[{"data":1,"prerenderedAt":4330},["ShallowReactive",2],{"header-counts":3,"footer-counts":6,"playbook-onboarding\u002Fdify-getting-started":9,"playbook-tools-onboarding\u002Fdify-getting-started":1342},{"tools":4,"reviews":5},77,25,{"tools":4,"reviews":5,"playbooks":7,"news":8},22,13,{"id":10,"title":11,"body":12,"category":1322,"cover":1323,"description":1324,"extension":1325,"meta":1326,"navigation":1155,"path":1327,"published":1328,"relatedTools":1329,"seo":1333,"stem":1334,"tags":1335,"updated":1328,"__hash__":1341},"playbook\u002Fplaybook\u002Fonboarding\u002Fdify-getting-started.md","Dify 上手指南：从零搭建 AI Agent 工作流，私有部署实战",{"type":13,"value":14,"toc":1290},"minimark",[15,19,35,42,45,48,83,107,111,114,184,195,209,219,257,267,270,281,303,309,315,321,324,327,402,408,441,447,451,454,458,605,609,612,708,714,717,723,737,746,750,753,756,772,822,836,839,845,865,871,880,883,916,920,923,940,945,1017,1023,1034,1037,1040,1043,1046,1095,1098,1104,1121,1124,1141,1144,1188,1192,1196,1204,1208,1214,1218,1224,1228,1234,1238,1244,1247,1286],[16,17,18],"h2",{"id":18},"适用人群",[20,21,22,23,27,28,27,31,34],"p",{},"这份指南适合三类用户：",[24,25,26],"strong",{},"准备用 Dify 搭企业 AI 应用平台的开发者","、",[24,29,30],{},"已经装了 Dify 但工作流编排不知道从哪下手的团队",[24,32,33],{},"在 Dify \u002F Coze \u002F FastGPT 之间做技术选型、需要跑一遍 POC 的架构师","。",[20,36,37,38,41],{},"目标是在 ",[24,39,40],{},"1 小时内完成私有部署 → 创建第一个应用 → 编排一个真实工作流 → 发布 API","，并且知道生产环境要踩哪些坑。",[16,43,44],{"id":44},"前置条件",[20,46,47],{},"在开始之前，确认以下条件：",[49,50,51,59,69,76],"ul",{},[52,53,54,55,58],"li",{},"一台 Linux 服务器（阿里云 \u002F 腾讯云 \u002F 自建均可），最低 ",[24,56,57],{},"2C4G + 30GB 硬盘","。推荐 4C8G 起步，企业级日活上千需 8C16G+。",[52,60,61,62,65,66,34],{},"已装 ",[24,63,64],{},"Docker 24+"," 和 ",[24,67,68],{},"Docker Compose 2.20+",[52,70,71,72,75],{},"至少一个可用的 ",[24,73,74],{},"LLM API Key","——OpenAI \u002F Anthropic \u002F DeepSeek \u002F Qwen \u002F 智谱 \u002F 豆包均可，Dify 支持 40+ 提供商。",[52,77,78,79,82],{},"一个",[24,80,81],{},"真实想解决的业务场景","——不要用\"你好\"测试 Dify，浪费部署时间也看不出工作流编排的价值。",[84,85,90],"div",{"className":86},[87,88,89],"card","p-5","my-4",[20,91,92,95,96,101,102,106],{},[24,93,94],{},"部署前选型","：数据必须不出内网 + 工作流复杂 → Dify 自托管（本指南）。个人 \u002F 小团队快速原型 → ",[97,98,100],"a",{"href":99},"\u002Fagent\u002Fplatform\u002Fcoze.html","Coze"," 云版更快。核心场景就是企业知识库 QA → ",[97,103,105],{"href":104},"\u002Fagent\u002Fplatform\u002Ffastgpt.html","FastGPT"," RAG 精度更专。先用 Dify 云版 Sandbox 免费 200 次调用跑 1 周 POC，再决定要不要自托管。",[16,108,110],{"id":109},"docker-部署10-分钟","Docker 部署（10 分钟）",[20,112,113],{},"Dify 社区版是 Apache 2.0 协议，完全免费可商用。官方提供 Docker Compose 一键部署：",[115,116,121],"pre",{"className":117,"code":118,"language":119,"meta":120,"style":120},"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\n","bash","",[122,123,124,140,150,162,177],"code",{"__ignoreMap":120},[125,126,129,133,137],"span",{"class":127,"line":128},"line",1,[125,130,132],{"class":131},"sScJk","git",[125,134,136],{"class":135},"sZZnC"," clone",[125,138,139],{"class":135}," https:\u002F\u002Fgithub.com\u002Flanggenius\u002Fdify.git\n",[125,141,143,147],{"class":127,"line":142},2,[125,144,146],{"class":145},"sj4cs","cd",[125,148,149],{"class":135}," dify\u002Fdocker\n",[125,151,153,156,159],{"class":127,"line":152},3,[125,154,155],{"class":131},"cp",[125,157,158],{"class":135}," .env.example",[125,160,161],{"class":135}," .env\n",[125,163,165,168,171,174],{"class":127,"line":164},4,[125,166,167],{"class":131},"docker",[125,169,170],{"class":135}," compose",[125,172,173],{"class":135}," up",[125,175,176],{"class":145}," -d\n",[125,178,180],{"class":127,"line":179},5,[125,181,183],{"class":182},"sJ8bj","# 默认 http:\u002F\u002Flocalhost\n",[20,185,186,187,190,191,194],{},"启动后访问 ",[122,188,189],{},"http:\u002F\u002F\u003C服务器IP>","，首次进入会引导创建管理员账号。",[24,192,193],{},"登录后第一件事","：改默认密码 + 在\"设置 → 模型供应商\"里配置至少一个 LLM 和一个 Embedding 模型。",[20,196,197,200,201,204,205,208],{},[24,198,199],{},"端口冲突处理","：如果服务器有其他服务占用 80\u002F443 端口，编辑 ",[122,202,203],{},"docker-compose.yaml"," 里 Nginx 服务的 ",[122,206,207],{},"ports:","，把宿主机端口改成 8080 或其他没冲突的端口。",[20,210,211,214,215,218],{},[24,212,213],{},"国内镜像加速","：Docker 镜像在国内拉取可能很慢。在 ",[122,216,217],{},".env"," 文件里或 Docker daemon 配置里加阿里云 \u002F 网易 registry 镜像加速：",[115,220,222],{"className":117,"code":221,"language":119,"meta":120,"style":120},"# \u002Fetc\u002Fdocker\u002Fdaemon.json\n{\n  \"registry-mirrors\": [\"https:\u002F\u002Fregistry.cn-hangzhou.aliyuncs.com\"]\n}\n",[122,223,224,229,235,252],{"__ignoreMap":120},[125,225,226],{"class":127,"line":128},[125,227,228],{"class":182},"# \u002Fetc\u002Fdocker\u002Fdaemon.json\n",[125,230,231],{"class":127,"line":142},[125,232,234],{"class":233},"sVt8B","{\n",[125,236,237,240,243,246,249],{"class":127,"line":152},[125,238,239],{"class":131},"  \"registry-mirrors\"",[125,241,242],{"class":145},":",[125,244,245],{"class":233}," [",[125,247,248],{"class":135},"\"https:\u002F\u002Fregistry.cn-hangzhou.aliyuncs.com\"",[125,250,251],{"class":233},"]\n",[125,253,254],{"class":127,"line":164},[125,255,256],{"class":233},"}\n",[20,258,259,260,263,264,34],{},"改完执行 ",[122,261,262],{},"sudo systemctl restart docker"," 后重新 ",[122,265,266],{},"docker compose up -d",[16,268,269],{"id":269},"配置模型",[20,271,272,273,276,277,280],{},"进入后台 → ",[24,274,275],{},"设置 → 模型供应商","，Dify 通过插件市场接入 40+ 提供商。",[24,278,279],{},"必须同时配置两类模型","：",[282,283,284,290],"ol",{},[52,285,286,289],{},[24,287,288],{},"对话模型 LLM","：GPT-5 \u002F Claude Sonnet 4 \u002F DeepSeek-V3 \u002F Qwen \u002F 豆包-Pro \u002F Kimi K2 都行",[52,291,292,280,295,298,299,302],{},[24,293,294],{},"嵌入模型 Embedding",[122,296,297],{},"text-embedding-3-large","（OpenAI）、",[122,300,301],{},"bge-large-zh-v1.5","（中文首选，可本地部署）",[20,304,305,308],{},[24,306,307],{},"常见坑","：只配对话模型没配嵌入模型，上传知识库后无法索引，界面显示\"处理中\"永远不结束。请务必两类都配。",[20,310,311,314],{},[24,312,313],{},"国产模型原生支持","是 Dify 在国内 toB 场景的关键优势——不像 FastGPT 需要中转，Dify 直接接 DeepSeek \u002F Qwen \u002F 智谱 \u002F 文心。同一个工作流里可以\"GPT-5 做复杂推理 + DeepSeek 跑日常省成本 + Ollama 本地跑敏感数据\"，按任务步骤选最合适的模型。",[20,316,317,320],{},[24,318,319],{},"预算敏感组合推荐","：DeepSeek-V3（对话）+ bge-large-zh（嵌入 + 本地），token 便宜、中文强、完全国内闭环。",[16,322,323],{"id":323},"创建第一个应用",[20,325,326],{},"Dify 1.0 把应用拆成四种类型：",[328,329,330,346],"table",{},[331,332,333],"thead",{},[334,335,336,340,343],"tr",{},[337,338,339],"th",{},"类型",[337,341,342],{},"适合场景",[337,344,345],{},"编排范式",[347,348,349,363,376,389],"tbody",{},[334,350,351,357,360],{},[352,353,354],"td",{},[24,355,356],{},"Chatbot",[352,358,359],{},"简单对话机器人",[352,361,362],{},"prompt + tools",[334,364,365,370,373],{},[352,366,367],{},[24,368,369],{},"Agent",[352,371,372],{},"自主多步任务",[352,374,375],{},"ReAct \u002F Function Calling",[334,377,378,383,386],{},[352,379,380],{},[24,381,382],{},"Chatflow",[352,384,385],{},"对话型工作流（多轮 + 分支）",[352,387,388],{},"节点 DAG，带聊天上下文",[334,390,391,396,399],{},[352,392,393],{},[24,394,395],{},"Workflow",[352,397,398],{},"单次输入→输出（API 模式）",[352,400,401],{},"节点 DAG，无对话状态",[20,403,404,407],{},[24,405,406],{},"第一个应用选 Chatbot","，10 分钟跑通：",[282,409,410,417,420,426,432,438],{},[52,411,412,413,416],{},"左侧\"工作室\" → ",[24,414,415],{},"\"创建空白应用\""," → 选\"聊天助手\"",[52,418,419],{},"起个名字（如\"技术文档助手\"）",[52,421,422,425],{},[24,423,424],{},"编排 prompt","：写清楚角色、能力边界、输出格式",[52,427,428,431],{},[24,429,430],{},"可选：关联知识库","（下一步会讲）",[52,433,434,437],{},[24,435,436],{},"调试","：右侧预览窗口输入问题，看回答效果",[52,439,440],{},"右上**\"发布\"** → 选渠道（Web App \u002F API \u002F 嵌入网页）",[20,442,443,446],{},[24,444,445],{},"关键认知","：Chatbot 只是热身。Dify 的真正价值在 Workflow 和 Chatflow——当你需要\"多步骤逻辑 + 条件分支 + 外部工具调用\"时，才需要工作流编排。",[16,448,450],{"id":449},"工作流编排dify-的核心","工作流编排：Dify 的核心",[20,452,453],{},"这是 Dify 最强的一块，也是它和 FastGPT 拉开差距的地方。进入\"工作室\" → 创建\"工作流\"应用，你会看到一块可视化画布。",[455,456,457],"h3",{"id":457},"节点类型",[328,459,460,473],{},[331,461,462],{},[334,463,464,467,470],{},[337,465,466],{},"节点",[337,468,469],{},"作用",[337,471,472],{},"实战要点",[347,474,475,488,501,514,527,540,553,566,579,592],{},[334,476,477,482,485],{},[352,478,479],{},[24,480,481],{},"开始 \u002F 结束",[352,483,484],{},"输入输出",[352,486,487],{},"定义入参变量和输出格式",[334,489,490,495,498],{},[352,491,492],{},[24,493,494],{},"LLM",[352,496,497],{},"调大模型",[352,499,500],{},"可选模型 \u002F 温度 \u002F 系统 prompt",[334,502,503,508,511],{},[352,504,505],{},[24,506,507],{},"知识检索",[352,509,510],{},"RAG 搜索",[352,512,513],{},"关联知识库，返回 Top-K chunk",[334,515,516,521,524],{},[352,517,518],{},[24,519,520],{},"代码执行",[352,522,523],{},"Python \u002F JS",[352,525,526],{},"数据转换、格式化、计算",[334,528,529,534,537],{},[352,530,531],{},[24,532,533],{},"条件分支",[352,535,536],{},"IF \u002F ELSE",[352,538,539],{},"按变量值路由到不同分支",[334,541,542,547,550],{},[352,543,544],{},[24,545,546],{},"HTTP 请求",[352,548,549],{},"调外部 API",[352,551,552],{},"对接 ERP \u002F 飞书 \u002F 自家系统",[334,554,555,560,563],{},[352,556,557],{},[24,558,559],{},"迭代",[352,561,562],{},"循环处理",[352,564,565],{},"批量处理列表数据",[334,567,568,573,576],{},[352,569,570],{},[24,571,572],{},"变量聚合",[352,574,575],{},"合并多分支",[352,577,578],{},"把 IF\u002FELSE 分支结果统一",[334,580,581,586,589],{},[352,582,583],{},[24,584,585],{},"参数提取",[352,587,588],{},"从文本提取结构化数据",[352,590,591],{},"提取意图 \u002F 实体",[334,593,594,599,602],{},[352,595,596],{},[24,597,598],{},"问题分类",[352,600,601],{},"意图路由",[352,603,604],{},"A 意图走分支 A，B 走分支 B",[455,606,608],{"id":607},"编排一个真实工作流智能客服路由","编排一个真实工作流：智能客服路由",[20,610,611],{},"以\"电商智能客服\"为例，演示 LLM 节点 + 知识检索 + 条件分支的组合：",[282,613,614,624,643,691,700],{},[52,615,616,619,620,623],{},[24,617,618],{},"开始节点","：输入变量 ",[122,621,622],{},"user_question","（用户提问）",[52,625,626,629,630,633,634,633,637,633,640],{},[24,627,628],{},"问题分类节点","：用 LLM 判断意图——",[122,631,632],{},"售后退换"," \u002F ",[122,635,636],{},"物流查询",[122,638,639],{},"商品咨询",[122,641,642],{},"其他",[52,644,645,647,648],{},[24,646,533],{},"：按分类结果路由\n",[49,649,650,663,674,684],{},[52,651,652,654,655,658,659,662],{},[122,653,632],{}," → ",[24,656,657],{},"知识检索节点","（查退换政策知识库）→ ",[24,660,661],{},"LLM 节点","（基于检索结果生成回复）",[52,664,665,654,667,670,671,673],{},[122,666,636],{},[24,668,669],{},"HTTP 请求节点","（调物流 API 查运单）→ ",[24,672,661],{},"（格式化物流信息）",[52,675,676,654,678,680,681,683],{},[122,677,639],{},[24,679,657],{},"（查商品 FAQ）→ ",[24,682,661],{},"（生成回复）",[52,685,686,654,688,690],{},[122,687,642],{},[24,689,661],{},"（兜底回复 + 建议转人工）",[52,692,693,696,697],{},[24,694,695],{},"变量聚合节点","：把四个分支的结果统一成 ",[122,698,699],{},"answer",[52,701,702,705,706],{},[24,703,704],{},"结束节点","：输出 ",[122,707,699],{},[20,709,710,711,34],{},"这个工作流把\"意图路由 + RAG 检索 + 外部 API + 兜底策略\"串成一条链路。",[24,712,713],{},"用代码写这套逻辑至少 200 行，用 Dify 拖拽 20 分钟搞定，且每一步可视化可观测",[455,715,716],{"id":716},"代码执行节点的边界",[20,718,719,720,280],{},"代码节点用 Sandbox 执行 Python \u002F JS，但有",[24,721,722],{},"严格限制",[49,724,725,728,731,734],{},[52,726,727],{},"执行超时默认 10 秒，复杂计算会超时",[52,729,730],{},"内存限制小（默认 128MB），大列表操作会 OOM",[52,732,733],{},"不能安装第三方库，只能用标准库",[52,735,736],{},"网络请求受限（Sandbox 隔离）",[20,738,739,742,743,745],{},[24,740,741],{},"生产建议","：代码节点只做轻量数据转换（JSON 解析、格式化、简单计算）。复杂逻辑改成 ",[24,744,669],{},"调外部服务——把重逻辑放在你自己的 API 里，Dify 只负责编排。",[16,747,749],{"id":748},"rag-知识库配置","RAG 知识库配置",[20,751,752],{},"RAG 是 Dify 的重要能力，虽然极致精度弱于 FastGPT，但 1.0 的混合检索已经补上了主要短板。",[455,754,755],{"id":755},"创建知识库",[282,757,758,764,767],{},[52,759,760,761],{},"进入\"知识库\" → ",[24,762,763],{},"\"创建知识库\"",[52,765,766],{},"上传文档（PDF \u002F Word \u002F Markdown \u002F TXT \u002F 网页 URL）",[52,768,769,280],{},[24,770,771],{},"分块设置",[328,773,774,787],{},[331,775,776],{},[334,777,778,781,784],{},[337,779,780],{},"参数",[337,782,783],{},"默认",[337,785,786],{},"说明",[347,788,789,800,811],{},[334,790,791,794,797],{},[352,792,793],{},"分块长度",[352,795,796],{},"500",[352,798,799],{},"每块字符数，长文档调高到 800-1000",[334,801,802,805,808],{},[352,803,804],{},"分块重叠",[352,806,807],{},"50",[352,809,810],{},"块之间重叠，避免语义断裂",[334,812,813,816,819],{},[352,814,815],{},"分段模式",[352,817,818],{},"自动",[352,820,821],{},"代码 \u002F 法律文档改\"按章节\"",[282,823,824,830],{"start":164},[52,825,826,829],{},[24,827,828],{},"索引方式","：选\"高质量\"（用 embedding 模型向量化），不要选\"经济\"（仅关键词）",[52,831,832,835],{},[24,833,834],{},"检索设置","：选\"混合检索\"（向量 + 全文 + 重排），召回率最高",[455,837,838],{"id":838},"混合检索调参",[20,840,841,842,280],{},"1.0 的 RAG 从纯向量升级到",[24,843,844],{},"混合检索 + 重排序",[49,846,847,853,859],{},[52,848,849,852],{},[24,850,851],{},"向量检索","：语义相似度，适合意图理解",[52,854,855,858],{},[24,856,857],{},"全文检索","：BM25 关键词匹配，适合专业术语、编号",[52,860,861,864],{},[24,862,863],{},"重排序 Rerank","：把初步召回的前 20 段进一步排序，选出最相关的 5-10 段",[20,866,867,870],{},[24,868,869],{},"Rerank 模型选择","：BGE Reranker（本地，中文好）或 Cohere Rerank（云端，多语言）。Rerank 对精度提升明显，但每次调用增加 200-500ms 延迟。",[20,872,873,876,877,879],{},[24,874,875],{},"社区版 vs 企业版差距","：多路召回 + 重排在企业版才完整解锁。社区版默认是基础语义检索，RAG 精度上限有限。如果你的核心场景就是知识库 QA 且要极致精度，",[97,878,105],{"href":104}," 社区版的 RAG 链路每步都可调，上限更高。",[455,881,882],{"id":882},"知识库避坑",[49,884,885,898,904,910],{},[52,886,887,890,891,893,894,897],{},[24,888,889],{},"文件大小限制","：社区版默认 15MB，超过会失败。改 ",[122,892,217],{}," 的 ",[122,895,896],{},"UPLOAD_FILE_SIZE_LIMIT"," 并重启容器",[52,899,900,903],{},[24,901,902],{},"大 PDF 处理慢","：单文件 > 50MB 时切分占用大量内存，先本地拆成小 PDF 再上传",[52,905,906,909],{},[24,907,908],{},"切片太碎","：默认 500 字符对代码 \u002F 长条款不友好，改成 800 字符 + overlap 100",[52,911,912,915],{},[24,913,914],{},"只调 prompt 不调切片","：答案不准 90% 是切片问题，先看召回段是否正确，再改 prompt",[16,917,919],{"id":918},"api-发布","API 发布",[20,921,922],{},"Dify 是 API-first 平台，每个应用自动暴露 REST API。",[282,924,925,931,934,937],{},[52,926,927,928],{},"进入应用 → ",[24,929,930],{},"\"访问 API\"",[52,932,933],{},"右上**\"API Server\"** → 获取 API Key",[52,935,936],{},"查看自动生成的 OpenAPI Schema 和调用示例",[52,938,939],{},"复制 curl \u002F Python \u002F Node.js 示例代码直接集成",[20,941,942,280],{},[24,943,944],{},"Workflow 应用的 API 调用",[115,946,948],{"className":117,"code":947,"language":119,"meta":120,"style":120},"curl -X POST 'https:\u002F\u002Fyour-dify\u002Fv1\u002Fworkflows\u002Frun' \\\n  -H 'Authorization: Bearer app-xxxxx' \\\n  -H 'Content-Type: application\u002Fjson' \\\n  -d '{\n    \"inputs\": {\"user_question\": \"我的订单什么时候发货？\"},\n    \"response_mode\": \"blocking\",\n    \"user\": \"user-123\"\n  }'\n",[122,949,950,967,977,986,994,999,1005,1011],{"__ignoreMap":120},[125,951,952,955,958,961,964],{"class":127,"line":128},[125,953,954],{"class":131},"curl",[125,956,957],{"class":145}," -X",[125,959,960],{"class":135}," POST",[125,962,963],{"class":135}," 'https:\u002F\u002Fyour-dify\u002Fv1\u002Fworkflows\u002Frun'",[125,965,966],{"class":145}," \\\n",[125,968,969,972,975],{"class":127,"line":142},[125,970,971],{"class":145},"  -H",[125,973,974],{"class":135}," 'Authorization: Bearer app-xxxxx'",[125,976,966],{"class":145},[125,978,979,981,984],{"class":127,"line":152},[125,980,971],{"class":145},[125,982,983],{"class":135}," 'Content-Type: application\u002Fjson'",[125,985,966],{"class":145},[125,987,988,991],{"class":127,"line":164},[125,989,990],{"class":145},"  -d",[125,992,993],{"class":135}," '{\n",[125,995,996],{"class":127,"line":179},[125,997,998],{"class":135},"    \"inputs\": {\"user_question\": \"我的订单什么时候发货？\"},\n",[125,1000,1002],{"class":127,"line":1001},6,[125,1003,1004],{"class":135},"    \"response_mode\": \"blocking\",\n",[125,1006,1008],{"class":127,"line":1007},7,[125,1009,1010],{"class":135},"    \"user\": \"user-123\"\n",[125,1012,1014],{"class":127,"line":1013},8,[125,1015,1016],{"class":135},"  }'\n",[20,1018,1019,1022],{},[24,1020,1021],{},"多平台发布","：除了 API，Dify 还支持发布到 Web App、Slack、Discord、微信公众号企业版。一次配置多处触达，不用每个平台单独写集成代码。",[20,1024,1025,1028,1029,1033],{},[24,1026,1027],{},"MCP 双向支持","：1.0 起 Dify 可作为 MCP Server 暴露工具——让 ",[97,1030,1032],{"href":1031},"\u002Fcoding\u002Fcli\u002Fclaude-code.html","Claude Code"," \u002F Cursor 调用 Dify 里的 workflow；也能消费外部 MCP Server——在 workflow 里调 GitHub \u002F Slack \u002F 自家内部系统。这让 Dify 不再是孤岛，而是 MCP 生态的中间层。",[16,1035,1036],{"id":1036},"私有化部署注意事项",[20,1038,1039],{},"私有化是 Dify 最深的护城河，但运维成本不能忽视。",[455,1041,1042],{"id":1042},"组件与资源",[20,1044,1045],{},"Dify 的 Docker Compose 包含多个组件：API Server、Web 前端、Worker、Sandbox、PostgreSQL、Redis、向量库（Weaviate）。链路比 FastGPT 长，资源占用更大。",[328,1047,1048,1060],{},[331,1049,1050],{},[334,1051,1052,1055,1058],{},[337,1053,1054],{},"规模",[337,1056,1057],{},"推荐配置",[337,1059,786],{},[347,1061,1062,1073,1084],{},[334,1063,1064,1067,1070],{},[352,1065,1066],{},"POC \u002F 小团队",[352,1068,1069],{},"2C4G + 30GB",[352,1071,1072],{},"纯外接 API 模式",[334,1074,1075,1078,1081],{},[352,1076,1077],{},"中型团队",[352,1079,1080],{},"4C8G + 50GB",[352,1082,1083],{},"推荐",[334,1085,1086,1089,1092],{},[352,1087,1088],{},"企业级",[352,1090,1091],{},"8C16G + 100GB+",[352,1093,1094],{},"单机日活上千",[455,1096,1097],{"id":1097},"版本升级",[20,1099,1100,1103],{},[24,1101,1102],{},"大版本升级会破坏数据库 schema","。跨大版本（如 0.x → 1.x）务必：",[282,1105,1106,1112,1115,1118],{},[52,1107,1108,1109],{},"先备份 PostgreSQL 卷：",[122,1110,1111],{},"docker exec dify-db pg_dump dify > backup.sql",[52,1113,1114],{},"在 staging 环境验证升级流程",[52,1116,1117],{},"生产升级前确认 release notes 的 breaking changes",[52,1119,1120],{},"升级后跑一遍核心 workflow 确认无 regression",[455,1122,1123],{"id":1123},"环境变量",[20,1125,1126,1128,1129,1132,1133,1136,1137,1140],{},[122,1127,217],{}," 改完要 ",[122,1130,1131],{},"docker compose down && docker compose up -d","，",[24,1134,1135],{},"不是"," ",[122,1138,1139],{},"docker compose restart","——后者不重新加载 env。这是最常见的\"改了配置不生效\"的原因。",[455,1142,1143],{"id":1143},"安全加固",[49,1145,1148,1158,1164,1170,1176,1182],{"className":1146},[1147],"contains-task-list",[52,1149,1152,1157],{"className":1150},[1151],"task-list-item",[1153,1154],"input",{"disabled":1155,"type":1156},true,"checkbox"," 管理员默认密码已改",[52,1159,1161,1163],{"className":1160},[1151],[1153,1162],{"disabled":1155,"type":1156}," 服务器只开必要端口，80\u002F443 走 Nginx 反代 + HTTPS",[52,1165,1167,1169],{"className":1166},[1151],[1153,1168],{"disabled":1155,"type":1156}," PostgreSQL 数据卷已定期备份",[52,1171,1173,1175],{"className":1172},[1151],[1153,1174],{"disabled":1155,"type":1156}," LLM API Key 走 secret 管理，不写到代码里",[52,1177,1179,1181],{"className":1178},[1151],[1153,1180],{"disabled":1155,"type":1156}," 限流：单用户 QPM、单应用 QPS 都有上限",[52,1183,1185,1187],{"className":1184},[1151],[1153,1186],{"disabled":1155,"type":1156}," 内网 DNS 已配好，员工能通过友好域名访问",[16,1189,1191],{"id":1190},"_5-个可直接复用的工作流模板","5 个可直接复用的工作流模板",[455,1193,1195],{"id":1194},"_1-智能客服路由","1. 智能客服路由",[115,1197,1202],{"className":1198,"code":1200,"language":1201,"meta":120},[1199],"language-text","开始(user_question) → 问题分类(售后\u002F物流\u002F商品\u002F其他) → 条件分支\n  ├ 售后 → 知识检索(退换政策) → LLM(生成回复)\n  ├ 物流 → HTTP请求(查运单API) → LLM(格式化)\n  ├ 商品 → 知识检索(商品FAQ) → LLM(生成回复)\n  └ 其他 → LLM(兜底+转人工)\n→ 变量聚合(answer) → 结束\n","text",[122,1203,1200],{"__ignoreMap":120},[455,1205,1207],{"id":1206},"_2-文档问答-引用溯源","2. 文档问答 + 引用溯源",[115,1209,1212],{"className":1210,"code":1211,"language":1201,"meta":120},[1199],"开始(question) → 知识检索(Top-5 chunk) → LLM(基于chunk回答+标注引用段号) → 结束(answer + sources)\n",[122,1213,1211],{"__ignoreMap":120},[455,1215,1217],{"id":1216},"_3-多模型-ab-测试","3. 多模型 A\u002FB 测试",[115,1219,1222],{"className":1220,"code":1221,"language":1201,"meta":120},[1199],"开始(input) → 条件分支(按用户ID取模分流)\n  ├ 分支A → LLM(GPT-5)\n  └ 分支B → LLM(DeepSeek-V3)\n→ 变量聚合 → 结束(记录哪个模型+用户满意度)\n",[122,1223,1221],{"__ignoreMap":120},[455,1225,1227],{"id":1226},"_4-数据-etl-报告生成","4. 数据 ETL + 报告生成",[115,1229,1232],{"className":1230,"code":1231,"language":1201,"meta":120},[1199],"开始(raw_data) → 代码执行(清洗+统计) → LLM(生成分析报告) → HTTP请求(发飞书) → 结束\n",[122,1233,1231],{"__ignoreMap":120},[455,1235,1237],{"id":1236},"_5-意图路由-agent-自主决策","5. 意图路由 + Agent 自主决策",[115,1239,1242],{"className":1240,"code":1241,"language":1201,"meta":120},[1199],"开始(user_input) → 问题分类(简单\u002F复杂)\n  ├ 简单 → LLM(直接回答) → 结束\n  └ 复杂 → Agent节点(自主调工具+多步推理) → 结束\n",[122,1243,1241],{"__ignoreMap":120},[16,1245,1246],{"id":1246},"相关阅读",[49,1248,1249,1255,1261,1267,1273],{},[52,1250,1251],{},[97,1252,1254],{"href":1253},"\u002Freview\u002Fdify-deep-review.html","Dify 深度评测：开源 AI Agent 平台私有部署首选？",[52,1256,1257],{},[97,1258,1260],{"href":1259},"\u002Fcompare\u002Ffastgpt-vs-dify.html","FastGPT vs Dify：国内企业级 RAG 与 Agent 平台怎么选",[52,1262,1263],{},[97,1264,1266],{"href":1265},"\u002Fplaybook\u002Fonboarding\u002Ffastgpt-getting-started.html","FastGPT 部署与知识库搭建实战",[52,1268,1269],{},[97,1270,1272],{"href":1271},"\u002Fplaybook\u002Fonboarding\u002Fcoze-getting-started.html","Coze 上手指南：从零搭建第一个 AI Bot",[52,1274,1275,1279,1280,1279,1283],{},[97,1276,1278],{"href":1277},"\u002Fagent\u002Fplatform\u002Fdify.html","Dify 工具卡"," · ",[97,1281,1282],{"href":104},"FastGPT 工具卡",[97,1284,1285],{"href":99},"Coze 工具卡",[1287,1288,1289],"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);}html pre.shiki code .sVt8B, html code.shiki .sVt8B{--shiki-default:#24292E;--shiki-dark:#E1E4E8}",{"title":120,"searchDepth":152,"depth":152,"links":1291},[1292,1293,1294,1295,1296,1297,1302,1307,1308,1314,1321],{"id":18,"depth":142,"text":18},{"id":44,"depth":142,"text":44},{"id":109,"depth":142,"text":110},{"id":269,"depth":142,"text":269},{"id":323,"depth":142,"text":323},{"id":449,"depth":142,"text":450,"children":1298},[1299,1300,1301],{"id":457,"depth":152,"text":457},{"id":607,"depth":152,"text":608},{"id":716,"depth":152,"text":716},{"id":748,"depth":142,"text":749,"children":1303},[1304,1305,1306],{"id":755,"depth":152,"text":755},{"id":838,"depth":152,"text":838},{"id":882,"depth":152,"text":882},{"id":918,"depth":142,"text":919},{"id":1036,"depth":142,"text":1036,"children":1309},[1310,1311,1312,1313],{"id":1042,"depth":152,"text":1042},{"id":1097,"depth":152,"text":1097},{"id":1123,"depth":152,"text":1123},{"id":1143,"depth":152,"text":1143},{"id":1190,"depth":142,"text":1191,"children":1315},[1316,1317,1318,1319,1320],{"id":1194,"depth":152,"text":1195},{"id":1206,"depth":152,"text":1207},{"id":1216,"depth":152,"text":1217},{"id":1226,"depth":152,"text":1227},{"id":1236,"depth":152,"text":1237},{"id":1246,"depth":142,"text":1246},"onboarding","\u002Fimg\u002Ftools\u002Fdify.webp","Dify 从零上手教程：Docker Compose 十分钟私有部署、创建第一个应用、工作流可视化编排（LLM 节点 \u002F 知识检索 \u002F 代码执行 \u002F 条件分支）、RAG 知识库配置与混合检索调参、API 发布与多端触达、私有化部署运维注意事项，以及 5 个可直接复用的工作流模板。","md",{},"\u002Fplaybook\u002Fonboarding\u002Fdify-getting-started","2026-07-04",[1330,1331,1332],"agent\u002Fplatform\u002Fdify","agent\u002Fplatform\u002Ffastgpt","agent\u002Fplatform\u002Fcoze",{"title":11,"description":1324},"playbook\u002Fonboarding\u002Fdify-getting-started",[1336,1337,1338,1339,1340],"Dify","AI Agent","工作流","私有部署","开源","HvREfArpp1ocyPymr09GM2VqnOnD2KR4y7YKl8fNrbw",[1343,2252,3256],{"id":1344,"title":100,"alternatives":1345,"api_compatible":1347,"body":1348,"category":2174,"chinese_friendly":179,"cover":2175,"description":2176,"domestic":2177,"extension":1325,"faq":2178,"free":2177,"github":2178,"languages":2179,"meta":2182,"models":2183,"navigation":1155,"notSuitable":2189,"opensource":2177,"path":2193,"pillar":2194,"platforms":2195,"priceTable":2197,"pricing":2218,"published":2219,"relatedPlaybooks":2178,"relatedReviews":2178,"score":2220,"self_host":2177,"seo":2221,"seoTitle":2222,"slug":1332,"sources":2223,"stem":2234,"suitable":2235,"tagline":2241,"tags":2242,"updated":2249,"verdict":2250,"website":1368,"__hash__":2251},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Fcoze.md",[1330,1331,1346],"agent\u002Fplatform\u002Fyuanqi",[],{"type":13,"value":1349,"toc":2158},[1350,1354,1392,1398,1401,1405,1408,1422,1431,1481,1490,1494,1497,1528,1536,1540,1543,1563,1567,1649,1658,1661,1664,1684,1693,1697,1733,1736,1739,1878,1893,1923,1926,1998,2002,2005,2022,2025,2055,2057,2121,2124,2151],[16,1351,1353],{"id":1352},"tldr","TL;DR",[84,1355,1357,1379],{"className":1356},[87,88,89],[20,1358,1359,1362,1363,1378],{},[24,1360,1361],{},"一句话："," 字节跳动出品的低代码 Agent 平台，",[24,1364,1365,1366,1372,1373],{},"国内版 ",[97,1367,1371],{"href":1368,"rel":1369},"https:\u002F\u002Fwww.coze.cn",[1370],"nofollow","coze.cn","（中文叫\"扣子\"）+ 国际版 ",[97,1374,1377],{"href":1375,"rel":1376},"https:\u002F\u002Fwww.coze.com",[1370],"coze.com"," 双轨运营。国内版深度集成飞书 \u002F 抖音生态、原生接入豆包；国际版集成 OpenAI \u002F Claude \u002F Gemini。",[20,1380,1381,1382,1385,1386,1391],{},"最大价值是 ",[24,1383,1384],{},"零代码、最快上手","——文档详细 + 中文社区活跃 + 模板多，根据 ",[97,1387,1390],{"href":1388,"rel":1389},"https:\u002F\u002Fwww.cnblogs.com\u002Fuulucias\u002Fp\u002F19449008",[1370],"博客园 2026-01 选型指南"," 引用的真实案例，\"某电商公司用 Coze 搭建客服机器人，3 天上线，月成本 \u003C 1000 元\"。",[1393,1394,1395],"blockquote",{},[20,1396,1397],{},"来源说明：本文基于 coze.cn \u002F coze.com 官方页面、docs.coze.com 文档、第三方选型评测（cnblogs \u002F besthub \u002F aibotgo）综合整理。字节产品迭代很快，价格 \u002F 功能请以最新官方页面为准。",[16,1399,1400],{"id":1400},"核心特性",[455,1402,1404],{"id":1403},"可视化工作流最大卖点","可视化工作流（最大卖点）",[20,1406,1407],{},"Coze 把 Agent 拆成两层：",[49,1409,1410,1416],{},[52,1411,1412,1415],{},[24,1413,1414],{},"Bot \u002F 智能体","：对话式 AI，配置 prompt + 知识库 + 插件",[52,1417,1418,1421],{},[24,1419,1420],{},"工作流（Workflow）","：DAG 节点编排，可被 Bot 调用，也可独立部署",[20,1423,1424,1425,1430],{},"工作流节点类型（基于 ",[97,1426,1429],{"href":1427,"rel":1428},"https:\u002F\u002Fdeveloper.volcengine.com\u002Farticles\u002F7530117616687480851",[1370],"火山引擎社区 2025 实战","）：",[49,1432,1433,1439,1445,1451,1457,1463,1469,1475],{},[52,1434,1435,1438],{},[24,1436,1437],{},"开始 \u002F 结束节点","：输入输出",[52,1440,1441,1444],{},[24,1442,1443],{},"大模型节点","：调豆包 \u002F GPT \u002F Claude 任一模型",[52,1446,1447,1450],{},[24,1448,1449],{},"代码节点","：内嵌 Python \u002F JavaScript（飞书插件常需要数据格式转换）",[52,1452,1453,1456],{},[24,1454,1455],{},"循环节点","：批量处理多条数据",[52,1458,1459,1462],{},[24,1460,1461],{},"条件节点","：分支判断",[52,1464,1465,1468],{},[24,1466,1467],{},"插件节点","：调 Coze 插件市场的工具",[52,1470,1471,1474],{},[24,1472,1473],{},"知识库节点","：RAG 检索",[52,1476,1477,1480],{},[24,1478,1479],{},"HTTP 节点","：调外部 API",[20,1482,1483,1484,1489],{},"典型用例（参考 ",[97,1485,1488],{"href":1486,"rel":1487},"https:\u002F\u002Fwww.toutiao.com\u002Farticle\u002F7469986334686315017",[1370],"今日头条 涛哥讲AI 2025-02 教程","）：读飞书多维表格 → 批量调大模型转写为小红书风格 → 写回飞书。整个流程零代码完成。",[455,1491,1493],{"id":1492},"插件市场200-官方-海量第三方","插件市场（200+ 官方 + 海量第三方）",[20,1495,1496],{},"Coze 的\"插件\"是封装好的 API 工具，比如：",[49,1498,1499,1505,1511,1517,1522],{},[52,1500,1501,1504],{},[24,1502,1503],{},"飞书多维表格","：增删改查记录（国内 toB 场景的杀手锏）",[52,1506,1507,1510],{},[24,1508,1509],{},"图像生成","：调豆包 \u002F SD \u002F DALL-E",[52,1512,1513,1516],{},[24,1514,1515],{},"联网搜索","：实时网页检索",[52,1518,1519,1521],{},[24,1520,520],{},"：在线运行 Python",[52,1523,1524,1527],{},[24,1525,1526],{},"第三方 SaaS","：微博、抖音、bilibili、Notion……",[20,1529,1530,1535],{},[97,1531,1534],{"href":1532,"rel":1533},"https:\u002F\u002Fkouziai.github.io\u002F",[1370],"扣子空间介绍"," 提到：\"插件数量突破 500 个\"——可信度待官方确认，但量级在百级别是确定的。",[455,1537,1539],{"id":1538},"bot-商店-多平台一键发布","Bot 商店 + 多平台一键发布",[20,1541,1542],{},"发布渠道：",[49,1544,1545,1548,1551,1554,1557,1560],{},[52,1546,1547],{},"飞书机器人（一键绑）",[52,1549,1550],{},"抖音 \u002F 头条号",[52,1552,1553],{},"微信小程序 \u002F 公众号（部分需企业认证）",[52,1555,1556],{},"自定义网页嵌入",[52,1558,1559],{},"API 接口（提供 OpenAPI 风格 REST 调用）",[52,1561,1562],{},"Discord（国际版）",[455,1564,1566],{"id":1565},"国内版-vs-国际版","国内版 vs 国际版",[328,1568,1569,1582],{},[331,1570,1571],{},[334,1572,1573,1576,1579],{},[337,1574,1575],{},"维度",[337,1577,1578],{},"扣子（coze.cn）",[337,1580,1581],{},"Coze（coze.com）",[347,1583,1584,1595,1606,1616,1627,1638],{},[334,1585,1586,1589,1592],{},[352,1587,1588],{},"主力模型",[352,1590,1591],{},"豆包 Pro \u002F DeepSeek \u002F Qwen \u002F Kimi",[352,1593,1594],{},"OpenAI \u002F Claude \u002F Gemini \u002F Cohere",[334,1596,1597,1600,1603],{},[352,1598,1599],{},"飞书 \u002F 抖音 \u002F 微信集成",[352,1601,1602],{},"✅ 原生",[352,1604,1605],{},"❌",[334,1607,1608,1611,1614],{},[352,1609,1610],{},"Discord \u002F Slack 集成",[352,1612,1613],{},"⚠️ 有限",[352,1615,1602],{},[334,1617,1618,1621,1624],{},[352,1619,1620],{},"数据存储位置",[352,1622,1623],{},"国内",[352,1625,1626],{},"海外",[334,1628,1629,1632,1635],{},[352,1630,1631],{},"支付",[352,1633,1634],{},"微信 \u002F 支付宝",[352,1636,1637],{},"海外信用卡",[334,1639,1640,1643,1646],{},[352,1641,1642],{},"内容合规",[352,1644,1645],{},"严格审核",[352,1647,1648],{},"宽松",[20,1650,1651,1654,1655,34],{},[24,1652,1653],{},"实践建议","：国内 toC \u002F toB 用扣子，海外项目 \u002F 接 GPT 用 coze.com。两边账号 \u002F 工作流 ",[24,1656,1657],{},"不互通",[16,1659,1660],{"id":1660},"价格与运行成本",[20,1662,1663],{},"国内版（扣子）：",[49,1665,1666,1672,1678],{},[52,1667,1668,1671],{},[24,1669,1670],{},"免费版","：免费模型有日额度（豆包 lite 等），适合个人玩 \u002F Demo",[52,1673,1674,1677],{},[24,1675,1676],{},"专业版","：按调用计费，模型 + 高并发，单 token 价比直连 API 略贵但省事",[52,1679,1680,1683],{},[24,1681,1682],{},"企业版","：议价，含 VPC、私有化（限定场景）、SLA",[20,1685,1686,1687,1692],{},"国际版（coze.com）的定价模式据 ",[97,1688,1691],{"href":1689,"rel":1690},"https:\u002F\u002Fdocs.coze.com\u002F",[1370],"官方文档"," 描述：\"按你访问和使用的功能分别计费，每个功能有自己的计费模型\"——目前没有简单的\"$X\u002F月\"档位，类似按 token \u002F 工具调用的 metered billing。",[16,1694,1696],{"id":1695},"上手-10-分钟","上手 10 分钟",[282,1698,1699,1712,1715,1718,1721,1724,1727,1730],{},[52,1700,1701,1702,1706,1707,1711],{},"打开 ",[97,1703,1705],{"href":1368,"rel":1704},[1370],"www.coze.cn","（国内）或 ",[97,1708,1710],{"href":1375,"rel":1709},[1370],"www.coze.com","（国际），用飞书 \u002F 抖音账号 \u002F Google 账号登录",[52,1713,1714],{},"左侧\"工作空间\" → \"+创建 Bot\"，起个名字",[52,1716,1717],{},"选模型（国内推荐豆包 Pro，国际推 Claude Sonnet 4）",[52,1719,1720],{},"写 Bot 角色 prompt",[52,1722,1723],{},"可选：上传 PDF \u002F 文档建知识库",[52,1725,1726],{},"测试一下对话效果",[52,1728,1729],{},"右上\"发布\" → 选渠道（飞书 \u002F 抖音 \u002F API \u002F Web）",[52,1731,1732],{},"拿到调用 URL \u002F 飞书机器人 webhook",[20,1734,1735],{},"进阶：在\"资源库\"创建工作流，拖节点 → 调试 → 在 Bot 里\"添加工作流\"引用。",[16,1737,1738],{"id":1738},"与同类怎么选",[328,1740,1741,1763],{},[331,1742,1743],{},[334,1744,1745,1747,1749,1753,1757],{},[337,1746,1575],{},[337,1748,100],{},[337,1750,1751],{},[97,1752,1336],{"href":1277},[337,1754,1755],{},[97,1756,105],{"href":104},[337,1758,1759],{},[97,1760,1762],{"href":1761},"\u002Fagent\u002Fplatform\u002Fyuanqi.html","元器 yuanqi",[347,1764,1765,1778,1792,1808,1824,1837,1851,1865],{},[334,1766,1767,1769,1771,1774,1776],{},[352,1768,1340],{},[352,1770,1605],{},[352,1772,1773],{},"✅",[352,1775,1773],{},[352,1777,1605],{},[334,1779,1780,1782,1785,1787,1789],{},[352,1781,1339],{},[352,1783,1784],{},"⚠️ 仅企业版",[352,1786,1773],{},[352,1788,1773],{},[352,1790,1791],{},"⚠️",[334,1793,1794,1797,1800,1803,1805],{},[352,1795,1796],{},"上手难度",[352,1798,1799],{},"★ 最简单",[352,1801,1802],{},"★★★",[352,1804,1802],{},[352,1806,1807],{},"★★",[334,1809,1810,1813,1816,1819,1822],{},[352,1811,1812],{},"工作流编排",[352,1814,1815],{},"★★★★☆",[352,1817,1818],{},"★★★★★",[352,1820,1821],{},"★★★☆☆",[352,1823,1821],{},[334,1825,1826,1829,1831,1833,1835],{},[352,1827,1828],{},"RAG 精度",[352,1830,1821],{},[352,1832,1815],{},[352,1834,1818],{},[352,1836,1821],{},[334,1838,1839,1842,1844,1846,1848],{},[352,1840,1841],{},"字节生态",[352,1843,1818],{},[352,1845,1605],{},[352,1847,1605],{},[352,1849,1850],{},"❌（腾讯系）",[334,1852,1853,1856,1859,1861,1863],{},[352,1854,1855],{},"插件市场",[352,1857,1858],{},"★★★★★ 200+",[352,1860,1821],{},[352,1862,1821],{},[352,1864,1821],{},[334,1866,1867,1870,1872,1874,1876],{},[352,1868,1869],{},"中文社区",[352,1871,1818],{},[352,1873,1815],{},[352,1875,1815],{},[352,1877,1815],{},[20,1879,1880,1883,1884,65,1889,1430],{},[24,1881,1882],{},"怎么选","（综合 ",[97,1885,1888],{"href":1886,"rel":1887},"https:\u002F\u002Fwww.besthub.dev\u002Farticles\u002Fcoze-vs-dify-vs-fastgpt-which-ai-agent-platform-fits-your-needs-fa59cf97b798",[1370],"BestHub 2025-07",[97,1890,1892],{"href":1388,"rel":1891},[1370],"博客园 2026-01",[49,1894,1895,1901,1908,1915],{},[52,1896,1897,1900],{},[24,1898,1899],{},"快速验证 \u002F 不懂代码 \u002F 1-2 天出原型"," → Coze",[52,1902,1903,654,1906],{},[24,1904,1905],{},"数据安全要求高 \u002F 复杂业务流程 \u002F 有技术团队",[97,1907,1336],{"href":1277},[52,1909,1910,654,1913],{},[24,1911,1912],{},"核心场景就是企业知识库 QA",[97,1914,105],{"href":104},[52,1916,1917,654,1920],{},[24,1918,1919],{},"QQ \u002F 微信生态 + 腾讯系",[97,1921,1922],{"href":1761},"元器",[16,1924,1925],{"id":1925},"避坑清单",[49,1927,1928,1938,1944,1959,1969,1975,1986,1992],{},[52,1929,1930,1933,1934,1937],{},[24,1931,1932],{},"国内版 vs 国际版的\"双账号陷阱\"","：扣子（coze.cn）和 Coze（coze.com）是",[24,1935,1936],{},"两套独立系统","，账号、Bot、工作流不互通；想\"国内调通后搬到海外\"需要重新搭",[52,1939,1940,1943],{},[24,1941,1942],{},"专业版按调用计费容易超预算","：上线前一定在测试环境跑量估算月成本，否则爆款 Bot 一夜烧爆账户",[52,1945,1946,1949,1950,1953,1954,1958],{},[24,1947,1948],{},"飞书多维表格插件数据格式坑","：写入多维表格需要 ",[122,1951,1952],{},"Array\u003CObject>"," 格式，代码节点要做转换（参考 ",[97,1955,1957],{"href":1427,"rel":1956},[1370],"火山引擎 2025 教程","）",[52,1960,1961,1964,1965,1968],{},[24,1962,1963],{},"工作流读取飞书表格默认 20 条","：要改 ",[122,1966,1967],{},"page_size","，最大 500 条；超过 500 要分页或循环",[52,1970,1971,1974],{},[24,1972,1973],{},"运行超时","：单工作流执行有时间上限，记录条数 > 50 时建议在 Bot 里\"异步\"调用，不要直接走工作流",[52,1976,1977,1980,1981,633,1983,1985],{},[24,1978,1979],{},"企业版\"私有化\"是有限的","：完全数据不出网仍建议 ",[97,1982,1336],{"href":1277},[97,1984,105],{"href":104}," 自托管",[52,1987,1988,1991],{},[24,1989,1990],{},"国际版接 Claude \u002F GPT 需要 BYOK","：自己绑海外信用卡，平台不代付",[52,1993,1994,1997],{},[24,1995,1996],{},"审核合规","：国内版对 prompt \u002F 输出有内容审核，金融 \u002F 医疗 \u002F 政治话题可能被拦",[16,1999,2001],{"id":2000},"适合-不适合","适合 \u002F 不适合",[20,2003,2004],{},"✅ 适合：",[49,2006,2007,2010,2013,2016,2019],{},[52,2008,2009],{},"产品 \u002F 运营 \u002F 非技术人员快速做 Bot",[52,2011,2012],{},"在飞书 \u002F 抖音 \u002F 头条生态内做集成",[52,2014,2015],{},"个人副业（小红书账号批量内容生成等）",[52,2017,2018],{},"中小企业客服 Bot（3 天上线）",[52,2020,2021],{},"想用豆包 \u002F DeepSeek 国产模型的人",[20,2023,2024],{},"❌ 不适合：",[49,2026,2027,2030,2033,2038,2045],{},[52,2028,2029],{},"金融 \u002F 政府 \u002F 医疗（数据敏感，需自托管）",[52,2031,2032],{},"复杂业务系统深度集成（自由度不够）",[52,2034,2035,2036,1958],{},"反感字节生态（去 ",[97,2037,1336],{"href":1277},[52,2039,2040,2041,633,2043,1958],{},"希望开源 \u002F 完全自主可控（去 ",[97,2042,1336],{"href":1277},[97,2044,105],{"href":104},[52,2046,2047,2048,633,2050,2054],{},"海外 toB SaaS 产品后端（",[97,2049,1336],{"href":1277},[97,2051,2053],{"href":2052},"\u002Fagent\u002Fplatform\u002Flangflow.html","Langflow"," 更合适）",[16,2056,1246],{"id":1246},[49,2058,2059,2074,2091,2110],{},[52,2060,2061,2062,633,2064,633,2066,633,2068,633,2070],{},"同类对比：",[97,2063,1336],{"href":1277},[97,2065,105],{"href":104},[97,2067,1922],{"href":1761},[97,2069,2053],{"href":2052},[97,2071,2073],{"href":2072},"\u002Fagent\u002Fplatform\u002Fn8n.html","n8n",[52,2075,2076,2077,633,2080,633,2084,633,2088],{},"概念：",[97,2078,1337],{"href":2079},"\u002Fwiki\u002Fai-agent.html",[97,2081,2083],{"href":2082},"\u002Fwiki\u002Frag.html","RAG",[97,2085,2087],{"href":2086},"\u002Fwiki\u002Ffunction-calling.html","Function Calling",[97,2089,2090],{"href":2079},"Multi-Agent",[52,2092,2093,2094,633,2098,633,2102,633,2106],{},"模型：",[97,2095,2097],{"href":2096},"\u002Fmodels\u002Fdoubao-1-5-pro.html","豆包 Doubao",[97,2099,2101],{"href":2100},"\u002Fmodels\u002Fdeepseek-v3.html","DeepSeek-V3",[97,2103,2105],{"href":2104},"\u002Fmodels\u002Fqwen-3.html","Qwen3",[97,2107,2109],{"href":2108},"\u002Fmodels\u002Fglm-5.2.html","GLM-5.2",[52,2111,2112,2113,633,2117],{},"进阶：",[97,2114,2116],{"href":2115},"\u002Fwiki\u002Fprompt-engineering.html","Prompt Engineering",[97,2118,2120],{"href":2119},"\u002Fwiki\u002Fcontext-engineering.html","Context Engineering",[16,2122,2123],{"id":2123},"来源",[49,2125,2126,2132,2138,2145,2148],{},[52,2127,2128,2129],{},"国内版：",[97,2130,1368],{"href":1368,"rel":2131},[1370],[52,2133,2134,2135],{},"国际版：",[97,2136,1375],{"href":1375,"rel":2137},[1370],[52,2139,2140,2141],{},"官方文档：",[97,2142,2143],{"href":2143,"rel":2144},"https:\u002F\u002Fdocs.coze.com",[1370],[52,2146,2147],{},"第三方选型评测：cnblogs.com \u002F besthub.dev \u002F aibotgo.net",[52,2149,2150],{},"实战教程：火山引擎社区、今日头条 涛哥讲AI",[20,2152,2153,2154,2157],{},"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 渠道与最新官方信息不一致，请通过 ",[97,2155,2156],{"href":2156},"\u002Fsubmit"," 反馈。",{"title":120,"searchDepth":152,"depth":152,"links":2159},[2160,2161,2167,2168,2169,2170,2171,2172,2173],{"id":1352,"depth":142,"text":1353},{"id":1400,"depth":142,"text":1400,"children":2162},[2163,2164,2165,2166],{"id":1403,"depth":152,"text":1404},{"id":1492,"depth":152,"text":1493},{"id":1538,"depth":152,"text":1539},{"id":1565,"depth":152,"text":1566},{"id":1660,"depth":142,"text":1660},{"id":1695,"depth":142,"text":1696},{"id":1738,"depth":142,"text":1738},{"id":1925,"depth":142,"text":1925},{"id":2000,"depth":142,"text":2001},{"id":1246,"depth":142,"text":1246},{"id":2123,"depth":142,"text":2123},"platform","\u002Fimg\u002Ftools\u002Fcoze.webp","Coze 扣子 2026 真实评测：字节低代码 AI Agent 平台，支持可视化工作流、插件市场、知识库和 Bot 发布。本文对比 Dify、FastGPT、n8n，梳理国内版\u002F国际版差异、价格、适合场景和避坑建议。",false,null,[2180,2181],"zh","en",{},[2184,2185,2186,2187,2188],"doubao-pro","deepseek-v3","qwen-max","gpt-4o (国际版)","claude (国际版)",[2190,2191,2192],"需要私有部署、数据不出网（去 Dify \u002F FastGPT）","需要深度定制（去 LangGraph \u002F n8n）","对字节生态依赖反感","\u002Ftools\u002Fagent\u002Fplatform\u002Fcoze","agent",[2196],"web",[2198,2204,2209,2213],{"plan":2199,"price":2200,"limit":2201,"cn_pay":2202,"note":2203},"免费版（扣子）","¥0","免费模型限额 + 基础功能","—","个人试水 \u002F 1-2 天 MVP",{"plan":1676,"price":2205,"limit":2206,"cn_pay":2207,"note":2208},"按调用计费","高级模型 + 高并发 + 商用授权","✅ 微信\u002F支付宝","上量后切",{"plan":1682,"price":2210,"limit":2211,"cn_pay":1773,"note":2212},"议价","VPC、合规、SLA、私有化（部分）","B 端落地",{"plan":2214,"price":2215,"cn_pay":2216,"note":2217,"limit":2202},"国际版 (coze.com)","免费起步 + 用量计费","需海外卡","可接 GPT\u002FClaude\u002FGemini","免费档 \u002F 专业版按调用计费 \u002F 企业版议价","2026-06-18",{"power":164,"ux":179,"price":164,"cn_support":179,"stability":164},{"title":100,"description":2176},"Coze 扣子评测 2026：字节 AI Agent 平台，对比 Dify",[2224,2226,2228,2230,2232],{"title":2225,"url":1368},"Coze 国内版（扣子）",{"title":2227,"url":1375},"Coze 国际版",{"title":2229,"url":2143},"Coze 官方文档",{"title":2231,"url":1388},"Coze vs Dify vs FastGPT 选型 2026",{"title":2233,"url":1886},"BestHub 三平台对比","tools\u002Fagent\u002Fplatform\u002Fcoze",[2236,2237,2238,2239,2240],"想 1 小时做出一个 Bot 的产品 \u002F 运营","需要发布到飞书 \u002F 微信 \u002F 抖音的 Bot","工作流可视化编排（不想写代码）","需要批量调用国内大模型 + 飞书多维表格的工作流","C 端 \u002F 轻量 toB 场景","字节出品的 Agent 搭建平台，国内 \u002F 国际双版本",[2243,2244,2245,2246,2247,2248],"agent-platform","low-code","workflow","bot-marketplace","bytedance","no-code","2026-06-24","想最快做出一个能用的 Bot，从 Coze 起步。要私有部署或开源协作，去 Dify \u002F FastGPT。","_ax6ctOSBd7giI3x-ID55Xvgk3Q0NZV5mixd0fcWkJg",{"id":2253,"title":1336,"alternatives":2254,"api_compatible":2178,"body":2257,"category":2174,"chinese_friendly":164,"cover":1323,"description":3204,"domestic":2177,"extension":1325,"faq":2178,"free":2177,"github":2525,"languages":3205,"meta":3207,"models":2178,"navigation":1155,"notSuitable":2178,"opensource":1155,"path":3208,"pillar":2194,"platforms":3209,"priceTable":3213,"pricing":3230,"published":2219,"relatedPlaybooks":2178,"relatedReviews":2178,"score":3231,"self_host":1155,"seo":3232,"seoTitle":3233,"slug":1330,"sources":3234,"stem":3246,"suitable":2178,"tagline":3247,"tags":3248,"updated":2249,"verdict":3254,"website":3150,"__hash__":3255},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Fdify.md",[1332,1331,2255,2256],"agent\u002Fplatform\u002Fn8n","agent\u002Fplatform\u002Flangflow",{"type":13,"value":2258,"toc":3182},[2259,2261,2278,2283,2285,2289,2292,2346,2349,2353,2356,2370,2386,2390,2393,2404,2408,2416,2427,2431,2434,2436,2440,2449,2510,2517,2521,2529,2571,2574,2593,2597,2600,2660,2666,2668,2758,2761,2790,2792,2920,2932,2967,2969,3037,3039,3041,3061,3063,3092,3094,3141,3143,3174,3179],[16,2260,1353],{"id":1352},[84,2262,2264,2269],{"className":2263},[87,88,89],[20,2265,2266,2268],{},[24,2267,1361],{}," 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。",[20,2270,1381,2271,2274,2275,2277],{},[24,2272,2273],{},"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 ",[97,2276,100],{"href":99}," 折腾，新手得读 1-2 小时文档。",[1393,2279,2280],{},[20,2281,2282],{},"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。",[16,2284,1400],{"id":1400},[455,2286,2288],{"id":2287},"可视化工作流chatflow-workflow","可视化工作流（Chatflow + Workflow）",[20,2290,2291],{},"Dify 把 LLM 应用拆成两种\"应用类型\"：",[328,2293,2294,2304],{},[331,2295,2296],{},[334,2297,2298,2300,2302],{},[337,2299,339],{},[337,2301,342],{},[337,2303,345],{},[347,2305,2306,2316,2326,2336],{},[334,2307,2308,2312,2314],{},[352,2309,2310],{},[24,2311,356],{},[352,2313,359],{},[352,2315,362],{},[334,2317,2318,2322,2324],{},[352,2319,2320],{},[24,2321,369],{},[352,2323,372],{},[352,2325,375],{},[334,2327,2328,2332,2334],{},[352,2329,2330],{},[24,2331,382],{},[352,2333,385],{},[352,2335,388],{},[334,2337,2338,2342,2344],{},[352,2339,2340],{},[24,2341,395],{},[352,2343,398],{},[352,2345,401],{},[20,2347,2348],{},"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\"。",[455,2350,2352],{"id":2351},"rag-知识库","RAG 知识库",[20,2354,2355],{},"内置完整 RAG 链路：",[282,2357,2358,2361,2364,2367],{},[52,2359,2360],{},"上传文档（PDF \u002F Word \u002F Markdown \u002F 网页）",[52,2362,2363],{},"自动分块 + embedding（可配置分段策略和 embedding 模型）",[52,2365,2366],{},"混合检索（向量 + 全文 + 重排）",[52,2368,2369],{},"引用溯源（回答末尾自动附原文片段）",[20,2371,2372,2373,2378,2379,2382,2383,2385],{},"注意：根据 ",[97,2374,2377],{"href":2375,"rel":2376},"https:\u002F\u002Fzhuanlan.zhihu.com\u002Fp\u002F1887141987838309480",[1370],"知乎 LLM 实战笔记 2025-03 对比"," 的实测，Dify ",[24,2380,2381],{},"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 ",[97,2384,105],{"href":104},"（实测准确率高 10+ 个百分点），Dify 胜在工作流而非纯 RAG。",[455,2387,2389],{"id":2388},"模型生态40-提供商","模型生态：40+ 提供商",[20,2391,2392],{},"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。",[20,2394,2395,2396,2398,2399,2403],{},"国产模型原生支持（不像 ",[97,2397,105],{"href":104}," 需要 ",[97,2400,2402],{"href":2401},"\u002Fcoding\u002Fapi\u002Fone-api.html","OneAPI"," 中转），是 Dify 在国内 toB 场景流行的关键。",[455,2405,2407],{"id":2406},"mcp-协议支持","MCP 协议支持",[20,2409,2410,2411,2415],{},"Dify 较早接入了 ",[97,2412,2414],{"href":2413},"\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：",[49,2417,2418,2421,2424],{},[52,2419,2420],{},"通过 MCP 调本地 PostgreSQL \u002F SQLite",[52,2422,2423],{},"通过 MCP 调 GitHub \u002F Slack \u002F Linear",[52,2425,2426],{},"通过 MCP 调自家内部系统（写一个 MCP Server 即可）",[455,2428,2430],{"id":2429},"api-first","API-first",[20,2432,2433],{},"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。",[16,2435,1660],{"id":1660},[455,2437,2439],{"id":2438},"云版difyai","云版（dify.ai）",[20,2441,2442,2443,2448],{},"根据 ",[97,2444,2447],{"href":2445,"rel":2446},"https:\u002F\u002Fwww.tooljunction.io\u002Fai-tools\u002Fdify-ai",[1370],"tooljunction.io 2026 评测"," 引用的官方定价：",[328,2450,2451,2464],{},[331,2452,2453],{},[334,2454,2455,2458,2461],{},[337,2456,2457],{},"套餐",[337,2459,2460],{},"价格",[337,2462,2463],{},"主要限制",[347,2465,2466,2477,2488,2499],{},[334,2467,2468,2471,2474],{},[352,2469,2470],{},"Sandbox",[352,2472,2473],{},"免费",[352,2475,2476],{},"200 次模型调用，1 app，5MB 知识库",[334,2478,2479,2482,2485],{},[352,2480,2481],{},"Professional",[352,2483,2484],{},"$59\u002F月起",[352,2486,2487],{},"5000 调用\u002F月，多 app，50MB 知识库",[334,2489,2490,2493,2496],{},[352,2491,2492],{},"Team",[352,2494,2495],{},"$159\u002F月起",[352,2497,2498],{},"团队协作、SSO",[334,2500,2501,2504,2507],{},[352,2502,2503],{},"Enterprise",[352,2505,2506],{},"联系销售",[352,2508,2509],{},"定制 SLA、私有云",[20,2511,2512,2513,2516],{},"注意：云版价格只是 Dify 平台费，",[24,2514,2515],{},"模型 API 费用另算","（自带 OpenAI \u002F Anthropic key）。",[455,2518,2520],{"id":2519},"自托管推荐","自托管（推荐）",[20,2522,2523,2528],{},[97,2524,2527],{"href":2525,"rel":2526},"https:\u002F\u002Fgithub.com\u002Flanggenius\u002Fdify",[1370],"官方 GitHub 仓库"," 提供 Docker Compose 部署，社区版完全免费可商用：",[115,2530,2532],{"className":117,"code":2531,"language":119,"meta":120,"style":120},"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",[122,2533,2534,2542,2548,2556,2566],{"__ignoreMap":120},[125,2535,2536,2538,2540],{"class":127,"line":128},[125,2537,132],{"class":131},[125,2539,136],{"class":135},[125,2541,139],{"class":135},[125,2543,2544,2546],{"class":127,"line":142},[125,2545,146],{"class":145},[125,2547,149],{"class":135},[125,2549,2550,2552,2554],{"class":127,"line":152},[125,2551,155],{"class":131},[125,2553,158],{"class":135},[125,2555,161],{"class":135},[125,2557,2558,2560,2562,2564],{"class":127,"line":164},[125,2559,167],{"class":131},[125,2561,170],{"class":135},[125,2563,173],{"class":135},[125,2565,176],{"class":145},[125,2567,2568],{"class":127,"line":179},[125,2569,2570],{"class":182},"# 默认 http:\u002F\u002Flocalhost \u002F 端口可在 .env 调整\n",[20,2572,2573],{},"硬件门槛（社区共识，非官方硬性要求）：",[49,2575,2576,2582,2587],{},[52,2577,2578,2581],{},[24,2579,2580],{},"最低","：2 核 4G，纯外接 API 模式",[52,2583,2584,2586],{},[24,2585,1083],{},"：4 核 8G + 至少 30GB 磁盘（向量数据 + 文件存储）",[52,2588,2589,2592],{},[24,2590,2591],{},"企业","：8 核 16G+，单机日活上千",[455,2594,2596],{"id":2595},"真实-tco","真实 TCO",[20,2598,2599],{},"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）：",[328,2601,2602,2615],{},[331,2603,2604],{},[334,2605,2606,2609,2612],{},[337,2607,2608],{},"成本项",[337,2610,2611],{},"云版 Professional",[337,2613,2614],{},"自托管",[347,2616,2617,2628,2638,2649],{},[334,2618,2619,2622,2625],{},[352,2620,2621],{},"平台费",[352,2623,2624],{},"~$2,100（3 年）",[352,2626,2627],{},"$0",[334,2629,2630,2633,2635],{},[352,2631,2632],{},"服务器",[352,2634,2627],{},[352,2636,2637],{},"~$50\u002F月 × 36 = $1,800",[334,2639,2640,2643,2646],{},[352,2641,2642],{},"模型 API",[352,2644,2645],{},"与下同",[352,2647,2648],{},"与上同",[334,2650,2651,2654,2657],{},[352,2652,2653],{},"运维人力",[352,2655,2656],{},"0",[352,2658,2659],{},"约 0.2 人月",[20,2661,2662,2665],{},[24,2663,2664],{},"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。",[16,2667,1696],{"id":1695},[115,2669,2671],{"className":117,"code":2670,"language":119,"meta":120,"style":120},"# 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",[122,2672,2673,2678,2686,2692,2700,2710,2715,2720,2725,2730,2736,2741,2747,2752],{"__ignoreMap":120},[125,2674,2675],{"class":127,"line":128},[125,2676,2677],{"class":182},"# 1. 自托管（社区版）\n",[125,2679,2680,2682,2684],{"class":127,"line":142},[125,2681,132],{"class":131},[125,2683,136],{"class":135},[125,2685,139],{"class":135},[125,2687,2688,2690],{"class":127,"line":152},[125,2689,146],{"class":145},[125,2691,149],{"class":135},[125,2693,2694,2696,2698],{"class":127,"line":164},[125,2695,155],{"class":131},[125,2697,158],{"class":135},[125,2699,161],{"class":135},[125,2701,2702,2704,2706,2708],{"class":127,"line":179},[125,2703,167],{"class":131},[125,2705,170],{"class":135},[125,2707,173],{"class":135},[125,2709,176],{"class":145},[125,2711,2712],{"class":127,"line":1001},[125,2713,2714],{"emptyLinePlaceholder":1155},"\n",[125,2716,2717],{"class":127,"line":1007},[125,2718,2719],{"class":182},"# 2. 浏览器打开 http:\u002F\u002Flocalhost\n",[125,2721,2722],{"class":127,"line":1013},[125,2723,2724],{"class":182},"#    首次会让你创建 admin 账号\n",[125,2726,2728],{"class":127,"line":2727},9,[125,2729,2714],{"emptyLinePlaceholder":1155},[125,2731,2733],{"class":127,"line":2732},10,[125,2734,2735],{"class":182},"# 3. 进入\"设置 → 模型供应商\"，配置 OpenAI \u002F 国产模型 API key\n",[125,2737,2739],{"class":127,"line":2738},11,[125,2740,2714],{"emptyLinePlaceholder":1155},[125,2742,2744],{"class":127,"line":2743},12,[125,2745,2746],{"class":182},"# 4. 在主界面\"创建空白应用\"，选 Chatflow 或 Workflow\n",[125,2748,2749],{"class":127,"line":8},[125,2750,2751],{"class":182},"# 5. 拖入\"开始 → LLM → 结束\"节点试一下基础 prompt\n",[125,2753,2755],{"class":127,"line":2754},14,[125,2756,2757],{"class":182},"# 6. 满意了点右上\"发布\"，自动生成 API endpoint\n",[16,2759,2760],{"id":2760},"国内使用注意事项",[282,2762,2763,2769,2775,2781],{},[52,2764,2765,2768],{},[24,2766,2767],{},"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",[52,2770,2771,2774],{},[24,2772,2773],{},"自托管 + 国产模型"," = 完全国内闭环，是 Dify 在国内最大优势",[52,2776,2777,2780],{},[24,2778,2779],{},"Docker 镜像拉取","：国内可能慢，建议配 Docker registry 镜像（阿里云 \u002F 网易）",[52,2782,2783,2786,2787,2789],{},[24,2784,2785],{},"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 ",[97,2788,100],{"href":99}," 迁到 Dify",[16,2791,1738],{"id":1738},[328,2793,2794,2814],{},[331,2795,2796],{},[334,2797,2798,2800,2802,2806,2810],{},[337,2799,1575],{},[337,2801,1336],{},[337,2803,2804],{},[97,2805,100],{"href":99},[337,2807,2808],{},[97,2809,105],{"href":104},[337,2811,2812],{},[97,2813,2073],{"href":2072},[347,2815,2816,2829,2841,2854,2866,2879,2893,2906],{},[334,2817,2818,2820,2822,2824,2826],{},[352,2819,1340],{},[352,2821,1773],{},[352,2823,1605],{},[352,2825,1773],{},[352,2827,2828],{},"✅（fair-code）",[334,2830,2831,2833,2835,2837,2839],{},[352,2832,1339],{},[352,2834,1773],{},[352,2836,1605],{},[352,2838,1773],{},[352,2840,1773],{},[334,2842,2843,2845,2847,2850,2852],{},[352,2844,1796],{},[352,2846,1821],{},[352,2848,2849],{},"★★☆☆☆ 最简单",[352,2851,1821],{},[352,2853,1815],{},[334,2855,2856,2858,2860,2862,2864],{},[352,2857,1812],{},[352,2859,1818],{},[352,2861,1815],{},[352,2863,1821],{},[352,2865,1818],{},[334,2867,2868,2870,2872,2874,2876],{},[352,2869,1828],{},[352,2871,1815],{},[352,2873,1821],{},[352,2875,1818],{},[352,2877,2878],{},"★★☆☆☆",[334,2880,2881,2884,2886,2888,2891],{},[352,2882,2883],{},"模型生态",[352,2885,1818],{},[352,2887,1815],{},[352,2889,2890],{},"★★★☆☆（OneAPI 中转）",[352,2892,1815],{},[334,2894,2895,2898,2900,2902,2904],{},[352,2896,2897],{},"中文场景",[352,2899,1815],{},[352,2901,1818],{},[352,2903,1815],{},[352,2905,1821],{},[334,2907,2908,2911,2913,2916,2918],{},[352,2909,2910],{},"字节生态绑定",[352,2912,1605],{},[352,2914,2915],{},"✅（飞书\u002F抖音深度集成）",[352,2917,1605],{},[352,2919,1605],{},[20,2921,2922,2924,2925,65,2928,2931],{},[24,2923,1882],{},"（基于 ",[97,2926,1888],{"href":1886,"rel":2927},[1370],[97,2929,1892],{"href":1388,"rel":2930},[1370]," 两份选型指南综合）：",[49,2933,2934,2940,2947,2953,2960],{},[52,2935,2936,2939],{},[24,2937,2938],{},"数据必须不出内网 + 工作流复杂"," → Dify",[52,2941,2942,654,2945],{},[24,2943,2944],{},"个人 \u002F 小团队 \u002F 快速原型 + 字节生态",[97,2946,100],{"href":99},[52,2948,2949,654,2951],{},[24,2950,1912],{},[97,2952,105],{"href":104},[52,2954,2955,654,2958],{},[24,2956,2957],{},"重点是连接外部 SaaS（Slack \u002F Notion \u002F 数据库）",[97,2959,2073],{"href":2072},[52,2961,2962,654,2965],{},[24,2963,2964],{},"要画图式表达 LangChain pipeline",[97,2966,2053],{"href":2052},[16,2968,1925],{"id":1925},[49,2970,2971,2977,2992,3002,3013,3019,3025,3031],{},[52,2972,2973,2976],{},[24,2974,2975],{},"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。",[52,2978,2979,280,2984,2987,2988,2991],{},[24,2980,2981,2983],{},[122,2982,217],{}," 文件改完忘 restart",[122,2985,2986],{},"docker compose down && up -d","，不是 ",[122,2989,2990],{},"restart","——后者不重新加载 env。",[52,2993,2994,280,2996,3001],{},[24,2995,1102],{},[97,2997,3000],{"href":2998,"rel":2999},"https:\u002F\u002Fdocs.dify.ai\u002Fzh-hans",[1370],"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。",[52,3003,3004,3007,3008,893,3010,3012],{},[24,3005,3006],{},"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 ",[122,3009,217],{},[122,3011,896],{}," 并重启容器。",[52,3014,3015,3018],{},[24,3016,3017],{},"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。",[52,3020,3021,3024],{},[24,3022,3023],{},"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。",[52,3026,3027,3030],{},[24,3028,3029],{},"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。",[52,3032,3033,3036],{},[24,3034,3035],{},"国内 Docker 拉取镜像慢","：先配国内 registry，否则首次 pull 可能要 30+ 分钟。",[16,3038,2001],{"id":2000},[20,3040,2004],{},[49,3042,3043,3046,3049,3052,3055,3058],{},[52,3044,3045],{},"中大型企业 LLM 中台建设",[52,3047,3048],{},"需要私有化部署（金融 \u002F 医疗 \u002F 政府）",[52,3050,3051],{},"想做\"AI 工作流即产品\"的开发团队",[52,3053,3054],{},"同时需要 RAG + Agent + Workflow 三件套",[52,3056,3057],{},"想用国产模型 + 国际模型混合编排",[52,3059,3060],{},"已经接受 Docker + 一定运维投入",[20,3062,2024],{},[49,3064,3065,3071,3077,3080,3086],{},[52,3066,3067,3068,3070],{},"纯个人玩家做对话机器人（",[97,3069,100],{"href":99}," 更快）",[52,3072,3073,3074,3076],{},"只想做企业知识库 QA（",[97,3075,105],{"href":104}," RAG 更专）",[52,3078,3079],{},"团队完全没运维能力（云版还行，自托管会踩坑）",[52,3081,3082,3083,3085],{},"需要深度对接字节飞书 \u002F 抖音（",[97,3084,100],{"href":99}," 原生）",[52,3087,3088,3089,3091],{},"工作流核心是连接 100+ SaaS（",[97,3090,2073],{"href":2072}," 节点更全）",[16,3093,1246],{"id":1246},[49,3095,3096,3106,3118,3133],{},[52,3097,2061,3098,633,3100,633,3102,633,3104],{},[97,3099,100],{"href":99},[97,3101,105],{"href":104},[97,3103,2073],{"href":2072},[97,3105,2053],{"href":2052},[52,3107,3108,3109,633,3111,633,3113,633,3116],{},"概念基础：",[97,3110,1337],{"href":2079},[97,3112,2083],{"href":2082},[97,3114,3115],{"href":2413},"MCP",[97,3117,2087],{"href":2086},[52,3119,3120,3121,633,3125,633,3129,633,3131],{},"模型选型：",[97,3122,3124],{"href":3123},"\u002Fmodels\u002Fgpt-5.html","GPT-5",[97,3126,3128],{"href":3127},"\u002Fmodels\u002Fclaude-sonnet-4.html","Claude Sonnet 4",[97,3130,2101],{"href":2100},[97,3132,2109],{"href":2108},[52,3134,2112,3135,633,3139],{},[97,3136,3138],{"href":3137},"\u002Fwiki\u002Ffine-tuning-vs-rag.html","Fine-tuning vs RAG",[97,3140,2120],{"href":2119},[16,3142,2123],{"id":2123},[49,3144,3145,3152,3158,3164,3171],{},[52,3146,3147,3148],{},"官网：",[97,3149,3150],{"href":3150,"rel":3151},"https:\u002F\u002Fdify.ai",[1370],[52,3153,3154,3155],{},"中文文档：",[97,3156,2998],{"href":2998,"rel":3157},[1370],[52,3159,3160,3161],{},"GitHub：",[97,3162,2525],{"href":2525,"rel":3163},[1370],[52,3165,3166,3167],{},"官方定价：",[97,3168,3169],{"href":3169,"rel":3170},"https:\u002F\u002Fdify.ai\u002Fpricing",[1370],[52,3172,3173],{},"第三方评测：tooljunction.io \u002F chatforest.com \u002F besthub.dev \u002F joshuaopolko.com \u002F 知乎 LLM 实战笔记",[20,3175,3176,3177,2157],{},"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 ",[97,3178,2156],{"href":2156},[1287,3180,3181],{},"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":120,"searchDepth":152,"depth":152,"links":3183},[3184,3185,3192,3197,3198,3199,3200,3201,3202,3203],{"id":1352,"depth":142,"text":1353},{"id":1400,"depth":142,"text":1400,"children":3186},[3187,3188,3189,3190,3191],{"id":2287,"depth":152,"text":2288},{"id":2351,"depth":152,"text":2352},{"id":2388,"depth":152,"text":2389},{"id":2406,"depth":152,"text":2407},{"id":2429,"depth":152,"text":2430},{"id":1660,"depth":142,"text":1660,"children":3193},[3194,3195,3196],{"id":2438,"depth":152,"text":2439},{"id":2519,"depth":152,"text":2520},{"id":2595,"depth":152,"text":2596},{"id":1695,"depth":142,"text":1696},{"id":2760,"depth":142,"text":2760},{"id":1738,"depth":142,"text":1738},{"id":1925,"depth":142,"text":1925},{"id":2000,"depth":142,"text":2001},{"id":1246,"depth":142,"text":1246},{"id":2123,"depth":142,"text":2123},"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。",[2180,2181,3206],"ja",{},"\u002Ftools\u002Fagent\u002Fplatform\u002Fdify",[3210,3211,3212,167],"windows","macos","linux",[3214,3218,3222,3226],{"plan":3215,"price":2627,"features":3216,"notes":3217},"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":3219,"price":2627,"features":3220,"notes":3221},"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":3223,"price":2484,"features":3224,"notes":3225},"Cloud Professional","更高调用额度 + 团队协作 + 商用支持","商用云首选",{"plan":3227,"price":3228,"features":3229,"notes":2506},"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",{"power":179,"ux":164,"price":179,"cn_support":164,"stability":164},{"title":1336,"description":3204},"Dify 评测 2026：开源 LLMOps 与 AI Agent 平台，自托管指南",[3235,3237,3239,3241,3243],{"title":3236,"url":2998},"Dify 官方文档（中文）",{"title":3238,"url":2525},"Dify GitHub",{"title":3240,"url":3169},"Dify 官方定价",{"title":3242,"url":1886},"Coze vs Dify vs FastGPT 选型",{"title":3244,"url":3245},"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",[2243,3249,3250,3251,2245,3252,3253],"opensource","self-host","rag","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。","0LsPnncTjEa2rm_jLZvwR6L0Zv1hFNDKVRmDyloL7uo",{"id":3257,"title":105,"alternatives":3258,"api_compatible":3261,"body":3263,"category":2174,"chinese_friendly":179,"cover":4262,"description":4263,"domestic":2177,"extension":1325,"faq":2178,"free":2177,"github":4225,"languages":4264,"meta":4265,"models":4266,"navigation":1155,"notSuitable":4270,"opensource":1155,"path":4274,"pillar":2194,"platforms":4275,"priceTable":4276,"pricing":4298,"published":2219,"relatedPlaybooks":4299,"relatedReviews":4301,"score":4303,"self_host":1155,"seo":4304,"seoTitle":2178,"slug":1331,"sources":4305,"stem":4316,"suitable":4317,"tagline":4323,"tags":4324,"updated":2249,"verdict":4328,"website":4219,"__hash__":4329},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Ffastgpt.md",[1330,1332,3259,3260],"agent\u002Fplatform\u002Fragflow","agent\u002Fplatform\u002Fanythingllm",[3262],"openai",{"type":13,"value":3264,"toc":4244},[3265,3267,3296,3307,3309,3313,3316,3330,3333,3359,3363,3370,3402,3409,3413,3466,3473,3476,3479,3482,3489,3535,3541,3545,3574,3578,3684,3687,3690,3694,3702,3823,3838,3840,3995,4003,4037,4043,4045,4118,4120,4122,4142,4144,4162,4164,4211,4213,4236,4241],[16,3266,1353],{"id":1352},[84,3268,3270,3285],{"className":3269},[87,88,89],[20,3271,3272,3274,3275,3280,3281,3284],{},[24,3273,1361],{}," labring 团队开源的 LLM 知识库 RAG 平台，27k+ GitHub star（截至 2026-03 数据，",[97,3276,3279],{"href":3277,"rel":3278},"https:\u002F\u002Fcloud.tencent.com\u002Fdeveloper\u002Farticle\u002F2632669",[1370],"腾讯云 2026-03 教程"," 引用），Apache 2.0 许可证可商用。",[24,3282,3283],{},"核心优势是 RAG 链路工程做得极细","——问题预处理、混合检索、重排序、上下文组装、答案生成每一步都可视化调参。",[20,3286,3287,3288,3291,3292,3295],{},"最大价值在 ",[24,3289,3290],{},"国内企业知识库 + 私有部署"," 场景。代价是 ",[24,3293,3294],{},"配置门槛","：docker 基础 + 网络知识 + 一定运维能力。",[1393,3297,3298],{},[20,3299,3300,3301,3306],{},"来源说明：本文基于 fastgpt.io 官方页面、github.com\u002Flabring\u002FFastGPT 仓库、",[97,3302,3305],{"href":3303,"rel":3304},"https:\u002F\u002Fwww.nanhuantech.com\u002Fzh\u002Fai-reviews\u002Ffastgpt-2025-review",[1370],"南环 AI 2026-05 评测","、腾讯云开发者社区 2026-03 部署教程综合整理。版本迭代较快，命令和价格请以最新官方文档为准。",[16,3308,1400],{"id":1400},[455,3310,3312],{"id":3311},"知识库管理核心能力","知识库管理（核心能力）",[20,3314,3315],{},"支持文件类型：",[49,3317,3318,3321,3324,3327],{},[52,3319,3320],{},"文档：PDF \u002F Word \u002F Markdown \u002F TXT \u002F HTML",[52,3322,3323],{},"表格：Excel \u002F CSV",[52,3325,3326],{},"网页：URL 抓取 + 定时同步",[52,3328,3329],{},"API：通过接口推送内容",[20,3331,3332],{},"处理流程：上传 → 文本切分 → 向量化 → 存储 → 可用于问答。支持：",[49,3334,3335,3341,3347,3353],{},[52,3336,3337,3340],{},[24,3338,3339],{},"文件夹分组","：不同主题 \u002F 部门分类",[52,3342,3343,3346],{},[24,3344,3345],{},"多种分块策略","：默认按段落 \u002F 按 token 数 \u002F 自定义",[52,3348,3349,3352],{},[24,3350,3351],{},"批量导入","：脚本化大批量同步",[52,3354,3355,3358],{},[24,3356,3357],{},"定时同步","：网页源自动更新",[455,3360,3362],{"id":3361},"rag-流程编排最强卖点","RAG 流程编排（最强卖点）",[20,3364,3365,3369],{},[97,3366,3368],{"href":3303,"rel":3367},[1370],"南环 AI 2026 评测"," 总结的 FastGPT RAG 链路：",[282,3371,3372,3378,3384,3390,3396],{},[52,3373,3374,3377],{},[24,3375,3376],{},"问题预处理","：改写 \u002F 扩展 \u002F 错词纠正（提升召回率）",[52,3379,3380,3383],{},[24,3381,3382],{},"检索策略","：语义检索 \u002F 关键词 BM25 \u002F 混合检索，可调相似度阈值",[52,3385,3386,3389],{},[24,3387,3388],{},"重排序（Rerank）","：对初步检索结果二次排序，提升相关性",[52,3391,3392,3395],{},[24,3393,3394],{},"上下文组装","：最优 chunk + 问题 → prompt",[52,3397,3398,3401],{},[24,3399,3400],{},"答案生成","：调大模型基于检索结果回答 + 引用标注",[20,3403,3404,3405,3408],{},"每一步都可视化调参，这是 FastGPT 比 Coze \u002F Dify 在 ",[24,3406,3407],{},"纯知识库 QA 精度","上更高的原因。",[455,3410,3412],{"id":3411},"多模型支持不绑定厂商","多模型支持（不绑定厂商）",[328,3414,3415,3425],{},[331,3416,3417],{},[334,3418,3419,3422],{},[337,3420,3421],{},"模型类别",[337,3423,3424],{},"支持",[347,3426,3427,3435,3442,3450,3458],{},[334,3428,3429,3432],{},[352,3430,3431],{},"国产闭源",[352,3433,3434],{},"豆包 \u002F 通义千问 \u002F 文心一言 \u002F 智谱 GLM \u002F Moonshot Kimi \u002F MiniMax",[334,3436,3437,3439],{},[352,3438,1340],{},[352,3440,3441],{},"LLaMA \u002F Qwen \u002F ChatGLM \u002F DeepSeek 等可自部署",[334,3443,3444,3447],{},[352,3445,3446],{},"OpenAI 系",[352,3448,3449],{},"GPT-5 \u002F GPT-5 mini \u002F o3",[334,3451,3452,3455],{},[352,3453,3454],{},"Claude 系",[352,3456,3457],{},"Sonnet 4 \u002F Opus 4 \u002F Haiku",[334,3459,3460,3463],{},[352,3461,3462],{},"嵌入 \u002F 重排",[352,3464,3465],{},"BGE \u002F m3e \u002F OpenAI text-embedding-3",[20,3467,3468,3469,3472],{},"可以在 ",[24,3470,3471],{},"应用级别","为不同知识库 \u002F 不同场景配置不同模型，做\"低成本 embedding + 高质量 LLM 生成\"组合。",[455,3474,3475],{"id":3475},"工作流与高级编排",[20,3477,3478],{},"新版本（v4.14.x）支持类似 Dify 的工作流节点编排——条件分支、循环、HTTP 调用、代码节点。能做\"分类 → 路由到不同子知识库 → 不同模型回答\"这类复杂场景。",[455,3480,3481],{"id":3481},"多向量库选择",[20,3483,3484,3488],{},[97,3485,3487],{"href":3277,"rel":3486},[1370],"腾讯云教程"," 公开的 4 种向量后端：",[328,3490,3491,3501],{},[331,3492,3493],{},[334,3494,3495,3498],{},[337,3496,3497],{},"后端",[337,3499,3500],{},"适用",[347,3502,3503,3511,3519,3527],{},[334,3504,3505,3508],{},[352,3506,3507],{},"PgVector",[352,3509,3510],{},"5000 万索引以下，新手 \u002F 小规模",[334,3512,3513,3516],{},[352,3514,3515],{},"Milvus",[352,3517,3518],{},"亿级以上，高性能",[334,3520,3521,3524],{},[352,3522,3523],{},"Zilliz Cloud",[352,3525,3526],{},"Milvus 全托管 SaaS",[334,3528,3529,3532],{},[352,3530,3531],{},"SeekDB \u002F OceanBase",[352,3533,3534],{},"企业级国产化",[20,3536,3537,3538,34],{},"部署时选对应 ",[122,3539,3540],{},"docker-compose.{pgvector|milvus|...}.yml",[455,3542,3544],{"id":3543},"api-与-mcp","API 与 MCP",[49,3546,3547,3553,3559,3568],{},[52,3548,3549,3552],{},[24,3550,3551],{},"对话 API","：流式 \u002F 非流式 HTTP，OpenAI 兼容",[52,3554,3555,3558],{},[24,3556,3557],{},"知识库检索 API","：单独调检索（不走生成）做 hybrid pipeline",[52,3560,3561,3564,3565,3567],{},[24,3562,3563],{},"MCP Server","：3005 端口暴露 MCP SSE 服务，可被 ",[97,3566,1032],{"href":1031}," 等客户端直接接入",[52,3569,3570,3573],{},[24,3571,3572],{},"Webhook","：回调通知",[16,3575,3577],{"id":3576},"部署-10-分钟docker","部署 10 分钟（Docker）",[115,3579,3581],{"className":117,"code":3580,"language":119,"meta":120,"style":120},"# 克隆代码\ngit clone https:\u002F\u002Fgithub.com\u002Flabring\u002FFastGPT.git\ncd FastGPT\n\n# 切到最新稳定版（参考 GitHub releases）\ngit switch -c 4.14.7.2\n\n# 选向量库版本（个人 \u002F 小规模选 pg）\ncd deploy\u002Fdocker\u002Fcn\nwget https:\u002F\u002Fdoc.fastgpt.cn\u002Fdeploy\u002Fconfig\u002Fconfig.json\n\n# 启动\ndocker-compose -f docker-compose.pg.yml up -d\n\n# 访问 http:\u002F\u002F\u003Cip>:3000，默认账号 root \u002F 1234\n",[122,3582,3583,3588,3597,3604,3608,3613,3626,3630,3635,3642,3650,3654,3659,3674,3678],{"__ignoreMap":120},[125,3584,3585],{"class":127,"line":128},[125,3586,3587],{"class":182},"# 克隆代码\n",[125,3589,3590,3592,3594],{"class":127,"line":142},[125,3591,132],{"class":131},[125,3593,136],{"class":135},[125,3595,3596],{"class":135}," https:\u002F\u002Fgithub.com\u002Flabring\u002FFastGPT.git\n",[125,3598,3599,3601],{"class":127,"line":152},[125,3600,146],{"class":145},[125,3602,3603],{"class":135}," FastGPT\n",[125,3605,3606],{"class":127,"line":164},[125,3607,2714],{"emptyLinePlaceholder":1155},[125,3609,3610],{"class":127,"line":179},[125,3611,3612],{"class":182},"# 切到最新稳定版（参考 GitHub releases）\n",[125,3614,3615,3617,3620,3623],{"class":127,"line":1001},[125,3616,132],{"class":131},[125,3618,3619],{"class":135}," switch",[125,3621,3622],{"class":145}," -c",[125,3624,3625],{"class":145}," 4.14.7.2\n",[125,3627,3628],{"class":127,"line":1007},[125,3629,2714],{"emptyLinePlaceholder":1155},[125,3631,3632],{"class":127,"line":1013},[125,3633,3634],{"class":182},"# 选向量库版本（个人 \u002F 小规模选 pg）\n",[125,3636,3637,3639],{"class":127,"line":2727},[125,3638,146],{"class":145},[125,3640,3641],{"class":135}," deploy\u002Fdocker\u002Fcn\n",[125,3643,3644,3647],{"class":127,"line":2732},[125,3645,3646],{"class":131},"wget",[125,3648,3649],{"class":135}," https:\u002F\u002Fdoc.fastgpt.cn\u002Fdeploy\u002Fconfig\u002Fconfig.json\n",[125,3651,3652],{"class":127,"line":2738},[125,3653,2714],{"emptyLinePlaceholder":1155},[125,3655,3656],{"class":127,"line":2743},[125,3657,3658],{"class":182},"# 启动\n",[125,3660,3661,3664,3667,3670,3672],{"class":127,"line":8},[125,3662,3663],{"class":131},"docker-compose",[125,3665,3666],{"class":145}," -f",[125,3668,3669],{"class":135}," docker-compose.pg.yml",[125,3671,173],{"class":135},[125,3673,176],{"class":145},[125,3675,3676],{"class":127,"line":2754},[125,3677,2714],{"emptyLinePlaceholder":1155},[125,3679,3681],{"class":127,"line":3680},15,[125,3682,3683],{"class":182},"# 访问 http:\u002F\u002F\u003Cip>:3000，默认账号 root \u002F 1234\n",[20,3685,3686],{},"最低配置：2C4G + 20GB 硬盘 + Docker 28+ + Docker Compose 2.20+。",[20,3688,3689],{},"进入后台 → 账号 → 模型提供商 → 配置至少 1 个对话模型 + 1 个嵌入模型 → 即可开始建知识库。",[16,3691,3693],{"id":3692},"云版-vs-自托管对比","云版 vs 自托管对比",[20,3695,3696,3701],{},[97,3697,3700],{"href":3698,"rel":3699},"https:\u002F\u002Ffastgpt.io\u002Fzh\u002Fprice",[1370],"fastgpt.io 官方定价"," 公开数据：",[328,3703,3704,3729],{},[331,3705,3706],{},[334,3707,3708,3710,3712,3715,3718,3721,3723,3726],{},[337,3709,2457],{},[337,3711,2460],{},[337,3713,3714],{},"AI 积分",[337,3716,3717],{},"知识库索引",[337,3719,3720],{},"团队",[337,3722,369],{},[337,3724,3725],{},"知识库",[337,3727,3728],{},"QPM",[347,3730,3731,3755,3779,3803],{},[334,3732,3733,3735,3737,3740,3743,3746,3749,3752],{},[352,3734,2473],{},[352,3736,2200],{},[352,3738,3739],{},"100",[352,3741,3742],{},"600",[352,3744,3745],{},"1",[352,3747,3748],{},"10",[352,3750,3751],{},"3",[352,3753,3754],{},"30",[334,3756,3757,3760,3763,3766,3769,3772,3774,3776],{},[352,3758,3759],{},"基础",[352,3761,3762],{},"¥99\u002F月",[352,3764,3765],{},"4000",[352,3767,3768],{},"6000",[352,3770,3771],{},"5",[352,3773,807],{},[352,3775,3754],{},[352,3777,3778],{},"300",[334,3780,3781,3784,3787,3790,3793,3795,3798,3800],{},[352,3782,3783],{},"高级",[352,3785,3786],{},"¥599\u002F月",[352,3788,3789],{},"25000",[352,3791,3792],{},"36000",[352,3794,807],{},[352,3796,3797],{},"200",[352,3799,3739],{},[352,3801,3802],{},"1500",[334,3804,3805,3808,3810,3813,3815,3817,3819,3821],{},[352,3806,3807],{},"定制",[352,3809,2210],{},[352,3811,3812],{},"弹性",[352,3814,3812],{},[352,3816,3812],{},[352,3818,3812],{},[352,3820,3812],{},[352,3822,3812],{},[20,3824,3825,3828,3829,3832,3833,3837],{},[24,3826,3827],{},"云版适合","：不想运维、量小、要快速上线\n",[24,3830,3831],{},"自托管适合","：量大（10 万+ 日问答）、数据敏感、要深度定制——按 ",[97,3834,3836],{"href":3303,"rel":3835},[1370],"南环评测"," 估算：\"日均 10 万次问答的企业场景，商业 SaaS 年费数十万，自建 FastGPT + 开源模型只需数万硬件投入\"",[16,3839,1738],{"id":1738},[328,3841,3842,3864],{},[331,3843,3844],{},[334,3845,3846,3848,3850,3854,3858,3861],{},[337,3847,1575],{},[337,3849,105],{},[337,3851,3852],{},[97,3853,1336],{"href":1277},[337,3855,3856],{},[97,3857,100],{"href":99},[337,3859,3860],{},"RAGFlow",[337,3862,3863],{},"AnythingLLM",[347,3865,3866,3886,3902,3917,3934,3948,3965,3980],{},[334,3867,3868,3871,3874,3877,3880,3883],{},[352,3869,3870],{},"核心定位",[352,3872,3873],{},"知识库 QA",[352,3875,3876],{},"综合 LLMOps",[352,3878,3879],{},"Bot + 工作流",[352,3881,3882],{},"文档解析+RAG",[352,3884,3885],{},"桌面级 KB",[334,3887,3888,3890,3893,3895,3897,3899],{},[352,3889,1340],{},[352,3891,3892],{},"✅ Apache 2.0",[352,3894,3892],{},[352,3896,1605],{},[352,3898,3892],{},[352,3900,3901],{},"✅ MIT",[334,3903,3904,3906,3909,3911,3913,3915],{},[352,3905,1339],{},[352,3907,3908],{},"★★★★★ docker",[352,3910,1818],{},[352,3912,1784],{},[352,3914,1815],{},[352,3916,1818],{},[334,3918,3919,3922,3925,3927,3929,3932],{},[352,3920,3921],{},"RAG 深度",[352,3923,3924],{},"★★★★★ 最细",[352,3926,1815],{},[352,3928,1821],{},[352,3930,3931],{},"★★★★★ 文档解析最强",[352,3933,1821],{},[334,3935,3936,3938,3940,3942,3944,3946],{},[352,3937,1338],{},[352,3939,1815],{},[352,3941,1818],{},[352,3943,1815],{},[352,3945,1821],{},[352,3947,2878],{},[334,3949,3950,3953,3956,3958,3961,3963],{},[352,3951,3952],{},"上手",[352,3954,3955],{},"★★★☆☆ 需 docker",[352,3957,1815],{},[352,3959,3960],{},"★★★★★ 最简单",[352,3962,1821],{},[352,3964,1815],{},[334,3966,3967,3970,3972,3974,3976,3978],{},[352,3968,3969],{},"中文优化",[352,3971,1818],{},[352,3973,1815],{},[352,3975,1818],{},[352,3977,1815],{},[352,3979,1821],{},[334,3981,3982,3984,3987,3989,3991,3993],{},[352,3983,1021],{},[352,3985,3986],{},"⚠️ API 为主",[352,3988,1815],{},[352,3990,1818],{},[352,3992,1791],{},[352,3994,1791],{},[20,3996,3997,1883,3999,1430],{},[24,3998,1882],{},[97,4000,4002],{"href":3303,"rel":4001},[1370],"南环 AI 评测",[49,4004,4005,4011,4018,4025,4031],{},[52,4006,4007,4010],{},[24,4008,4009],{},"核心需求是 RAG 精度"," → FastGPT",[52,4012,4013,654,4016],{},[24,4014,4015],{},"需要丰富插件 + 复杂工作流 + 多平台发布",[97,4017,1336],{"href":1277},[52,4019,4020,654,4023],{},[24,4021,4022],{},"零代码、快速发布到飞书 \u002F 微信",[97,4024,100],{"href":99},[52,4026,4027,4030],{},[24,4028,4029],{},"文档解析（含 OCR \u002F 表格 \u002F 公式）是瓶颈"," → RAGFlow",[52,4032,4033,4036],{},[24,4034,4035],{},"桌面 \u002F 单机使用"," → AnythingLLM",[20,4038,4039,4042],{},[24,4040,4041],{},"很多企业同时用","：FastGPT 做知识库底座 + Coze 做前端 Bot 发布 \u002F 工作流编排。",[16,4044,1925],{"id":1925},[49,4046,4047,4060,4069,4075,4085,4096,4102,4108],{},[52,4048,4049,280,4052,4055,4056,4059],{},[24,4050,4051],{},"docker-compose 镜像 tag 不一致",[97,4053,3487],{"href":3277,"rel":4054},[1370]," 实测的坑——某些版本编排文件的 image tag 与最新 release 不一致，启动报\"镜像找不到\"，手动改 ",[122,4057,4058],{},"image:"," 行为正确版本即可",[52,4061,4062,4065,4066,4068],{},[24,4063,4064],{},"3000 端口冲突","：默认占用 3000（主服务）\u002F 9000（S3 \u002F MinIO）\u002F 3005（MCP）；改 ",[122,4067,3663],{}," 的 ports 映射端口",[52,4070,4071,4074],{},[24,4072,4073],{},"PostgreSQL pgvector 不够用就换 Milvus","：单库索引超 5000 万时 pgvector 查询性能下降，切 Milvus",[52,4076,4077,4080,4081,4084],{},[24,4078,4079],{},"向量库选错代价大","：先评估索引量再选向量后端，迁移要重新 embedding 整库，按 ",[97,4082,3836],{"href":3303,"rel":4083},[1370],"：\"新手 \u002F 小规模 PgVector，中大规模 Milvus，企业 \u002F 国产 OceanBase\"",[52,4086,4087,280,4090,1132,4093],{},[24,4088,4089],{},"MinIO 默认密码",[122,4091,4092],{},"minioadmin\u002Fminioadmin",[24,4094,4095],{},"部署到公网前必须改",[52,4097,4098,4101],{},[24,4099,4100],{},"分段策略影响巨大","：默认分段对长法律 \u002F 医疗文档不友好，需调\"按章节\"或\"自定义\"",[52,4103,4104,4107],{},[24,4105,4106],{},"嵌入模型 ≠ 对话模型","：经常有人只配 GPT-4 没配 embedding 模型，知识库无法索引——必须同时配两类",[52,4109,4110,4113,4114,1958],{},[24,4111,4112],{},"云版 AI 积分会过期","：未用完不能跨月累积（按 ",[97,4115,4117],{"href":3698,"rel":4116},[1370],"fastgpt.io 定价 FAQ",[16,4119,2001],{"id":2000},[20,4121,2004],{},[49,4123,4124,4127,4130,4133,4136,4139],{},[52,4125,4126],{},"企业内部知识库（员工手册 \u002F 制度 \u002F 流程）",[52,4128,4129],{},"产品 FAQ \u002F 用户手册问答",[52,4131,4132],{},"医疗 \u002F 法律 \u002F 金融垂直领域知识系统",[52,4134,4135],{},"数据严格不出网 + Apache 2.0 商用",[52,4137,4138],{},"有 docker 运维基础的技术团队",[52,4140,4141],{},"需要把 RAG 当后端服务的开发者（API 接入业务系统）",[20,4143,2024],{},[49,4145,4146,4151,4156,4159],{},[52,4147,4148,4149,1958],{},"完全非技术用户（去 ",[97,4150,100],{"href":99},[52,4152,4153,4154,1958],{},"主要需求是工作流 + 插件集成（去 ",[97,4155,1336],{"href":1277},[52,4157,4158],{},"文档解析 \u002F OCR 是首要痛点（RAGFlow）",[52,4160,4161],{},"不想自己运维 + 量很小（FastGPT 云免费版起步即可）",[16,4163,1246],{"id":1246},[49,4165,4166,4175,4191,4205],{},[52,4167,2061,4168,633,4170,4172,4173],{},[97,4169,1336],{"href":1277},[97,4171,100],{"href":99}," \u002F RAGFlow \u002F AnythingLLM \u002F ",[97,4174,2073],{"href":2072},[52,4176,2076,4177,633,4179,633,4183,633,4186,633,4189],{},[97,4178,2083],{"href":2082},[97,4180,4182],{"href":4181},"\u002Fwiki\u002Fembedding.html","Embedding",[97,4184,4185],{"href":4181},"Vector Database",[97,4187,4188],{"href":2082},"Reranker",[97,4190,1337],{"href":2079},[52,4192,2093,4193,633,4195,633,4197,633,4199,633,4203],{},[97,4194,2101],{"href":2100},[97,4196,2105],{"href":2104},[97,4198,2109],{"href":2108},[97,4200,4202],{"href":4201},"\u002Fmodels\u002Fkimi-k2.html","Kimi K2",[97,4204,2097],{"href":2096},[52,4206,2112,4207,633,4209],{},[97,4208,2120],{"href":2119},[97,4210,2116],{"href":2115},[16,4212,2123],{"id":2123},[49,4214,4215,4221,4227,4233],{},[52,4216,3147,4217],{},[97,4218,4219],{"href":4219,"rel":4220},"https:\u002F\u002Ffastgpt.io",[1370],[52,4222,3160,4223],{},[97,4224,4225],{"href":4225,"rel":4226},"https:\u002F\u002Fgithub.com\u002Flabring\u002FFastGPT",[1370],[52,4228,4229,4230],{},"定价：",[97,4231,3698],{"href":3698,"rel":4232},[1370],[52,4234,4235],{},"第三方评测：南环 AI \u002F 腾讯云开发者社区 \u002F 飞书 AGI 掘金知识库",[20,4237,4238,4239,2157],{},"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 命令 \u002F 功能与最新官方信息不一致，请通过 ",[97,4240,2156],{"href":2156},[1287,4242,4243],{},"html pre.shiki code .sJ8bj, html code.shiki .sJ8bj{--shiki-default:#6A737D;--shiki-dark:#6A737D}html pre.shiki code .sScJk, html code.shiki .sScJk{--shiki-default:#6F42C1;--shiki-dark:#B392F0}html pre.shiki code .sZZnC, html code.shiki .sZZnC{--shiki-default:#032F62;--shiki-dark:#9ECBFF}html pre.shiki code .sj4cs, html code.shiki .sj4cs{--shiki-default:#005CC5;--shiki-dark:#79B8FF}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"title":120,"searchDepth":152,"depth":152,"links":4245},[4246,4247,4255,4256,4257,4258,4259,4260,4261],{"id":1352,"depth":142,"text":1353},{"id":1400,"depth":142,"text":1400,"children":4248},[4249,4250,4251,4252,4253,4254],{"id":3311,"depth":152,"text":3312},{"id":3361,"depth":152,"text":3362},{"id":3411,"depth":152,"text":3412},{"id":3475,"depth":152,"text":3475},{"id":3481,"depth":152,"text":3481},{"id":3543,"depth":152,"text":3544},{"id":3576,"depth":142,"text":3577},{"id":3692,"depth":142,"text":3693},{"id":1738,"depth":142,"text":1738},{"id":1925,"depth":142,"text":1925},{"id":2000,"depth":142,"text":2001},{"id":1246,"depth":142,"text":1246},{"id":2123,"depth":142,"text":2123},"\u002Fimg\u002Ftools\u002Ffastgpt.webp","FastGPT 真实评测：开源 LLM 知识库 RAG 平台，labring 团队出品，27k+ GitHub star。一键 docker-compose 部署、RAG 流程编排可视化、多向量库支持。AIHO 编辑部基于官方文档与社区资料整理，含与 Dify\u002FCoze 对比、避坑指南。",[2180,2181],{},[2185,2186,2184,4267,4268,4269],"gpt-4o","claude-sonnet-4","kimi",[4271,4272,4273],"完全零代码 \u002F 不懂 docker 的用户（去 Coze）","Bot 多平台一键发布场景（Coze 强项）","插件 \u002F 工作流复杂集成（去 Dify）","\u002Ftools\u002Fagent\u002Fplatform\u002Ffastgpt",[3210,3211,3212],[4277,4281,4286,4290,4294],{"plan":4278,"price":2473,"limit":4279,"cn_pay":2202,"note":4280},"Self-host 开源","全功能 + 全数据本地","Apache 2.0 可商用",{"plan":4282,"price":4283,"limit":4284,"cn_pay":2202,"note":4285},"云免费版","¥0\u002F月","100 AI 积分 + 600 索引 + 3 知识库","试水",{"plan":4287,"price":3762,"limit":4288,"cn_pay":2207,"note":4289},"云基础版","4000 积分 + 6000 索引 + 50 Agent","中小团队 SaaS",{"plan":4291,"price":3786,"limit":4292,"cn_pay":1773,"note":4293},"云高级版","25000 积分 + 36000 索引 + 50 成员 + 200 Agent + 1500 QPM","企业级生产",{"plan":4295,"price":2210,"limit":4296,"cn_pay":1773,"note":4297},"云定制版","弹性资源 + 深度技术支持 + 专属客户经理","中大型企业","自托管开源免费 \u002F 云版 ¥0-¥599\u002F月",[4300],"onboarding\u002Ffastgpt-getting-started",[4302],"fastgpt-deep-review",{"power":164,"ux":164,"price":179,"cn_support":179,"stability":164},{"title":105,"description":4263},[4306,4308,4310,4312,4314],{"title":4307,"url":4219},"FastGPT 官网",{"title":4309,"url":4225},"FastGPT GitHub",{"title":4311,"url":3698},"FastGPT 定价页",{"title":4313,"url":3303},"FastGPT 2025 测评（南环 AI）",{"title":4315,"url":3277},"FastGPT 部署教程（腾讯云）","tools\u002Fagent\u002Fplatform\u002Ffastgpt",[4318,4319,4320,4321,4322],"企业内部知识库（员工手册、规章、流程）","产品文档智能问答（FAQ \u002F 用户手册）","垂直领域知识库（医疗、法律、金融）","数据严格不出网的合规场景","需要精细 RAG 流程编排（重排序、混合检索、阈值调节）","开源知识库问答系统，国内私有部署友好",[2243,3249,3250,3251,4325,4326,4327],"china","knowledge-base","labring","国内企业知识库私有化首选。RAG 召回工程做得很细，可视化调试好用，docker-compose 一键部署。生态插件不如 Dify 丰富。","RC9i5Gg2wX0bUcrC3Lfq5pHAeWLGi9KP2E_gZTefpe4",1783173060472]