,133],{},"Cloud",[120,135,136],{},"按量付费",[120,138,139],{},"托管服务，免运维",[103,141,142,145,148],{},[120,143,144],{},"Enterprise",[120,146,147],{},"联系销售",[120,149,150],{},"私有部署 + 技术支持 + 定制",[152,153,154],"blockquote",{},[25,155,156],{},"价格信息基于 2026-07 官网，可能调整。",[20,158,160],{"id":159},"体验与评测资料整理","体验与评测（资料整理）",[152,162,163],{},[25,164,165,166],{},"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",[41,167,168],{},"亮点：",[35,170,171,174,177,180,183,186],{},[38,172,173],{},"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",[38,175,176],{},"扫描件 OCR 效果好，中文印刷体识别准确率高",[38,178,179],{},"引用溯源到页码 + 原文片段，回答可信度高",[38,181,182],{},"混合检索 + Rerank 召回精度明显优于纯向量检索",[38,184,185],{},"中国团队出品，中文文档和社区支持好，Issue 响应快",[38,187,188],{},"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",[25,190,191],{},[41,192,193],{},"踩坑：",[35,195,196,199,202,205,208,211,214],{},[38,197,198],{},"资源消耗大——Elasticsearch + Redis + MinIO + RAGFlow 本身，至少 16GB 内存",[38,200,201],{},"部署较重，Docker Compose 起来 5+ 容器，配置复杂",[38,203,204],{},"大文件解析慢——100 页 PDF 解析 + 切片可能 5-10 分钟",[38,206,207],{},"UI 仍有粗糙处，文档管理界面交互不够流畅",[38,209,210],{},"Agent 能力弱——RAG 问答是强项，复杂工具调用 \u002F 多步推理不如 Dify",[38,212,213],{},"解析失败的重试机制不完善，偶尔卡在 parsing 状态",[38,215,216],{},"版本迭代快，升级需注意数据迁移",[20,218,219],{"id":219},"上手",[221,222,223,226,233,239,246,249],"ol",{},[38,224,225],{},"系统准备：确保 16GB+ 内存 + Docker + Docker Compose",[38,227,228,229],{},"克隆仓库：",[230,231,232],"code",{},"git clone https:\u002F\u002Fgithub.com\u002Finfiniflow\u002Fragflow.git",[38,234,235,236],{},"启动服务：",[230,237,238],{},"cd ragflow\u002Fdocker && docker compose up -d",[38,240,241,242,245],{},"访问 ",[230,243,244],{},"http:\u002F\u002Flocalhost:80","，注册管理员账号",[38,247,248],{},"配置模型：Settings → Model Providers 添加 LLM + Embedding + Rerank",[38,250,251],{},"创建知识库 → 上传文档 → 等待解析完成 → 开始问答",[20,253,254],{"id":254},"对比",[97,256,257,275],{},[100,258,259],{},[103,260,261,264,266,269,272],{},[106,262,263],{},"维度",[106,265,11],{},[106,267,268],{},"Dify",[106,270,271],{},"FastGPT",[106,273,274],{},"AnythingLLM",[115,276,277,294,309,324,338,351,365,379],{},[103,278,279,282,285,288,291],{},[120,280,281],{},"文档解析",[120,283,284],{},"✅ 最强",[120,286,287],{},"中",[120,289,290],{},"强",[120,292,293],{},"弱",[103,295,296,299,302,305,307],{},[120,297,298],{},"表格识别",[120,300,301],{},"✅",[120,303,304],{},"❌",[120,306,301],{},[120,308,304],{},[103,310,311,314,317,319,322],{},[120,312,313],{},"OCR",[120,315,316],{},"✅ 内置",[120,318,304],{},[120,320,321],{},"需配置",[120,323,321],{},[103,325,326,329,332,334,336],{},[120,327,328],{},"召回精度",[120,330,331],{},"高",[120,333,331],{},[120,335,331],{},[120,337,287],{},[103,339,340,342,345,347,349],{},[120,341,67],{},[120,343,344],{},"✅ 页码+片段",[120,346,301],{},[120,348,301],{},[120,350,301],{},[103,352,353,356,358,360,362],{},[120,354,355],{},"Agent 能力",[120,357,293],{},[120,359,290],{},[120,361,287],{},[120,363,364],{},"基础",[103,366,367,370,372,374,376],{},[120,368,369],{},"资源消耗",[120,371,331],{},[120,373,287],{},[120,375,287],{},[120,377,378],{},"低",[103,380,381,384,387,390,392],{},[120,382,383],{},"中文支持",[120,385,386],{},"✅ 优秀",[120,388,389],{},"好",[120,391,389],{},[120,393,394],{},"一般",[20,396,397],{"id":397},"避坑",[35,399,400,406,416,422,428,434,440],{},[38,401,402,405],{},[41,403,404],{},"资源一定要够","：低于 16GB 内存别部署，ES + Redis + MinIO 都吃内存",[38,407,408,411,412,415],{},[41,409,410],{},"Elasticsearch 配置","：默认 JVM 堆偏小，大知识库调 ",[230,413,414],{},"ES_JAVA_OPTS"," 到 4-8GB",[38,417,418,421],{},[41,419,420],{},"大文件拆分上传","：超过 100 页的 PDF 拆成小文件，解析更稳定",[38,423,424,427],{},[41,425,426],{},"解析失败检查格式","：加密 PDF \u002F 损坏文件会卡住，上传前检查",[38,429,430,433],{},[41,431,432],{},"Rerank 模型别省","：召回精度提升的关键，用 bge-reranker 或 Cohere Rerank",[38,435,436,439],{},[41,437,438],{},"不要当 Agent 平台用","：RAG 问答是核心，复杂工具调用上 Dify",[38,441,442,445],{},[41,443,444],{},"定期备份","：ES 数据 + MinIO 文件，升级前完整快照",[20,447,449],{"id":448},"适合-不适合","适合 \u002F 不适合",[35,451,452,455,458,461,464,467,470,473,476,479],{},[38,453,454],{},"✅ 需要精准文档问答的企业知识库",[38,456,457],{},"✅ 复杂文档（表格 \u002F 图文 \u002F 扫描件）RAG 场景",[38,459,460],{},"✅ 中文 RAG 需求（国产模型 + 中文 OCR）",[38,462,463],{},"✅ 对召回率和引用溯源要求高的业务",[38,465,466],{},"✅ 有运维能力的团队私有化部署",[38,468,469],{},"❌ 需要复杂 Agent 编排（用 Dify）",[38,471,472],{},"❌ 资源有限的小服务器（至少 16GB 内存）",[38,474,475],{},"❌ 需要精美 C 端 UI 的产品",[38,477,478],{},"❌ 无运维能力的团队（用 Cloud 版或 FastGPT）",[38,480,481],{},"❌ 纯英文简单文档场景（AnythingLLM 更轻量）",[20,483,485],{"id":484},"faq","FAQ",[25,487,488,491],{},[41,489,490],{},"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。",[25,493,494,497],{},[41,495,496],{},"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。",[25,499,500,503],{},[41,501,502],{},"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。",[25,505,506,509],{},[41,507,508],{},"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。",[20,511,512],{"id":512},"相关阅读",[25,514,515,519,520,519,524],{},[516,517,274],"a",{"href":518},"\u002Fagent\u002Fplatform\u002Fanythingllm.html"," · ",[516,521,523],{"href":522},"\u002Fagent\u002Fplatform\u002Fflowise.html","Flowise",[516,525,527],{"href":526},"\u002Fcoding\u002Fapi\u002Flangfuse.html","Langfuse",[20,529,530],{"id":530},"来源",[152,532,533],{},[25,534,535],{},"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。",[35,537,538,546],{},[38,539,540],{},[516,541,545],{"href":542,"rel":543},"https:\u002F\u002Fragflow.io",[544],"nofollow","官网",[38,547,548],{},[516,549,552],{"href":550,"rel":551},"https:\u002F\u002Fgithub.com\u002Finfiniflow\u002Fragflow",[544],"GitHub",{"title":554,"searchDepth":555,"depth":555,"links":556},"",3,[557,559,560,561,562,563,564,565,566,567,568],{"id":22,"depth":558,"text":23},2,{"id":33,"depth":558,"text":33},{"id":95,"depth":558,"text":95},{"id":159,"depth":558,"text":160},{"id":219,"depth":558,"text":219},{"id":254,"depth":558,"text":254},{"id":397,"depth":558,"text":397},{"id":448,"depth":558,"text":449},{"id":484,"depth":558,"text":485},{"id":512,"depth":558,"text":512},{"id":530,"depth":558,"text":530},"platform",4,"\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。",false,"md",[576,577],"en","zh","2026-07-30",{},true,"\u002Ftools\u002Fagent\u002Fplatform\u002Fragflow","agent",[584,585],"linux","docker","Free \u002F 开源（Apache 2.0）\u002F Cloud","2026-07-05",{"power":570,"ux":555,"price":589,"cn_support":570,"stability":555},5,{"title":11,"description":572},"RAGFlow - 开源 RAG 引擎评测与部署 | AIHO","agent\u002Fplatform\u002Fragflow",[594,595],{"title":545,"url":542},{"title":552,"url":550},"tools\u002Fagent\u002Fplatform\u002Fragflow","开源 RAG 引擎，深度文档解析 + 高召回率",[599,600,601,602,603],"agent-platform","opensource","rag","document-parsing","self-host","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",{"id":607,"title":268,"alternatives":608,"api_compatible":8,"body":612,"category":569,"chinese_friendly":570,"cover":1658,"description":1659,"domestic":573,"extension":574,"faq":8,"free":573,"github":904,"languages":1660,"lastVerified":8,"meta":1662,"models":8,"navigation":580,"notSuitable":8,"opensource":580,"path":1663,"pillar":582,"platforms":1664,"priceTable":1667,"pricing":1684,"published":1685,"relatedPlaybooks":8,"relatedReviews":1686,"score":1691,"self_host":580,"seo":1692,"seoTitle":1693,"slug":13,"sources":1694,"stem":1706,"suitable":8,"tagline":1707,"tags":1708,"updated":1712,"verdict":1713,"website":1600,"__hash__":1714},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Fdify.md",[609,14,610,611],"agent\u002Fplatform\u002Fcoze","agent\u002Fplatform\u002Fn8n","agent\u002Fplatform\u002Flangflow",{"type":17,"value":613,"toc":1636},[614,616,641,646,649,654,657,726,729,733,736,750,767,771,774,785,789,797,808,812,815,818,822,831,889,896,900,908,971,974,994,998,1001,1060,1066,1070,1164,1167,1196,1199,1337,1355,1394,1397,1470,1472,1475,1495,1498,1527,1529,1591,1593,1624,1632],[20,615,23],{"id":22},[617,618,623,629],"div",{"className":619},[620,621,622],"card","p-5","my-4",[25,624,625,628],{},[41,626,627],{},"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。",[25,630,631,632,635,636,640],{},"最大价值是 ",[41,633,634],{},"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 ",[516,637,639],{"href":638},"\u002Fagent\u002Fplatform\u002Fcoze.html","Coze"," 折腾，新手得读 1-2 小时文档。",[152,642,643],{},[25,644,645],{},"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。",[20,647,648],{"id":648},"核心特性",[650,651,653],"h3",{"id":652},"可视化工作流chatflow-workflow","可视化工作流（Chatflow + Workflow）",[25,655,656],{},"Dify 把 LLM 应用拆成两种\"应用类型\"：",[97,658,659,672],{},[100,660,661],{},[103,662,663,666,669],{},[106,664,665],{},"类型",[106,667,668],{},"适合场景",[106,670,671],{},"编排范式",[115,673,674,687,700,713],{},[103,675,676,681,684],{},[120,677,678],{},[41,679,680],{},"Chatbot",[120,682,683],{},"简单对话机器人",[120,685,686],{},"prompt + tools",[103,688,689,694,697],{},[120,690,691],{},[41,692,693],{},"Agent",[120,695,696],{},"自主多步任务",[120,698,699],{},"ReAct \u002F Function Calling",[103,701,702,707,710],{},[120,703,704],{},[41,705,706],{},"Chatflow",[120,708,709],{},"对话型工作流（多轮 + 分支）",[120,711,712],{},"节点 DAG，带聊天上下文",[103,714,715,720,723],{},[120,716,717],{},[41,718,719],{},"Workflow",[120,721,722],{},"单次输入→输出（API 模式）",[120,724,725],{},"节点 DAG，无对话状态",[25,727,728],{},"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\"。",[650,730,732],{"id":731},"rag-知识库","RAG 知识库",[25,734,735],{},"内置完整 RAG 链路：",[221,737,738,741,744,747],{},[38,739,740],{},"上传文档（PDF \u002F Word \u002F Markdown \u002F 网页）",[38,742,743],{},"自动分块 + embedding（可配置分段策略和 embedding 模型）",[38,745,746],{},"混合检索（向量 + 全文 + 重排）",[38,748,749],{},"引用溯源（回答末尾自动附原文片段）",[25,751,752,753,758,759,762,763,766],{},"注意：根据 ",[516,754,757],{"href":755,"rel":756},"https:\u002F\u002Fzhuanlan.zhihu.com\u002Fp\u002F1887141987838309480",[544],"知乎 LLM 实战笔记 2025-03 对比"," 的实测，Dify ",[41,760,761],{},"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 ",[516,764,271],{"href":765},"\u002Fagent\u002Fplatform\u002Ffastgpt.html","（实测准确率高 10+ 个百分点），Dify 胜在工作流而非纯 RAG。",[650,768,770],{"id":769},"模型生态40-提供商","模型生态：40+ 提供商",[25,772,773],{},"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。",[25,775,776,777,779,780,784],{},"国产模型原生支持（不像 ",[516,778,271],{"href":765}," 需要 ",[516,781,783],{"href":782},"\u002Fcoding\u002Fapi\u002Fone-api.html","OneAPI"," 中转），是 Dify 在国内 toB 场景流行的关键。",[650,786,788],{"id":787},"mcp-协议支持","MCP 协议支持",[25,790,791,792,796],{},"Dify 较早接入了 ",[516,793,795],{"href":794},"\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：",[35,798,799,802,805],{},[38,800,801],{},"通过 MCP 调本地 PostgreSQL \u002F SQLite",[38,803,804],{},"通过 MCP 调 GitHub \u002F Slack \u002F Linear",[38,806,807],{},"通过 MCP 调自家内部系统（写一个 MCP Server 即可）",[650,809,811],{"id":810},"api-first","API-first",[25,813,814],{},"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。",[20,816,817],{"id":817},"价格与运行成本",[650,819,821],{"id":820},"云版difyai","云版（dify.ai）",[25,823,824,825,830],{},"根据 ",[516,826,829],{"href":827,"rel":828},"https:\u002F\u002Fwww.tooljunction.io\u002Fai-tools\u002Fdify-ai",[544],"tooljunction.io 2026 评测"," 引用的官方定价：",[97,832,833,845],{},[100,834,835],{},[103,836,837,840,842],{},[106,838,839],{},"套餐",[106,841,95],{},[106,843,844],{},"主要限制",[115,846,847,858,869,880],{},[103,848,849,852,855],{},[120,850,851],{},"Sandbox",[120,853,854],{},"免费",[120,856,857],{},"200 次模型调用，1 app，5MB 知识库",[103,859,860,863,866],{},[120,861,862],{},"Professional",[120,864,865],{},"$59\u002F月起",[120,867,868],{},"5000 调用\u002F月，多 app，50MB 知识库",[103,870,871,874,877],{},[120,872,873],{},"Team",[120,875,876],{},"$159\u002F月起",[120,878,879],{},"团队协作、SSO",[103,881,882,884,886],{},[120,883,144],{},[120,885,147],{},[120,887,888],{},"定制 SLA、私有云",[25,890,891,892,895],{},"注意：云版价格只是 Dify 平台费，",[41,893,894],{},"模型 API 费用另算","（自带 OpenAI \u002F Anthropic key）。",[650,897,899],{"id":898},"自托管推荐","自托管（推荐）",[25,901,902,907],{},[516,903,906],{"href":904,"rel":905},"https:\u002F\u002Fgithub.com\u002Flanggenius\u002Fdify",[544],"官方 GitHub 仓库"," 提供 Docker Compose 部署，社区版完全免费可商用：",[909,910,914],"pre",{"className":911,"code":912,"language":913,"meta":554,"style":554},"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",[230,915,916,932,941,952,965],{"__ignoreMap":554},[917,918,921,925,929],"span",{"class":919,"line":920},"line",1,[917,922,924],{"class":923},"sScJk","git",[917,926,928],{"class":927},"sZZnC"," clone",[917,930,931],{"class":927}," https:\u002F\u002Fgithub.com\u002Flanggenius\u002Fdify.git\n",[917,933,934,938],{"class":919,"line":558},[917,935,937],{"class":936},"sj4cs","cd",[917,939,940],{"class":927}," dify\u002Fdocker\n",[917,942,943,946,949],{"class":919,"line":555},[917,944,945],{"class":923},"cp",[917,947,948],{"class":927}," .env.example",[917,950,951],{"class":927}," .env\n",[917,953,954,956,959,962],{"class":919,"line":570},[917,955,585],{"class":923},[917,957,958],{"class":927}," compose",[917,960,961],{"class":927}," up",[917,963,964],{"class":936}," -d\n",[917,966,967],{"class":919,"line":589},[917,968,970],{"class":969},"sJ8bj","# 默认 http:\u002F\u002Flocalhost \u002F 端口可在 .env 调整\n",[25,972,973],{},"硬件门槛（社区共识，非官方硬性要求）：",[35,975,976,982,988],{},[38,977,978,981],{},[41,979,980],{},"最低","：2 核 4G，纯外接 API 模式",[38,983,984,987],{},[41,985,986],{},"推荐","：4 核 8G + 至少 30GB 磁盘（向量数据 + 文件存储）",[38,989,990,993],{},[41,991,992],{},"企业","：8 核 16G+，单机日活上千",[650,995,997],{"id":996},"真实-tco","真实 TCO",[25,999,1000],{},"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）：",[97,1002,1003,1016],{},[100,1004,1005],{},[103,1006,1007,1010,1013],{},[106,1008,1009],{},"成本项",[106,1011,1012],{},"云版 Professional",[106,1014,1015],{},"自托管",[115,1017,1018,1028,1038,1049],{},[103,1019,1020,1023,1026],{},[120,1021,1022],{},"平台费",[120,1024,1025],{},"~$2,100（3 年）",[120,1027,125],{},[103,1029,1030,1033,1035],{},[120,1031,1032],{},"服务器",[120,1034,125],{},[120,1036,1037],{},"~$50\u002F月 × 36 = $1,800",[103,1039,1040,1043,1046],{},[120,1041,1042],{},"模型 API",[120,1044,1045],{},"与下同",[120,1047,1048],{},"与上同",[103,1050,1051,1054,1057],{},[120,1052,1053],{},"运维人力",[120,1055,1056],{},"0",[120,1058,1059],{},"约 0.2 人月",[25,1061,1062,1065],{},[41,1063,1064],{},"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。",[20,1067,1069],{"id":1068},"上手-10-分钟","上手 10 分钟",[909,1071,1073],{"className":911,"code":1072,"language":913,"meta":554,"style":554},"# 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",[230,1074,1075,1080,1088,1094,1102,1112,1118,1124,1130,1135,1141,1146,1152,1158],{"__ignoreMap":554},[917,1076,1077],{"class":919,"line":920},[917,1078,1079],{"class":969},"# 1. 自托管（社区版）\n",[917,1081,1082,1084,1086],{"class":919,"line":558},[917,1083,924],{"class":923},[917,1085,928],{"class":927},[917,1087,931],{"class":927},[917,1089,1090,1092],{"class":919,"line":555},[917,1091,937],{"class":936},[917,1093,940],{"class":927},[917,1095,1096,1098,1100],{"class":919,"line":570},[917,1097,945],{"class":923},[917,1099,948],{"class":927},[917,1101,951],{"class":927},[917,1103,1104,1106,1108,1110],{"class":919,"line":589},[917,1105,585],{"class":923},[917,1107,958],{"class":927},[917,1109,961],{"class":927},[917,1111,964],{"class":936},[917,1113,1115],{"class":919,"line":1114},6,[917,1116,1117],{"emptyLinePlaceholder":580},"\n",[917,1119,1121],{"class":919,"line":1120},7,[917,1122,1123],{"class":969},"# 2. 浏览器打开 http:\u002F\u002Flocalhost\n",[917,1125,1127],{"class":919,"line":1126},8,[917,1128,1129],{"class":969},"#    首次会让你创建 admin 账号\n",[917,1131,1133],{"class":919,"line":1132},9,[917,1134,1117],{"emptyLinePlaceholder":580},[917,1136,1138],{"class":919,"line":1137},10,[917,1139,1140],{"class":969},"# 3. 进入\"设置 → 模型供应商\"，配置 OpenAI \u002F 国产模型 API key\n",[917,1142,1144],{"class":919,"line":1143},11,[917,1145,1117],{"emptyLinePlaceholder":580},[917,1147,1149],{"class":919,"line":1148},12,[917,1150,1151],{"class":969},"# 4. 在主界面\"创建空白应用\"，选 Chatflow 或 Workflow\n",[917,1153,1155],{"class":919,"line":1154},13,[917,1156,1157],{"class":969},"# 5. 拖入\"开始 → LLM → 结束\"节点试一下基础 prompt\n",[917,1159,1161],{"class":919,"line":1160},14,[917,1162,1163],{"class":969},"# 6. 满意了点右上\"发布\"，自动生成 API endpoint\n",[20,1165,1166],{"id":1166},"国内使用注意事项",[221,1168,1169,1175,1181,1187],{},[38,1170,1171,1174],{},[41,1172,1173],{},"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",[38,1176,1177,1180],{},[41,1178,1179],{},"自托管 + 国产模型"," = 完全国内闭环，是 Dify 在国内最大优势",[38,1182,1183,1186],{},[41,1184,1185],{},"Docker 镜像拉取","：国内可能慢，建议配 Docker registry 镜像（阿里云 \u002F 网易）",[38,1188,1189,1192,1193,1195],{},[41,1190,1191],{},"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 ",[516,1194,639],{"href":638}," 迁到 Dify",[20,1197,1198],{"id":1198},"与同类怎么选",[97,1200,1201,1223],{},[100,1202,1203],{},[103,1204,1205,1207,1209,1213,1217],{},[106,1206,263],{},[106,1208,268],{},[106,1210,1211],{},[516,1212,639],{"href":638},[106,1214,1215],{},[516,1216,271],{"href":765},[106,1218,1219],{},[516,1220,1222],{"href":1221},"\u002Fagent\u002Fplatform\u002Fn8n.html","n8n",[115,1224,1225,1239,1252,1268,1282,1296,1310,1323],{},[103,1226,1227,1230,1232,1234,1236],{},[120,1228,1229],{},"开源",[120,1231,301],{},[120,1233,304],{},[120,1235,301],{},[120,1237,1238],{},"✅（fair-code）",[103,1240,1241,1244,1246,1248,1250],{},[120,1242,1243],{},"私有部署",[120,1245,301],{},[120,1247,304],{},[120,1249,301],{},[120,1251,301],{},[103,1253,1254,1257,1260,1263,1265],{},[120,1255,1256],{},"上手难度",[120,1258,1259],{},"★★★☆☆",[120,1261,1262],{},"★★☆☆☆ 最简单",[120,1264,1259],{},[120,1266,1267],{},"★★★★☆",[103,1269,1270,1273,1276,1278,1280],{},[120,1271,1272],{},"工作流编排",[120,1274,1275],{},"★★★★★",[120,1277,1267],{},[120,1279,1259],{},[120,1281,1275],{},[103,1283,1284,1287,1289,1291,1293],{},[120,1285,1286],{},"RAG 精度",[120,1288,1267],{},[120,1290,1259],{},[120,1292,1275],{},[120,1294,1295],{},"★★☆☆☆",[103,1297,1298,1301,1303,1305,1308],{},[120,1299,1300],{},"模型生态",[120,1302,1275],{},[120,1304,1267],{},[120,1306,1307],{},"★★★☆☆（OneAPI 中转）",[120,1309,1267],{},[103,1311,1312,1315,1317,1319,1321],{},[120,1313,1314],{},"中文场景",[120,1316,1267],{},[120,1318,1275],{},[120,1320,1267],{},[120,1322,1259],{},[103,1324,1325,1328,1330,1333,1335],{},[120,1326,1327],{},"字节生态绑定",[120,1329,304],{},[120,1331,1332],{},"✅（飞书\u002F抖音深度集成）",[120,1334,304],{},[120,1336,304],{},[25,1338,1339,1342,1343,1348,1349,1354],{},[41,1340,1341],{},"怎么选","（基于 ",[516,1344,1347],{"href":1345,"rel":1346},"https:\u002F\u002Fwww.besthub.dev\u002Farticles\u002Fcoze-vs-dify-vs-fastgpt-which-ai-agent-platform-fits-your-needs-fa59cf97b798",[544],"BestHub 2025-07"," 和 ",[516,1350,1353],{"href":1351,"rel":1352},"https:\u002F\u002Fwww.cnblogs.com\u002Fuulucias\u002Fp\u002F19449008",[544],"博客园 2026-01"," 两份选型指南综合）：",[35,1356,1357,1363,1371,1378,1385],{},[38,1358,1359,1362],{},[41,1360,1361],{},"数据必须不出内网 + 工作流复杂"," → Dify",[38,1364,1365,1368,1369],{},[41,1366,1367],{},"个人 \u002F 小团队 \u002F 快速原型 + 字节生态"," → ",[516,1370,639],{"href":638},[38,1372,1373,1368,1376],{},[41,1374,1375],{},"核心场景就是企业知识库 QA",[516,1377,271],{"href":765},[38,1379,1380,1368,1383],{},[41,1381,1382],{},"重点是连接外部 SaaS（Slack \u002F Notion \u002F 数据库）",[516,1384,1222],{"href":1221},[38,1386,1387,1368,1390],{},[41,1388,1389],{},"要画图式表达 LangChain pipeline",[516,1391,1393],{"href":1392},"\u002Fagent\u002Fplatform\u002Flangflow.html","Langflow",[20,1395,1396],{"id":1396},"避坑清单",[35,1398,1399,1405,1422,1433,1446,1452,1458,1464],{},[38,1400,1401,1404],{},[41,1402,1403],{},"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。",[38,1406,1407,1413,1414,1417,1418,1421],{},[41,1408,1409,1412],{},[230,1410,1411],{},".env"," 文件改完忘 restart","：",[230,1415,1416],{},"docker compose down && up -d","，不是 ",[230,1419,1420],{},"restart","——后者不重新加载 env。",[38,1423,1424,1413,1427,1432],{},[41,1425,1426],{},"大版本升级会破坏数据库 schema",[516,1428,1431],{"href":1429,"rel":1430},"https:\u002F\u002Fdocs.dify.ai\u002Fzh-hans",[544],"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。",[38,1434,1435,1438,1439,1441,1442,1445],{},[41,1436,1437],{},"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 ",[230,1440,1411],{}," 的 ",[230,1443,1444],{},"UPLOAD_FILE_SIZE_LIMIT"," 并重启容器。",[38,1447,1448,1451],{},[41,1449,1450],{},"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。",[38,1453,1454,1457],{},[41,1455,1456],{},"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。",[38,1459,1460,1463],{},[41,1461,1462],{},"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。",[38,1465,1466,1469],{},[41,1467,1468],{},"国内 Docker 拉取镜像慢","：先配国内 registry，否则首次 pull 可能要 30+ 分钟。",[20,1471,449],{"id":448},[25,1473,1474],{},"✅ 适合：",[35,1476,1477,1480,1483,1486,1489,1492],{},[38,1478,1479],{},"中大型企业 LLM 中台建设",[38,1481,1482],{},"需要私有化部署（金融 \u002F 医疗 \u002F 政府）",[38,1484,1485],{},"想做\"AI 工作流即产品\"的开发团队",[38,1487,1488],{},"同时需要 RAG + Agent + Workflow 三件套",[38,1490,1491],{},"想用国产模型 + 国际模型混合编排",[38,1493,1494],{},"已经接受 Docker + 一定运维投入",[25,1496,1497],{},"❌ 不适合：",[35,1499,1500,1506,1512,1515,1521],{},[38,1501,1502,1503,1505],{},"纯个人玩家做对话机器人（",[516,1504,639],{"href":638}," 更快）",[38,1507,1508,1509,1511],{},"只想做企业知识库 QA（",[516,1510,271],{"href":765}," RAG 更专）",[38,1513,1514],{},"团队完全没运维能力（云版还行，自托管会踩坑）",[38,1516,1517,1518,1520],{},"需要深度对接字节飞书 \u002F 抖音（",[516,1519,639],{"href":638}," 原生）",[38,1522,1523,1524,1526],{},"工作流核心是连接 100+ SaaS（",[516,1525,1222],{"href":1221}," 节点更全）",[20,1528,512],{"id":512},[35,1530,1531,1543,1561,1580],{},[38,1532,1533,1534,1536,1537,1536,1539,1536,1541],{},"同类对比：",[516,1535,639],{"href":638}," \u002F ",[516,1538,271],{"href":765},[516,1540,1222],{"href":1221},[516,1542,1393],{"href":1392},[38,1544,1545,1546,1536,1550,1536,1554,1536,1557],{},"概念基础：",[516,1547,1549],{"href":1548},"\u002Fwiki\u002Fai-agent.html","AI Agent",[516,1551,1553],{"href":1552},"\u002Fwiki\u002Frag.html","RAG",[516,1555,1556],{"href":794},"MCP",[516,1558,1560],{"href":1559},"\u002Fwiki\u002Ffunction-calling.html","Function Calling",[38,1562,1563,1564,1536,1568,1536,1572,1536,1576],{},"模型选型：",[516,1565,1567],{"href":1566},"\u002Fmodels\u002Fgpt-5.html","GPT-5",[516,1569,1571],{"href":1570},"\u002Fmodels\u002Fclaude-sonnet-4.html","Claude Sonnet 4",[516,1573,1575],{"href":1574},"\u002Fmodels\u002Fdeepseek-v3.html","DeepSeek-V3",[516,1577,1579],{"href":1578},"\u002Fmodels\u002Fglm-5.2.html","GLM-5.2",[38,1581,1582,1583,1536,1587],{},"进阶：",[516,1584,1586],{"href":1585},"\u002Fwiki\u002Ffine-tuning-vs-rag.html","Fine-tuning vs RAG",[516,1588,1590],{"href":1589},"\u002Fwiki\u002Fcontext-engineering.html","Context Engineering",[20,1592,530],{"id":530},[35,1594,1595,1602,1608,1614,1621],{},[38,1596,1597,1598],{},"官网：",[516,1599,1600],{"href":1600,"rel":1601},"https:\u002F\u002Fdify.ai",[544],[38,1603,1604,1605],{},"中文文档：",[516,1606,1429],{"href":1429,"rel":1607},[544],[38,1609,1610,1611],{},"GitHub：",[516,1612,904],{"href":904,"rel":1613},[544],[38,1615,1616,1617],{},"官方定价：",[516,1618,1619],{"href":1619,"rel":1620},"https:\u002F\u002Fdify.ai\u002Fpricing",[544],[38,1622,1623],{},"第三方评测：tooljunction.io \u002F chatforest.com \u002F besthub.dev \u002F joshuaopolko.com \u002F 知乎 LLM 实战笔记",[25,1625,1626,1627,1631],{},"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 ",[516,1628,1630],{"href":1629},"mailto:hello@aiho.net","反馈邮箱"," 反馈。",[1633,1634,1635],"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":554,"searchDepth":555,"depth":555,"links":1637},[1638,1639,1646,1651,1652,1653,1654,1655,1656,1657],{"id":22,"depth":558,"text":23},{"id":648,"depth":558,"text":648,"children":1640},[1641,1642,1643,1644,1645],{"id":652,"depth":555,"text":653},{"id":731,"depth":555,"text":732},{"id":769,"depth":555,"text":770},{"id":787,"depth":555,"text":788},{"id":810,"depth":555,"text":811},{"id":817,"depth":558,"text":817,"children":1647},[1648,1649,1650],{"id":820,"depth":555,"text":821},{"id":898,"depth":555,"text":899},{"id":996,"depth":555,"text":997},{"id":1068,"depth":558,"text":1069},{"id":1166,"depth":558,"text":1166},{"id":1198,"depth":558,"text":1198},{"id":1396,"depth":558,"text":1396},{"id":448,"depth":558,"text":449},{"id":512,"depth":558,"text":512},{"id":530,"depth":558,"text":530},"\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。",[577,576,1661],"ja",{},"\u002Ftools\u002Fagent\u002Fplatform\u002Fdify",[1665,1666,584,585],"windows","macos",[1668,1672,1676,1680],{"plan":1669,"price":125,"features":1670,"notes":1671},"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":1673,"price":125,"features":1674,"notes":1675},"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":1677,"price":865,"features":1678,"notes":1679},"Cloud Professional","更高调用额度 + 团队协作 + 商用支持","商用云首选",{"plan":1681,"price":1682,"features":1683,"notes":147},"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",[1687,1688,1689,1690],"coze-deep-review","coze-vs-dify","dify-deep-review","fastgpt-deep-review",{"power":589,"ux":570,"price":589,"cn_support":570,"stability":570},{"title":268,"description":1659},"Dify 评测 2026：开源 LLMOps 与 AI Agent 平台，自托管指南",[1695,1697,1699,1701,1703],{"title":1696,"url":1429},"Dify 官方文档（中文）",{"title":1698,"url":904},"Dify GitHub",{"title":1700,"url":1619},"Dify 官方定价",{"title":1702,"url":1345},"Coze vs Dify vs FastGPT 选型",{"title":1704,"url":1705},"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",[599,600,603,601,1709,1710,1711],"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",1785428432062]