[{"data":1,"prerenderedAt":3727},["ShallowReactive",2],{"header-counts":3,"footer-counts":6,"review-dify-deep-review":9,"review-related-dify-deep-review":770},{"tools":4,"reviews":5},77,25,{"tools":4,"reviews":5,"playbooks":7,"news":8},22,13,{"id":10,"title":11,"body":12,"cover":749,"description":750,"extension":751,"meta":752,"navigation":753,"path":754,"published":755,"relatedTools":756,"seo":760,"stem":761,"tags":762,"updated":755,"verdict":768,"__hash__":769},"review\u002Freview\u002Fdify-deep-review.md","Dify 深度评测：开源 AI Agent 平台私有部署首选？",{"type":13,"value":14,"toc":736},"minimark",[15,19,32,45,55,59,66,76,86,92,96,104,179,182,192,206,212,216,223,236,250,265,269,272,284,353,359,365,424,429,446,450,453,467,473,476,479,550,557,563,569,572,610,613,621,629,635,643,653,657,663,673,679,685,691,694,732],[16,17,18],"h2",{"id":18},"一句话结论",[20,21,22,23,27,28,31],"p",{},"如果你要做",[24,25,26],"strong",{},"企业级 AI 应用平台","——需要工作流编排、RAG 知识库、Agent、多模型混合调用，还要能私有部署——",[24,29,30],{},"Dify 在 2026 年是开源 LLMOps 平台的事实标准","。GitHub 13 万 star、累计 100 万+ 生产 app，1.0 版本把\"可视化工作流 + RAG 混合检索 + Agent + MCP 双向支持\"打包成一个 Docker Compose 能跑起来的东西。",[20,33,34,35,38,39,44],{},"但它不是万能的：",[24,36,37],{},"社区版和企业版差距比想象大、RAG 极致精度仍弱于 FastGPT、上手比 Coze 陡、自托管要懂运维","。它的甜点区是\"私有部署 + 复杂工作流 + 多模型混合\"，不是\"个人快速做个聊天机器人\"——后者去 ",[40,41,43],"a",{"href":42},"\u002Fagent\u002Fplatform\u002Fcoze.html","Coze"," 更快。",[46,47,52],"div",{"className":48},[49,50,51],"card","p-5","my-4",[20,53,54],{},"选型建议：数据必须不出内网 + 工作流复杂 → Dify 自托管。个人 \u002F 小团队快速原型 + 字节生态 → Coze。核心场景就是企业知识库 QA → FastGPT（RAG 精度更专）。重点连接 100+ 外部 SaaS → n8n。先用云版 Sandbox 跑 1 周 POC，再决定要不要上自托管。",[16,56,58],{"id":57},"dify-真正在解决的问题","Dify 真正在解决的问题",[20,60,61,62,65],{},"社区讨论\"为什么选 Dify\"经常停在\"它开源、它免费\"。但深一层看，Dify 是在解决",[24,63,64],{},"企业落地 LLM 应用的三个卡点","：",[20,67,68,71,72,75],{},[24,69,70],{},"第一个卡点：模型锁定。"," 用 ChatGPT API 做应用，绑死 OpenAI；用 Coze 做应用，绑死字节系模型。企业想\"今天用 GPT-5、明天切 Claude、后天接 DeepSeek 省钱\"，每换一家就要重写集成。Dify 通过插件市场接入 ",[24,73,74],{},"40+ 提供商","——OpenAI、Anthropic、Gemini、DeepSeek、Qwen、智谱、文心、豆包、Ollama 本地——同一个工作流里同时调多家，模型切换是配置而非重写。这是它在国内 toB 场景流行的关键。",[20,77,78,81,82,85],{},[24,79,80],{},"第二个卡点：数据合规。"," 金融、医疗、政府的数据不能出内网，但 SaaS 平台（Coze、OpenAI）的数据都要过云。Dify 的 ",[24,83,84],{},"Apache 2.0 开源自托管","让数据完全留在自己的服务器上，代码可审计、权限可自控。某些金融 \u002F 政府客户正是因此从 Coze 迁到 Dify。",[20,87,88,91],{},[24,89,90],{},"第三个卡点：LLM 应用的工程化。"," 从\"一段 prompt 调通\"到\"生产可用的 AI 应用\"，中间隔着工作流编排、RAG 检索、工具调用、监控告警、A\u002FB 测试一整套工程。Dify 把这套 LLMOps 能力可视化——拖拽编排、节点可观测、API 自动生成、多模型 A\u002FB——让 AI 应用从\"脚本\"变成\"可维护的系统\"。",[16,93,95],{"id":94},"工作流编排dify-的核心壁垒","工作流编排：Dify 的核心壁垒",[20,97,98,99,103],{},"工作流编排是 Dify 最强的一块，也是它和 ",[40,100,102],{"href":101},"\u002Fagent\u002Fplatform\u002Ffastgpt.html","FastGPT"," 拉开差距的地方。1.0 把应用拆成几种类型：",[105,106,107,123],"table",{},[108,109,110],"thead",{},[111,112,113,117,120],"tr",{},[114,115,116],"th",{},"类型",[114,118,119],{},"适合场景",[114,121,122],{},"编排范式",[124,125,126,140,153,166],"tbody",{},[111,127,128,134,137],{},[129,130,131],"td",{},[24,132,133],{},"Chatbot",[129,135,136],{},"简单对话机器人",[129,138,139],{},"prompt + tools",[111,141,142,147,150],{},[129,143,144],{},[24,145,146],{},"Agent",[129,148,149],{},"自主多步任务",[129,151,152],{},"ReAct \u002F Function Calling",[111,154,155,160,163],{},[129,156,157],{},[24,158,159],{},"Chatflow",[129,161,162],{},"对话型工作流（多轮 + 分支）",[129,164,165],{},"节点 DAG，带聊天上下文",[111,167,168,173,176],{},[129,169,170],{},[24,171,172],{},"Workflow",[129,174,175],{},"单次输入→输出（API 模式）",[129,177,178],{},"节点 DAG，无对话状态",[20,180,181],{},"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。这套节点的丰富度让 Dify 能拖拽出\"用 workflow 直接跑生产业务链路\"——不只是聊天机器人，而是\"AI 拆单 + 调 ERP + 通知飞书\"这种多步骤业务流程。",[20,183,184,187,188,191],{},[24,185,186],{},"1.0 的 Agent Workflow 升级","：把工作流从代码级提升到拖拽级，支持条件分支和循环节点。Agent 节点能自主决策、调用工具，配合条件分支做意图路由（A 意图走 Agent A、B 意图走 Agent B、其他走 Fallback）。这是对标 Coze 编排能力的关键一步，但 Dify 的差异化是",[24,189,190],{},"开源可自托管","。",[20,193,194,197,198,202,203,205],{},[24,195,196],{},"MCP 双向支持是 1.0 的隐藏大招","。Dify 既能作为 MCP Server 暴露工具（让 ",[40,199,201],{"href":200},"\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）。这意味着 Dify 不再是孤岛，而是能塞进 MCP 生态的中间层——这是 ",[40,204,43],{"href":42}," 给不了的开放性。",[20,207,208,211],{},[24,209,210],{},"多模型 A\u002FB 测试","：同一个 workflow 分流到不同模型对比效果，按用户满意度 \u002F 准确率 \u002F 延迟自动选优胜模型。这是生产级 LLMOps 的硬需求，让\"哪个模型更好\"从主观判断变成数据驱动。",[16,213,215],{"id":214},"rag-能力够用但不是最强","RAG 能力：够用但不是最强",[20,217,218,219,222],{},"Dify 内置完整 RAG 链路：上传文档（PDF \u002F Word \u002F Markdown \u002F 网页）→ 自动分块 + embedding → 混合检索（向量 + 全文 + 重排）→ 引用溯源。1.0 的 RAG 从纯向量升级到",[24,220,221],{},"混合检索 + 重排序","，召回率从 72% 提到 89%，Precision@5 从 65% 提到 91%——补上了之前 Dify RAG 的短板。",[20,224,225,228,229,235],{},[24,226,227],{},"但 RAG 极致精度仍弱于 FastGPT","。根据 ",[40,230,234],{"href":231,"rel":232},"https:\u002F\u002Fzhuanlan.zhihu.com\u002Fp\u002F1887141987838309480",[233],"nofollow","知乎 LLM 实战笔记对比"," 实测，Dify 社区版默认是基础语义检索，多路召回 + 重排在企业版才解锁。而 FastGPT 的 RAG 链路每一步都是可视化节点（问题预处理 → 检索策略 → 重排序 → 上下文组装 → 答案生成），每个环节可单独调参。做医疗、法律等复杂 domain 的 RAG，FastGPT 上限高一档。",[20,237,238,241,242,245,246,249],{},[24,239,240],{},"结论","：Dify 的 RAG 够用，且 1.0 补上了混合检索。但如果你的",[24,243,244],{},"核心场景就是企业知识库 QA","，且要在 RAG 精度上做深度调优，FastGPT 更专。Dify 胜在工作流而非纯 RAG——这也是为什么不少国内企业最终是 ",[24,247,248],{},"FastGPT 做知识库底座 + Dify 做业务应用层","的组合。",[20,251,252,255,256,260,261,264],{},[24,253,254],{},"避坑","：社区版 RAG 文件大小默认 15MB，超过会失败，要改 ",[257,258,259],"code",{},".env"," 的 ",[257,262,263],{},"UPLOAD_FILE_SIZE_LIMIT"," 并重启容器。代码节点的 Sandbox 性能差（启动慢、内存小），生产高频用建议改成 HTTP 节点调外部服务。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。",[16,266,268],{"id":267},"私有部署dify-最深的护城河","私有部署：Dify 最深的护城河",[20,270,271],{},"私有部署是 Dify 相对所有闭源 SaaS 平台最深的护城河，也是它能在金融 \u002F 医疗 \u002F 政府场景立足的根本。",[20,273,274,277,278,283],{},[24,275,276],{},"部署链路","：官方 ",[40,279,282],{"href":280,"rel":281},"https:\u002F\u002Fgithub.com\u002Flanggenius\u002Fdify",[233],"GitHub 仓库"," 提供 Docker Compose，社区版完全免费可商用：",[285,286,291],"pre",{"className":287,"code":288,"language":289,"meta":290,"style":290},"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","",[257,292,293,309,319,331,346],{"__ignoreMap":290},[294,295,298,302,306],"span",{"class":296,"line":297},"line",1,[294,299,301],{"class":300},"sScJk","git",[294,303,305],{"class":304},"sZZnC"," clone",[294,307,308],{"class":304}," https:\u002F\u002Fgithub.com\u002Flanggenius\u002Fdify.git\n",[294,310,312,316],{"class":296,"line":311},2,[294,313,315],{"class":314},"sj4cs","cd",[294,317,318],{"class":304}," dify\u002Fdocker\n",[294,320,322,325,328],{"class":296,"line":321},3,[294,323,324],{"class":300},"cp",[294,326,327],{"class":304}," .env.example",[294,329,330],{"class":304}," .env\n",[294,332,334,337,340,343],{"class":296,"line":333},4,[294,335,336],{"class":300},"docker",[294,338,339],{"class":304}," compose",[294,341,342],{"class":304}," up",[294,344,345],{"class":314}," -d\n",[294,347,349],{"class":296,"line":348},5,[294,350,352],{"class":351},"sJ8bj","# 默认 http:\u002F\u002Flocalhost\n",[20,354,355,358],{},[24,356,357],{},"硬件门槛","（社区共识）：最低 2 核 4G（纯外接 API 模式），推荐 4 核 8G + 30GB 磁盘，企业 8 核 16G+（单机日活上千）。",[20,360,361,364],{},[24,362,363],{},"真实 TCO 对比","（中小团队 3 年场景）：",[105,366,367,380],{},[108,368,369],{},[111,370,371,374,377],{},[114,372,373],{},"成本项",[114,375,376],{},"云版 Professional",[114,378,379],{},"自托管",[124,381,382,393,403,413],{},[111,383,384,387,390],{},[129,385,386],{},"平台费",[129,388,389],{},"~$2,100（3 年）",[129,391,392],{},"$0",[111,394,395,398,400],{},[129,396,397],{},"服务器",[129,399,392],{},[129,401,402],{},"~$50\u002F月 × 36 = $1,800",[111,404,405,408,411],{},[129,406,407],{},"模型 API",[129,409,410],{},"同",[129,412,410],{},[111,414,415,418,421],{},[129,416,417],{},"运维人力",[129,419,420],{},"0",[129,422,423],{},"约 0.2 人月",[20,425,426,428],{},[24,427,240],{},"：日活 \u003C 100 用云版省心；> 500 或数据敏感场景自托管 ROI 更好。而数据合规这条是云版给不了的——完全自托管时数据零外泄，这是某些客户从 Coze 迁到 Dify 的唯一原因。",[20,430,431,434,435,437,438,441,442,445],{},[24,432,433],{},"但要算清运维成本","：Dify 组件多（API、Web、Worker、Sandbox、DB、Redis、向量库），链路比 FastGPT 长。大版本升级会破坏数据库 schema，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 验证。",[257,436,259],{}," 改完要 ",[257,439,440],{},"docker compose down && up -d","，不是 ",[257,443,444],{},"restart","（后者不重新加载 env）。团队完全没运维能力的话，自托管会踩坑——这种场景云版还行。",[16,447,449],{"id":448},"模型支持40-提供商的开放生态","模型支持：40+ 提供商的开放生态",[20,451,452],{},"模型生态是 Dify 的另一核心壁垒。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。",[20,454,455,458,459,461,462,466],{},[24,456,457],{},"国产模型原生支持","是 Dify 在国内 toB 场景流行的关键——不像 ",[40,460,102],{"href":101}," 需要 ",[40,463,465],{"href":464},"\u002Fcoding\u002Fapi\u002Fone-api.html","OneAPI"," 中转，Dify 直接接 DeepSeek \u002F Qwen \u002F 智谱 \u002F 文心。这意味着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。",[20,468,469,472],{},[24,470,471],{},"API-first 是工程化的关键","。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。多平台发布（Web、API、Slack、Discord、微信公众号企业版）一次配置多处触达。",[16,474,475],{"id":475},"价格与运行成本",[20,477,478],{},"Dify 的价格结构对自托管团队极度友好：",[105,480,481,494],{},[108,482,483],{},[111,484,485,488,491],{},[114,486,487],{},"套餐",[114,489,490],{},"价格",[114,492,493],{},"主要限制",[124,495,496,506,517,528,539],{},[111,497,498,501,503],{},[129,499,500],{},"Self-hosted（开源版）",[129,502,392],{},[129,504,505],{},"全部核心功能，Apache 2.0 可商用",[111,507,508,511,514],{},[129,509,510],{},"Cloud Sandbox",[129,512,513],{},"免费",[129,515,516],{},"200 次调用，1 app，5MB 知识库",[111,518,519,522,525],{},[129,520,521],{},"Cloud Professional",[129,523,524],{},"$59\u002F月起",[129,526,527],{},"5000 调用\u002F月，多 app，50MB 知识库",[111,529,530,533,536],{},[129,531,532],{},"Cloud Team",[129,534,535],{},"$159\u002F月起",[129,537,538],{},"团队协作、SSO",[111,540,541,544,547],{},[129,542,543],{},"Enterprise",[129,545,546],{},"联系销售",[129,548,549],{},"定制 SLA、私有云",[20,551,552,553,556],{},"注意：云版价格只是 Dify 平台费，",[24,554,555],{},"模型 API 费用另算","（自带 OpenAI \u002F Anthropic key）。",[20,558,559,562],{},[24,560,561],{},"社区版 vs 企业版的差距比想象大","：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——RAG 精度、SSO、审计这些企业级能力社区版要么没有要么基础。这也是为什么\"数据必须不出内网 + 要企业级能力\"的客户最终会买企业版或自托管 + 社区版补充。",[20,564,565,568],{},[24,566,567],{},"国内使用","：云版 dify.ai 直连国内访问稳定但需国际信用卡；自托管 + 国产模型 = 完全国内闭环，是 Dify 在国内最大优势。Docker 镜像拉取国内可能慢，建议配阿里云 \u002F 网易 registry 镜像。",[16,570,571],{"id":571},"适用场景",[573,574,575,583,589,595,601,607],"ul",{},[576,577,578,579,582],"li",{},"✅ ",[24,580,581],{},"中大型企业 LLM 中台建设","——工作流 + RAG + Agent 三件套都要",[576,584,578,585,588],{},[24,586,587],{},"需要私有化部署","（金融 \u002F 医疗 \u002F 政府）——数据零外泄",[576,590,578,591,594],{},[24,592,593],{},"想做\"AI 工作流即产品\"的开发团队","——API-first、多平台发布",[576,596,578,597,600],{},[24,598,599],{},"多模型混合编排","——GPT + Claude + DeepSeek + 本地模型同流",[576,602,578,603,606],{},[24,604,605],{},"复杂业务工作流","——拖拽实现\"拆单 + 调 ERP + 通知飞书\"",[576,608,609],{},"✅ 已接受 Docker + 一定运维投入的团队",[16,611,612],{"id":612},"不推荐场景",[20,614,615,65,618,620],{},[24,616,617],{},"纯个人玩家做对话机器人",[40,619,43],{"href":42}," 更快、上手更简单。Dify 的部署和配置对个人玩家是负担，杀鸡用牛刀。",[20,622,623,65,626,628],{},[24,624,625],{},"只想做企业知识库 QA",[40,627,102],{"href":101}," 的 RAG 精度更专、链路每步可调。Dify 的 RAG 够用但不是最强，纯知识库场景 FastGPT 上限更高。",[20,630,631,634],{},[24,632,633],{},"团队完全没运维能力","：自托管会踩坑（Docker、PostgreSQL、向量库、env 配置、版本升级）。云版还行，但要数据不出内网就必须自托管——这种矛盾场景要么招运维、要么换 SaaS 方案。",[20,636,637,65,640,642],{},[24,638,639],{},"需要深度对接字节飞书 \u002F 抖音",[40,641,43],{"href":42}," 原生集成字节生态，Dify 没有这种深度绑定。如果你的业务全在飞书 \u002F 抖音里，Coze 更顺。",[20,644,645,65,648,652],{},[24,646,647],{},"工作流核心是连接 100+ 外部 SaaS",[40,649,651],{"href":650},"\u002Fagent\u002Fplatform\u002Fn8n.html","n8n"," 的节点更全（Slack \u002F Notion \u002F 数据库 \u002F 几百个 SaaS 连接器）。Dify 的 HTTP 节点能接外部 API，但 SaaS 连接器数量不如 n8n。",[16,654,656],{"id":655},"faq","FAQ",[20,658,659,662],{},[24,660,661],{},"Q：Dify 1.0 升级了什么？","\nA：三个核心升级：Agent Workflow 可视化编排（拖拽配置条件和循环）、MCP 双向支持（既能暴露工具给 Claude\u002FCursor，也能消费外部 MCP Server）、RAG 重构（混合检索 + 重排序，准确率提升约 30%）。叙事从\"管 LLM 调用\"变成\"编排 Agent 工作流\"。",[20,664,665,668,669,191],{},[24,666,667],{},"Q：Dify 和 FastGPT 怎么选？","\nA：核心是知识库 QA 选 FastGPT（RAG 精度更专、每步可调），综合 LLMOps + 复杂工作流 + 多平台发布选 Dify。不少国内企业的做法是 FastGPT 做知识库底座 + Dify 做业务应用层，通过 API 打通。详见 ",[40,670,672],{"href":671},"\u002Fcompare\u002Ffastgpt-vs-dify.html","FastGPT vs Dify 对比",[20,674,675,678],{},[24,676,677],{},"Q：社区版能商用吗？","\nA：能。Dify 社区版是 Apache 2.0 协议，完全免费可商用。但社区版和企业版差距比想象大——多路召回 \u002F 重排序 \u002F SSO \u002F 审计日志都在企业版。社区版做生产前要评估这些能力够不够。",[20,680,681,684],{},[24,682,683],{},"Q：自托管部署难吗？","\nA：Docker Compose 一键起，但组件多（API、Web、Worker、Sandbox、DB、Redis、向量库），链路比 FastGPT 长。大版本升级会破坏数据库 schema，要备份 + staging 验证。有 Docker + PostgreSQL 运维基础的团队能搞定，完全没运维能力会踩坑。",[20,686,687,690],{},[24,688,689],{},"Q：Dify 支持 MCP 吗？","\nA：1.0 起双向支持。Dify 可作为 MCP Server 暴露工具（让 Claude Code \u002F Cursor 调用 Dify workflow），也能消费外部 MCP Server（在 workflow 里调 GitHub \u002F Slack \u002F 自家系统）。这是 Dify 区别于 Coze 的开放性优势。",[16,692,693],{"id":693},"相关阅读",[573,695,696,702,707,713,722],{},[576,697,698],{},[40,699,701],{"href":700},"\u002Fagent\u002Fplatform\u002Fdify.html","Dify 工具卡：开源 LLMOps 平台",[576,703,704],{},[40,705,706],{"href":671},"FastGPT vs Dify：国内企业级 RAG 与 Agent 平台怎么选",[576,708,709],{},[40,710,712],{"href":711},"\u002Freview\u002Fcoze-vs-dify.html","Coze vs Dify：国内 Agent 平台对比",[576,714,715,718,719],{},[40,716,717],{"href":101},"FastGPT 工具卡"," · ",[40,720,721],{"href":42},"Coze 工具卡",[576,723,724,718,728],{},[40,725,727],{"href":726},"\u002Fwiki\u002Frag.html","什么是 RAG",[40,729,731],{"href":730},"\u002Fwiki\u002Fmcp.html","什么是 MCP",[733,734,735],"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":290,"searchDepth":321,"depth":321,"links":737},[738,739,740,741,742,743,744,745,746,747,748],{"id":18,"depth":311,"text":18},{"id":57,"depth":311,"text":58},{"id":94,"depth":311,"text":95},{"id":214,"depth":311,"text":215},{"id":267,"depth":311,"text":268},{"id":448,"depth":311,"text":449},{"id":475,"depth":311,"text":475},{"id":571,"depth":311,"text":571},{"id":612,"depth":311,"text":612},{"id":655,"depth":311,"text":656},{"id":693,"depth":311,"text":693},"\u002Fog\u002Freview\u002Fdify.png","Dify 真实评测：GitHub 13 万 star 的开源 LLMOps 平台，1.0 把工作流编排、RAG 知识库、Agent、MCP 双向支持打包成 Docker Compose 能跑起来的东西。本文写它真正解决的问题、工作流编排深度、RAG 混合检索能力、私有部署经济学、40+ 模型生态，以及 5 类不推荐场景。AIHO 编辑部基于官方文档与多份评测整理。","md",{},true,"\u002Freview\u002Fdify-deep-review","2026-07-04",[757,758,759],"agent\u002Fplatform\u002Fdify","agent\u002Fplatform\u002Ffastgpt","agent\u002Fplatform\u002Fcoze",{"title":11,"description":750},"review\u002Fdify-deep-review",[763,764,765,766,767],"Dify","AI Agent","开源","私有部署","深度评测","开源 LLMOps 平台的事实标准。GitHub 13 万 star、完全开源 + 模型不挑食 + MCP 双向支持是核心壁垒。比 Coze 工程化、上手陡一点；比 FastGPT 工作流强、RAG 略弱。私有部署 + 复杂工作流 + 多模型混合编排场景的首选，纯知识库 QA 场景 FastGPT 更专。","rpxaPstbcG-SkuDrDLJlKxTPjtbqChbUAQXIeSrAhT0",[771,1680,2649],{"id":772,"title":43,"alternatives":773,"api_compatible":775,"body":776,"category":1602,"chinese_friendly":348,"cover":1603,"description":1604,"domestic":1605,"extension":751,"faq":1606,"free":1605,"github":1606,"languages":1607,"meta":1610,"models":1611,"navigation":753,"notSuitable":1617,"opensource":1605,"path":1621,"pillar":1622,"platforms":1623,"priceTable":1625,"pricing":1646,"published":1647,"relatedPlaybooks":1606,"relatedReviews":1606,"score":1648,"self_host":1605,"seo":1649,"seoTitle":1650,"slug":759,"sources":1651,"stem":1662,"suitable":1663,"tagline":1669,"tags":1670,"updated":1677,"verdict":1678,"website":796,"__hash__":1679},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Fcoze.md",[757,758,774],"agent\u002Fplatform\u002Fyuanqi",[],{"type":13,"value":777,"toc":1586},[778,782,819,825,828,833,836,850,859,909,918,922,925,957,965,969,972,992,996,1078,1087,1089,1092,1112,1121,1125,1162,1165,1168,1307,1323,1353,1356,1429,1433,1436,1453,1456,1486,1488,1549,1552,1579],[16,779,781],{"id":780},"tldr","TL;DR",[46,783,785,806],{"className":784},[49,50,51],[20,786,787,790,791,805],{},[24,788,789],{},"一句话："," 字节跳动出品的低代码 Agent 平台，",[24,792,793,794,799,800],{},"国内版 ",[40,795,798],{"href":796,"rel":797},"https:\u002F\u002Fwww.coze.cn",[233],"coze.cn","（中文叫\"扣子\"）+ 国际版 ",[40,801,804],{"href":802,"rel":803},"https:\u002F\u002Fwww.coze.com",[233],"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。",[20,807,808,809,812,813,818],{},"最大价值是 ",[24,810,811],{},"零代码、最快上手","——文档详细 + 中文社区活跃 + 模板多，根据 ",[40,814,817],{"href":815,"rel":816},"https:\u002F\u002Fwww.cnblogs.com\u002Fuulucias\u002Fp\u002F19449008",[233],"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\"。",[820,821,822],"blockquote",{},[20,823,824],{},"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。",[16,826,827],{"id":827},"核心特性",[829,830,832],"h3",{"id":831},"可视化工作流最大卖点","可视化工作流（最大卖点）",[20,834,835],{},"Coze 把 Agent 拆成两层：",[573,837,838,844],{},[576,839,840,843],{},[24,841,842],{},"Bot \u002F 智能体","：对话式 AI，配置 prompt + 知识库 + 插件",[576,845,846,849],{},[24,847,848],{},"工作流（Workflow）","：DAG 节点编排，可被 Bot 调用，也可独立部署",[20,851,852,853,858],{},"工作流节点类型（基于 ",[40,854,857],{"href":855,"rel":856},"https:\u002F\u002Fdeveloper.volcengine.com\u002Farticles\u002F7530117616687480851",[233],"火山引擎社区 2025 实战","）：",[573,860,861,867,873,879,885,891,897,903],{},[576,862,863,866],{},[24,864,865],{},"开始 \u002F 结束节点","：输入输出",[576,868,869,872],{},[24,870,871],{},"大模型节点","：调豆包 \u002F GPT \u002F Claude 任一模型",[576,874,875,878],{},[24,876,877],{},"代码节点","：内嵌 Python \u002F JavaScript（飞书插件常需要数据格式转换）",[576,880,881,884],{},[24,882,883],{},"循环节点","：批量处理多条数据",[576,886,887,890],{},[24,888,889],{},"条件节点","：分支判断",[576,892,893,896],{},[24,894,895],{},"插件节点","：调 Coze 插件市场的工具",[576,898,899,902],{},[24,900,901],{},"知识库节点","：RAG 检索",[576,904,905,908],{},[24,906,907],{},"HTTP 节点","：调外部 API",[20,910,911,912,917],{},"典型用例（参考 ",[40,913,916],{"href":914,"rel":915},"https:\u002F\u002Fwww.toutiao.com\u002Farticle\u002F7469986334686315017",[233],"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。",[829,919,921],{"id":920},"插件市场200-官方-海量第三方","插件市场（200+ 官方 + 海量第三方）",[20,923,924],{},"Coze 的\"插件\"是封装好的 API 工具，比如：",[573,926,927,933,939,945,951],{},[576,928,929,932],{},[24,930,931],{},"飞书多维表格","：增删改查记录（国内 toB 场景的杀手锏）",[576,934,935,938],{},[24,936,937],{},"图像生成","：调豆包 \u002F SD \u002F DALL-E",[576,940,941,944],{},[24,942,943],{},"联网搜索","：实时网页检索",[576,946,947,950],{},[24,948,949],{},"代码执行","：在线运行 Python",[576,952,953,956],{},[24,954,955],{},"第三方 SaaS","：微博、抖音、bilibili、Notion……",[20,958,959,964],{},[40,960,963],{"href":961,"rel":962},"https:\u002F\u002Fkouziai.github.io\u002F",[233],"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。",[829,966,968],{"id":967},"bot-商店-多平台一键发布","Bot 商店 + 多平台一键发布",[20,970,971],{},"发布渠道：",[573,973,974,977,980,983,986,989],{},[576,975,976],{},"飞书机器人（一键绑）",[576,978,979],{},"抖音 \u002F 头条号",[576,981,982],{},"微信小程序 \u002F 公众号（部分需企业认证）",[576,984,985],{},"自定义网页嵌入",[576,987,988],{},"API 接口（提供 OpenAPI 风格 REST 调用）",[576,990,991],{},"Discord（国际版）",[829,993,995],{"id":994},"国内版-vs-国际版","国内版 vs 国际版",[105,997,998,1011],{},[108,999,1000],{},[111,1001,1002,1005,1008],{},[114,1003,1004],{},"维度",[114,1006,1007],{},"扣子（coze.cn）",[114,1009,1010],{},"Coze（coze.com）",[124,1012,1013,1024,1035,1045,1056,1067],{},[111,1014,1015,1018,1021],{},[129,1016,1017],{},"主力模型",[129,1019,1020],{},"豆包 Pro \u002F DeepSeek \u002F Qwen \u002F Kimi",[129,1022,1023],{},"OpenAI \u002F Claude \u002F Gemini \u002F Cohere",[111,1025,1026,1029,1032],{},[129,1027,1028],{},"飞书 \u002F 抖音 \u002F 微信集成",[129,1030,1031],{},"✅ 原生",[129,1033,1034],{},"❌",[111,1036,1037,1040,1043],{},[129,1038,1039],{},"Discord \u002F Slack 集成",[129,1041,1042],{},"⚠️ 有限",[129,1044,1031],{},[111,1046,1047,1050,1053],{},[129,1048,1049],{},"数据存储位置",[129,1051,1052],{},"国内",[129,1054,1055],{},"海外",[111,1057,1058,1061,1064],{},[129,1059,1060],{},"支付",[129,1062,1063],{},"微信 \u002F 支付宝",[129,1065,1066],{},"海外信用卡",[111,1068,1069,1072,1075],{},[129,1070,1071],{},"内容合规",[129,1073,1074],{},"严格审核",[129,1076,1077],{},"宽松",[20,1079,1080,1083,1084,191],{},[24,1081,1082],{},"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 ",[24,1085,1086],{},"不互通",[16,1088,475],{"id":475},[20,1090,1091],{},"国内版（扣子）：",[573,1093,1094,1100,1106],{},[576,1095,1096,1099],{},[24,1097,1098],{},"免费版","：免费模型有日额度（豆包 lite 等），适合个人玩 \u002F Demo",[576,1101,1102,1105],{},[24,1103,1104],{},"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",[576,1107,1108,1111],{},[24,1109,1110],{},"企业版","：议价，含 VPC、私有化（限定场景）、SLA",[20,1113,1114,1115,1120],{},"国际版（coze.com）的定价模式据 ",[40,1116,1119],{"href":1117,"rel":1118},"https:\u002F\u002Fdocs.coze.com\u002F",[233],"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。",[16,1122,1124],{"id":1123},"上手-10-分钟","上手 10 分钟",[1126,1127,1128,1141,1144,1147,1150,1153,1156,1159],"ol",{},[576,1129,1130,1131,1135,1136,1140],{},"打开 ",[40,1132,1134],{"href":796,"rel":1133},[233],"www.coze.cn","（国内）或 ",[40,1137,1139],{"href":802,"rel":1138},[233],"www.coze.com","（国际），用飞书 \u002F 抖音账号 \u002F Google 账号登录",[576,1142,1143],{},"左侧\"工作空间\" → \"+创建 Bot\"，起个名字",[576,1145,1146],{},"选模型（国内推荐豆包 Pro，国际推 Claude Sonnet 4）",[576,1148,1149],{},"写 Bot 角色 prompt",[576,1151,1152],{},"可选：上传 PDF \u002F 文档建知识库",[576,1154,1155],{},"测试一下对话效果",[576,1157,1158],{},"右上\"发布\" → 选渠道（飞书 \u002F 抖音 \u002F API \u002F Web）",[576,1160,1161],{},"拿到调用 URL \u002F 飞书机器人 webhook",[20,1163,1164],{},"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。",[16,1166,1167],{"id":1167},"与同类怎么选",[105,1169,1170,1192],{},[108,1171,1172],{},[111,1173,1174,1176,1178,1182,1186],{},[114,1175,1004],{},[114,1177,43],{},[114,1179,1180],{},[40,1181,763],{"href":700},[114,1183,1184],{},[40,1185,102],{"href":101},[114,1187,1188],{},[40,1189,1191],{"href":1190},"\u002Fagent\u002Fplatform\u002Fyuanqi.html","元器 yuanqi",[124,1193,1194,1207,1221,1237,1253,1266,1280,1294],{},[111,1195,1196,1198,1200,1203,1205],{},[129,1197,765],{},[129,1199,1034],{},[129,1201,1202],{},"✅",[129,1204,1202],{},[129,1206,1034],{},[111,1208,1209,1211,1214,1216,1218],{},[129,1210,766],{},[129,1212,1213],{},"⚠️ 仅企业版",[129,1215,1202],{},[129,1217,1202],{},[129,1219,1220],{},"⚠️",[111,1222,1223,1226,1229,1232,1234],{},[129,1224,1225],{},"上手难度",[129,1227,1228],{},"★ 最简单",[129,1230,1231],{},"★★★",[129,1233,1231],{},[129,1235,1236],{},"★★",[111,1238,1239,1242,1245,1248,1251],{},[129,1240,1241],{},"工作流编排",[129,1243,1244],{},"★★★★☆",[129,1246,1247],{},"★★★★★",[129,1249,1250],{},"★★★☆☆",[129,1252,1250],{},[111,1254,1255,1258,1260,1262,1264],{},[129,1256,1257],{},"RAG 精度",[129,1259,1250],{},[129,1261,1244],{},[129,1263,1247],{},[129,1265,1250],{},[111,1267,1268,1271,1273,1275,1277],{},[129,1269,1270],{},"字节生态",[129,1272,1247],{},[129,1274,1034],{},[129,1276,1034],{},[129,1278,1279],{},"❌（腾讯系）",[111,1281,1282,1285,1288,1290,1292],{},[129,1283,1284],{},"插件市场",[129,1286,1287],{},"★★★★★ 200+",[129,1289,1250],{},[129,1291,1250],{},[129,1293,1250],{},[111,1295,1296,1299,1301,1303,1305],{},[129,1297,1298],{},"中文社区",[129,1300,1247],{},[129,1302,1244],{},[129,1304,1244],{},[129,1306,1244],{},[20,1308,1309,1312,1313,1318,1319,858],{},[24,1310,1311],{},"怎么选","（综合 ",[40,1314,1317],{"href":1315,"rel":1316},"https:\u002F\u002Fwww.besthub.dev\u002Farticles\u002Fcoze-vs-dify-vs-fastgpt-which-ai-agent-platform-fits-your-needs-fa59cf97b798",[233],"BestHub 2025-07"," 和 ",[40,1320,1322],{"href":815,"rel":1321},[233],"博客园 2026-01",[573,1324,1325,1331,1339,1345],{},[576,1326,1327,1330],{},[24,1328,1329],{},"快速验证 \u002F 不懂代码 \u002F 1-2 天出原型"," → Coze",[576,1332,1333,1336,1337],{},[24,1334,1335],{},"数据安全要求高 \u002F 复杂业务流程 \u002F 有技术团队"," → ",[40,1338,763],{"href":700},[576,1340,1341,1336,1343],{},[24,1342,244],{},[40,1344,102],{"href":101},[576,1346,1347,1336,1350],{},[24,1348,1349],{},"QQ \u002F 微信生态 + 腾讯系",[40,1351,1352],{"href":1190},"元器",[16,1354,1355],{"id":1355},"避坑清单",[573,1357,1358,1368,1374,1389,1399,1405,1417,1423],{},[576,1359,1360,1363,1364,1367],{},[24,1361,1362],{},"国内版 vs 国际版的\"双账号陷阱\"","：扣子（coze.cn）和 Coze（coze.com）是",[24,1365,1366],{},"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",[576,1369,1370,1373],{},[24,1371,1372],{},"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",[576,1375,1376,1379,1380,1383,1384,1388],{},[24,1377,1378],{},"飞书多维表格插件数据格式坑","：写入多维表格需要 ",[257,1381,1382],{},"Array\u003CObject>"," 格式，代码节点要做转换（参考 ",[40,1385,1387],{"href":855,"rel":1386},[233],"火山引擎 2025 教程","）",[576,1390,1391,1394,1395,1398],{},[24,1392,1393],{},"工作流读取飞书表格默认 20 条","：要改 ",[257,1396,1397],{},"page_size","，最大 500 条；超过 500 要分页或循环",[576,1400,1401,1404],{},[24,1402,1403],{},"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",[576,1406,1407,1410,1411,1413,1414,1416],{},[24,1408,1409],{},"企业版\"私有化\"是有限的","：完全数据不出网仍建议 ",[40,1412,763],{"href":700}," \u002F ",[40,1415,102],{"href":101}," 自托管",[576,1418,1419,1422],{},[24,1420,1421],{},"国际版接 Claude \u002F GPT 需要 BYOK","：自己绑海外信用卡，平台不代付",[576,1424,1425,1428],{},[24,1426,1427],{},"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",[16,1430,1432],{"id":1431},"适合-不适合","适合 \u002F 不适合",[20,1434,1435],{},"✅ 适合：",[573,1437,1438,1441,1444,1447,1450],{},[576,1439,1440],{},"产品 \u002F 运营 \u002F 非技术人员快速做 Bot",[576,1442,1443],{},"在飞书 \u002F 抖音 \u002F 头条生态内做集成",[576,1445,1446],{},"个人副业（小红书账号批量内容生成等）",[576,1448,1449],{},"中小企业客服 Bot（3 天上线）",[576,1451,1452],{},"想用豆包 \u002F DeepSeek 国产模型的人",[20,1454,1455],{},"❌ 不适合：",[573,1457,1458,1461,1464,1469,1476],{},[576,1459,1460],{},"金融 \u002F 政府 \u002F 医疗（数据敏感，需自托管）",[576,1462,1463],{},"复杂业务系统深度集成（自由度不够）",[576,1465,1466,1467,1388],{},"反感字节生态（去 ",[40,1468,763],{"href":700},[576,1470,1471,1472,1413,1474,1388],{},"希望开源 \u002F 完全自主可控（去 ",[40,1473,763],{"href":700},[40,1475,102],{"href":101},[576,1477,1478,1479,1413,1481,1485],{},"海外 toB SaaS 产品后端（",[40,1480,763],{"href":700},[40,1482,1484],{"href":1483},"\u002Fagent\u002Fplatform\u002Flangflow.html","Langflow"," 更合适）",[16,1487,693],{"id":693},[573,1489,1490,1503,1519,1538],{},[576,1491,1492,1493,1413,1495,1413,1497,1413,1499,1413,1501],{},"同类对比：",[40,1494,763],{"href":700},[40,1496,102],{"href":101},[40,1498,1352],{"href":1190},[40,1500,1484],{"href":1483},[40,1502,651],{"href":650},[576,1504,1505,1506,1413,1509,1413,1512,1413,1516],{},"概念：",[40,1507,764],{"href":1508},"\u002Fwiki\u002Fai-agent.html",[40,1510,1511],{"href":726},"RAG",[40,1513,1515],{"href":1514},"\u002Fwiki\u002Ffunction-calling.html","Function Calling",[40,1517,1518],{"href":1508},"Multi-Agent",[576,1520,1521,1522,1413,1526,1413,1530,1413,1534],{},"模型：",[40,1523,1525],{"href":1524},"\u002Fmodels\u002Fdoubao-1-5-pro.html","豆包 Doubao",[40,1527,1529],{"href":1528},"\u002Fmodels\u002Fdeepseek-v3.html","DeepSeek-V3",[40,1531,1533],{"href":1532},"\u002Fmodels\u002Fqwen-3.html","Qwen3",[40,1535,1537],{"href":1536},"\u002Fmodels\u002Fglm-5.2.html","GLM-5.2",[576,1539,1540,1541,1413,1545],{},"进阶：",[40,1542,1544],{"href":1543},"\u002Fwiki\u002Fprompt-engineering.html","Prompt Engineering",[40,1546,1548],{"href":1547},"\u002Fwiki\u002Fcontext-engineering.html","Context Engineering",[16,1550,1551],{"id":1551},"来源",[573,1553,1554,1560,1566,1573,1576],{},[576,1555,1556,1557],{},"国内版：",[40,1558,796],{"href":796,"rel":1559},[233],[576,1561,1562,1563],{},"国际版：",[40,1564,802],{"href":802,"rel":1565},[233],[576,1567,1568,1569],{},"官方文档：",[40,1570,1571],{"href":1571,"rel":1572},"https:\u002F\u002Fdocs.coze.com",[233],[576,1574,1575],{},"第三方选型评测：cnblogs.com \u002F besthub.dev \u002F aibotgo.net",[576,1577,1578],{},"实战教程：火山引擎社区、今日头条 涛哥讲AI",[20,1580,1581,1582,1585],{},"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 ",[40,1583,1584],{"href":1584},"\u002Fsubmit"," 反馈。",{"title":290,"searchDepth":321,"depth":321,"links":1587},[1588,1589,1595,1596,1597,1598,1599,1600,1601],{"id":780,"depth":311,"text":781},{"id":827,"depth":311,"text":827,"children":1590},[1591,1592,1593,1594],{"id":831,"depth":321,"text":832},{"id":920,"depth":321,"text":921},{"id":967,"depth":321,"text":968},{"id":994,"depth":321,"text":995},{"id":475,"depth":311,"text":475},{"id":1123,"depth":311,"text":1124},{"id":1167,"depth":311,"text":1167},{"id":1355,"depth":311,"text":1355},{"id":1431,"depth":311,"text":1432},{"id":693,"depth":311,"text":693},{"id":1551,"depth":311,"text":1551},"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,[1608,1609],"zh","en",{},[1612,1613,1614,1615,1616],"doubao-pro","deepseek-v3","qwen-max","gpt-4o (国际版)","claude (国际版)",[1618,1619,1620],"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",[1624],"web",[1626,1632,1637,1641],{"plan":1627,"price":1628,"limit":1629,"cn_pay":1630,"note":1631},"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":1104,"price":1633,"limit":1634,"cn_pay":1635,"note":1636},"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":1110,"price":1638,"limit":1639,"cn_pay":1202,"note":1640},"议价","VPC、合规、SLA、私有化（部分）","B 端落地",{"plan":1642,"price":1643,"cn_pay":1644,"note":1645,"limit":1630},"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":333,"ux":348,"price":333,"cn_support":348,"stability":333},{"title":43,"description":1604},"Coze 扣子评测 2026：字节 AI Agent 平台，对比 Dify",[1652,1654,1656,1658,1660],{"title":1653,"url":796},"Coze 国内版（扣子）",{"title":1655,"url":802},"Coze 国际版",{"title":1657,"url":1571},"Coze 官方文档",{"title":1659,"url":815},"Coze vs Dify vs FastGPT 选型 2026",{"title":1661,"url":1315},"BestHub 三平台对比","tools\u002Fagent\u002Fplatform\u002Fcoze",[1664,1665,1666,1667,1668],"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",[1671,1672,1673,1674,1675,1676],"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":1681,"title":763,"alternatives":1682,"api_compatible":1606,"body":1685,"category":1602,"chinese_friendly":333,"cover":2598,"description":2599,"domestic":1605,"extension":751,"faq":1606,"free":1605,"github":280,"languages":2600,"meta":2602,"models":1606,"navigation":753,"notSuitable":1606,"opensource":753,"path":2603,"pillar":1622,"platforms":2604,"priceTable":2608,"pricing":2623,"published":1647,"relatedPlaybooks":1606,"relatedReviews":1606,"score":2624,"self_host":753,"seo":2625,"seoTitle":2626,"slug":757,"sources":2627,"stem":2639,"suitable":1606,"tagline":2640,"tags":2641,"updated":1677,"verdict":2647,"website":2545,"__hash__":2648},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Fdify.md",[759,758,1683,1684],"agent\u002Fplatform\u002Fn8n","agent\u002Fplatform\u002Flangflow",{"type":13,"value":1686,"toc":2576},[1687,1689,1706,1711,1713,1717,1720,1774,1777,1781,1784,1798,1813,1817,1820,1828,1832,1839,1850,1854,1857,1859,1863,1872,1922,1926,1930,1937,1979,1982,2002,2006,2009,2057,2062,2064,2157,2160,2189,2191,2319,2331,2366,2368,2434,2436,2438,2456,2458,2487,2489,2536,2538,2569,2574],[16,1688,781],{"id":780},[46,1690,1692,1697],{"className":1691},[49,50,51],[20,1693,1694,1696],{},[24,1695,789],{}," 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。",[20,1698,808,1699,1702,1703,1705],{},[24,1700,1701],{},"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 ",[40,1704,43],{"href":42}," 折腾，新手得读 1-2 小时文档。",[820,1707,1708],{},[20,1709,1710],{},"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。",[16,1712,827],{"id":827},[829,1714,1716],{"id":1715},"可视化工作流chatflow-workflow","可视化工作流（Chatflow + Workflow）",[20,1718,1719],{},"Dify 把 LLM 应用拆成两种\"应用类型\"：",[105,1721,1722,1732],{},[108,1723,1724],{},[111,1725,1726,1728,1730],{},[114,1727,116],{},[114,1729,119],{},[114,1731,122],{},[124,1733,1734,1744,1754,1764],{},[111,1735,1736,1740,1742],{},[129,1737,1738],{},[24,1739,133],{},[129,1741,136],{},[129,1743,139],{},[111,1745,1746,1750,1752],{},[129,1747,1748],{},[24,1749,146],{},[129,1751,149],{},[129,1753,152],{},[111,1755,1756,1760,1762],{},[129,1757,1758],{},[24,1759,159],{},[129,1761,162],{},[129,1763,165],{},[111,1765,1766,1770,1772],{},[129,1767,1768],{},[24,1769,172],{},[129,1771,175],{},[129,1773,178],{},[20,1775,1776],{},"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\"。",[829,1778,1780],{"id":1779},"rag-知识库","RAG 知识库",[20,1782,1783],{},"内置完整 RAG 链路：",[1126,1785,1786,1789,1792,1795],{},[576,1787,1788],{},"上传文档（PDF \u002F Word \u002F Markdown \u002F 网页）",[576,1790,1791],{},"自动分块 + embedding（可配置分段策略和 embedding 模型）",[576,1793,1794],{},"混合检索（向量 + 全文 + 重排）",[576,1796,1797],{},"引用溯源（回答末尾自动附原文片段）",[20,1799,1800,1801,1805,1806,1809,1810,1812],{},"注意：根据 ",[40,1802,1804],{"href":231,"rel":1803},[233],"知乎 LLM 实战笔记 2025-03 对比"," 的实测，Dify ",[24,1807,1808],{},"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 ",[40,1811,102],{"href":101},"（实测准确率高 10+ 个百分点），Dify 胜在工作流而非纯 RAG。",[829,1814,1816],{"id":1815},"模型生态40-提供商","模型生态：40+ 提供商",[20,1818,1819],{},"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。",[20,1821,1822,1823,461,1825,1827],{},"国产模型原生支持（不像 ",[40,1824,102],{"href":101},[40,1826,465],{"href":464}," 中转），是 Dify 在国内 toB 场景流行的关键。",[829,1829,1831],{"id":1830},"mcp-协议支持","MCP 协议支持",[20,1833,1834,1835,1838],{},"Dify 较早接入了 ",[40,1836,1837],{"href":730},"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：",[573,1840,1841,1844,1847],{},[576,1842,1843],{},"通过 MCP 调本地 PostgreSQL \u002F SQLite",[576,1845,1846],{},"通过 MCP 调 GitHub \u002F Slack \u002F Linear",[576,1848,1849],{},"通过 MCP 调自家内部系统（写一个 MCP Server 即可）",[829,1851,1853],{"id":1852},"api-first","API-first",[20,1855,1856],{},"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。",[16,1858,475],{"id":475},[829,1860,1862],{"id":1861},"云版difyai","云版（dify.ai）",[20,1864,1865,1866,1871],{},"根据 ",[40,1867,1870],{"href":1868,"rel":1869},"https:\u002F\u002Fwww.tooljunction.io\u002Fai-tools\u002Fdify-ai",[233],"tooljunction.io 2026 评测"," 引用的官方定价：",[105,1873,1874,1884],{},[108,1875,1876],{},[111,1877,1878,1880,1882],{},[114,1879,487],{},[114,1881,490],{},[114,1883,493],{},[124,1885,1886,1896,1905,1914],{},[111,1887,1888,1891,1893],{},[129,1889,1890],{},"Sandbox",[129,1892,513],{},[129,1894,1895],{},"200 次模型调用，1 app，5MB 知识库",[111,1897,1898,1901,1903],{},[129,1899,1900],{},"Professional",[129,1902,524],{},[129,1904,527],{},[111,1906,1907,1910,1912],{},[129,1908,1909],{},"Team",[129,1911,535],{},[129,1913,538],{},[111,1915,1916,1918,1920],{},[129,1917,543],{},[129,1919,546],{},[129,1921,549],{},[20,1923,552,1924,556],{},[24,1925,555],{},[829,1927,1929],{"id":1928},"自托管推荐","自托管（推荐）",[20,1931,1932,1936],{},[40,1933,1935],{"href":280,"rel":1934},[233],"官方 GitHub 仓库"," 提供 Docker Compose 部署，社区版完全免费可商用：",[285,1938,1940],{"className":287,"code":1939,"language":289,"meta":290,"style":290},"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",[257,1941,1942,1950,1956,1964,1974],{"__ignoreMap":290},[294,1943,1944,1946,1948],{"class":296,"line":297},[294,1945,301],{"class":300},[294,1947,305],{"class":304},[294,1949,308],{"class":304},[294,1951,1952,1954],{"class":296,"line":311},[294,1953,315],{"class":314},[294,1955,318],{"class":304},[294,1957,1958,1960,1962],{"class":296,"line":321},[294,1959,324],{"class":300},[294,1961,327],{"class":304},[294,1963,330],{"class":304},[294,1965,1966,1968,1970,1972],{"class":296,"line":333},[294,1967,336],{"class":300},[294,1969,339],{"class":304},[294,1971,342],{"class":304},[294,1973,345],{"class":314},[294,1975,1976],{"class":296,"line":348},[294,1977,1978],{"class":351},"# 默认 http:\u002F\u002Flocalhost \u002F 端口可在 .env 调整\n",[20,1980,1981],{},"硬件门槛（社区共识，非官方硬性要求）：",[573,1983,1984,1990,1996],{},[576,1985,1986,1989],{},[24,1987,1988],{},"最低","：2 核 4G，纯外接 API 模式",[576,1991,1992,1995],{},[24,1993,1994],{},"推荐","：4 核 8G + 至少 30GB 磁盘（向量数据 + 文件存储）",[576,1997,1998,2001],{},[24,1999,2000],{},"企业","：8 核 16G+，单机日活上千",[829,2003,2005],{"id":2004},"真实-tco","真实 TCO",[20,2007,2008],{},"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）：",[105,2010,2011,2021],{},[108,2012,2013],{},[111,2014,2015,2017,2019],{},[114,2016,373],{},[114,2018,376],{},[114,2020,379],{},[124,2022,2023,2031,2039,2049],{},[111,2024,2025,2027,2029],{},[129,2026,386],{},[129,2028,389],{},[129,2030,392],{},[111,2032,2033,2035,2037],{},[129,2034,397],{},[129,2036,392],{},[129,2038,402],{},[111,2040,2041,2043,2046],{},[129,2042,407],{},[129,2044,2045],{},"与下同",[129,2047,2048],{},"与上同",[111,2050,2051,2053,2055],{},[129,2052,417],{},[129,2054,420],{},[129,2056,423],{},[20,2058,2059,2061],{},[24,2060,240],{},"：日活 \u003C 100 用云版省心；> 500 或数据敏感场景自托管 ROI 更好。",[16,2063,1124],{"id":1123},[285,2065,2067],{"className":287,"code":2066,"language":289,"meta":290,"style":290},"# 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",[257,2068,2069,2074,2082,2088,2096,2106,2112,2118,2124,2129,2135,2140,2146,2151],{"__ignoreMap":290},[294,2070,2071],{"class":296,"line":297},[294,2072,2073],{"class":351},"# 1. 自托管（社区版）\n",[294,2075,2076,2078,2080],{"class":296,"line":311},[294,2077,301],{"class":300},[294,2079,305],{"class":304},[294,2081,308],{"class":304},[294,2083,2084,2086],{"class":296,"line":321},[294,2085,315],{"class":314},[294,2087,318],{"class":304},[294,2089,2090,2092,2094],{"class":296,"line":333},[294,2091,324],{"class":300},[294,2093,327],{"class":304},[294,2095,330],{"class":304},[294,2097,2098,2100,2102,2104],{"class":296,"line":348},[294,2099,336],{"class":300},[294,2101,339],{"class":304},[294,2103,342],{"class":304},[294,2105,345],{"class":314},[294,2107,2109],{"class":296,"line":2108},6,[294,2110,2111],{"emptyLinePlaceholder":753},"\n",[294,2113,2115],{"class":296,"line":2114},7,[294,2116,2117],{"class":351},"# 2. 浏览器打开 http:\u002F\u002Flocalhost\n",[294,2119,2121],{"class":296,"line":2120},8,[294,2122,2123],{"class":351},"#    首次会让你创建 admin 账号\n",[294,2125,2127],{"class":296,"line":2126},9,[294,2128,2111],{"emptyLinePlaceholder":753},[294,2130,2132],{"class":296,"line":2131},10,[294,2133,2134],{"class":351},"# 3. 进入\"设置 → 模型供应商\"，配置 OpenAI \u002F 国产模型 API key\n",[294,2136,2138],{"class":296,"line":2137},11,[294,2139,2111],{"emptyLinePlaceholder":753},[294,2141,2143],{"class":296,"line":2142},12,[294,2144,2145],{"class":351},"# 4. 在主界面\"创建空白应用\"，选 Chatflow 或 Workflow\n",[294,2147,2148],{"class":296,"line":8},[294,2149,2150],{"class":351},"# 5. 拖入\"开始 → LLM → 结束\"节点试一下基础 prompt\n",[294,2152,2154],{"class":296,"line":2153},14,[294,2155,2156],{"class":351},"# 6. 满意了点右上\"发布\"，自动生成 API endpoint\n",[16,2158,2159],{"id":2159},"国内使用注意事项",[1126,2161,2162,2168,2174,2180],{},[576,2163,2164,2167],{},[24,2165,2166],{},"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",[576,2169,2170,2173],{},[24,2171,2172],{},"自托管 + 国产模型"," = 完全国内闭环，是 Dify 在国内最大优势",[576,2175,2176,2179],{},[24,2177,2178],{},"Docker 镜像拉取","：国内可能慢，建议配 Docker registry 镜像（阿里云 \u002F 网易）",[576,2181,2182,2185,2186,2188],{},[24,2183,2184],{},"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 ",[40,2187,43],{"href":42}," 迁到 Dify",[16,2190,1167],{"id":1167},[105,2192,2193,2213],{},[108,2194,2195],{},[111,2196,2197,2199,2201,2205,2209],{},[114,2198,1004],{},[114,2200,763],{},[114,2202,2203],{},[40,2204,43],{"href":42},[114,2206,2207],{},[40,2208,102],{"href":101},[114,2210,2211],{},[40,2212,651],{"href":650},[124,2214,2215,2228,2240,2253,2265,2278,2292,2305],{},[111,2216,2217,2219,2221,2223,2225],{},[129,2218,765],{},[129,2220,1202],{},[129,2222,1034],{},[129,2224,1202],{},[129,2226,2227],{},"✅（fair-code）",[111,2229,2230,2232,2234,2236,2238],{},[129,2231,766],{},[129,2233,1202],{},[129,2235,1034],{},[129,2237,1202],{},[129,2239,1202],{},[111,2241,2242,2244,2246,2249,2251],{},[129,2243,1225],{},[129,2245,1250],{},[129,2247,2248],{},"★★☆☆☆ 最简单",[129,2250,1250],{},[129,2252,1244],{},[111,2254,2255,2257,2259,2261,2263],{},[129,2256,1241],{},[129,2258,1247],{},[129,2260,1244],{},[129,2262,1250],{},[129,2264,1247],{},[111,2266,2267,2269,2271,2273,2275],{},[129,2268,1257],{},[129,2270,1244],{},[129,2272,1250],{},[129,2274,1247],{},[129,2276,2277],{},"★★☆☆☆",[111,2279,2280,2283,2285,2287,2290],{},[129,2281,2282],{},"模型生态",[129,2284,1247],{},[129,2286,1244],{},[129,2288,2289],{},"★★★☆☆（OneAPI 中转）",[129,2291,1244],{},[111,2293,2294,2297,2299,2301,2303],{},[129,2295,2296],{},"中文场景",[129,2298,1244],{},[129,2300,1247],{},[129,2302,1244],{},[129,2304,1250],{},[111,2306,2307,2310,2312,2315,2317],{},[129,2308,2309],{},"字节生态绑定",[129,2311,1034],{},[129,2313,2314],{},"✅（飞书\u002F抖音深度集成）",[129,2316,1034],{},[129,2318,1034],{},[20,2320,2321,2323,2324,1318,2327,2330],{},[24,2322,1311],{},"（基于 ",[40,2325,1317],{"href":1315,"rel":2326},[233],[40,2328,1322],{"href":815,"rel":2329},[233]," 两份选型指南综合）：",[573,2332,2333,2339,2346,2352,2359],{},[576,2334,2335,2338],{},[24,2336,2337],{},"数据必须不出内网 + 工作流复杂"," → Dify",[576,2340,2341,1336,2344],{},[24,2342,2343],{},"个人 \u002F 小团队 \u002F 快速原型 + 字节生态",[40,2345,43],{"href":42},[576,2347,2348,1336,2350],{},[24,2349,244],{},[40,2351,102],{"href":101},[576,2353,2354,1336,2357],{},[24,2355,2356],{},"重点是连接外部 SaaS（Slack \u002F Notion \u002F 数据库）",[40,2358,651],{"href":650},[576,2360,2361,1336,2364],{},[24,2362,2363],{},"要画图式表达 LangChain pipeline",[40,2365,1484],{"href":1483},[16,2367,1355],{"id":1355},[573,2369,2370,2376,2388,2399,2410,2416,2422,2428],{},[576,2371,2372,2375],{},[24,2373,2374],{},"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。",[576,2377,2378,65,2383,441,2385,2387],{},[24,2379,2380,2382],{},[257,2381,259],{}," 文件改完忘 restart",[257,2384,440],{},[257,2386,444],{},"——后者不重新加载 env。",[576,2389,2390,65,2393,2398],{},[24,2391,2392],{},"大版本升级会破坏数据库 schema",[40,2394,2397],{"href":2395,"rel":2396},"https:\u002F\u002Fdocs.dify.ai\u002Fzh-hans",[233],"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。",[576,2400,2401,2404,2405,260,2407,2409],{},[24,2402,2403],{},"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 ",[257,2406,259],{},[257,2408,263],{}," 并重启容器。",[576,2411,2412,2415],{},[24,2413,2414],{},"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。",[576,2417,2418,2421],{},[24,2419,2420],{},"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。",[576,2423,2424,2427],{},[24,2425,2426],{},"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。",[576,2429,2430,2433],{},[24,2431,2432],{},"国内 Docker 拉取镜像慢","：先配国内 registry，否则首次 pull 可能要 30+ 分钟。",[16,2435,1432],{"id":1431},[20,2437,1435],{},[573,2439,2440,2442,2445,2447,2450,2453],{},[576,2441,581],{},[576,2443,2444],{},"需要私有化部署（金融 \u002F 医疗 \u002F 政府）",[576,2446,593],{},[576,2448,2449],{},"同时需要 RAG + Agent + Workflow 三件套",[576,2451,2452],{},"想用国产模型 + 国际模型混合编排",[576,2454,2455],{},"已经接受 Docker + 一定运维投入",[20,2457,1455],{},[573,2459,2460,2466,2472,2475,2481],{},[576,2461,2462,2463,2465],{},"纯个人玩家做对话机器人（",[40,2464,43],{"href":42}," 更快）",[576,2467,2468,2469,2471],{},"只想做企业知识库 QA（",[40,2470,102],{"href":101}," RAG 更专）",[576,2473,2474],{},"团队完全没运维能力（云版还行，自托管会踩坑）",[576,2476,2477,2478,2480],{},"需要深度对接字节飞书 \u002F 抖音（",[40,2479,43],{"href":42}," 原生）",[576,2482,2483,2484,2486],{},"工作流核心是连接 100+ SaaS（",[40,2485,651],{"href":650}," 节点更全）",[16,2488,693],{"id":693},[573,2490,2491,2501,2513,2528],{},[576,2492,1492,2493,1413,2495,1413,2497,1413,2499],{},[40,2494,43],{"href":42},[40,2496,102],{"href":101},[40,2498,651],{"href":650},[40,2500,1484],{"href":1483},[576,2502,2503,2504,1413,2506,1413,2508,1413,2511],{},"概念基础：",[40,2505,764],{"href":1508},[40,2507,1511],{"href":726},[40,2509,2510],{"href":730},"MCP",[40,2512,1515],{"href":1514},[576,2514,2515,2516,1413,2520,1413,2524,1413,2526],{},"模型选型：",[40,2517,2519],{"href":2518},"\u002Fmodels\u002Fgpt-5.html","GPT-5",[40,2521,2523],{"href":2522},"\u002Fmodels\u002Fclaude-sonnet-4.html","Claude Sonnet 4",[40,2525,1529],{"href":1528},[40,2527,1537],{"href":1536},[576,2529,1540,2530,1413,2534],{},[40,2531,2533],{"href":2532},"\u002Fwiki\u002Ffine-tuning-vs-rag.html","Fine-tuning vs RAG",[40,2535,1548],{"href":1547},[16,2537,1551],{"id":1551},[573,2539,2540,2547,2553,2559,2566],{},[576,2541,2542,2543],{},"官网：",[40,2544,2545],{"href":2545,"rel":2546},"https:\u002F\u002Fdify.ai",[233],[576,2548,2549,2550],{},"中文文档：",[40,2551,2395],{"href":2395,"rel":2552},[233],[576,2554,2555,2556],{},"GitHub：",[40,2557,280],{"href":280,"rel":2558},[233],[576,2560,2561,2562],{},"官方定价：",[40,2563,2564],{"href":2564,"rel":2565},"https:\u002F\u002Fdify.ai\u002Fpricing",[233],[576,2567,2568],{},"第三方评测：tooljunction.io \u002F chatforest.com \u002F besthub.dev \u002F joshuaopolko.com \u002F 知乎 LLM 实战笔记",[20,2570,2571,2572,1585],{},"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 ",[40,2573,1584],{"href":1584},[733,2575,735],{},{"title":290,"searchDepth":321,"depth":321,"links":2577},[2578,2579,2586,2591,2592,2593,2594,2595,2596,2597],{"id":780,"depth":311,"text":781},{"id":827,"depth":311,"text":827,"children":2580},[2581,2582,2583,2584,2585],{"id":1715,"depth":321,"text":1716},{"id":1779,"depth":321,"text":1780},{"id":1815,"depth":321,"text":1816},{"id":1830,"depth":321,"text":1831},{"id":1852,"depth":321,"text":1853},{"id":475,"depth":311,"text":475,"children":2587},[2588,2589,2590],{"id":1861,"depth":321,"text":1862},{"id":1928,"depth":321,"text":1929},{"id":2004,"depth":321,"text":2005},{"id":1123,"depth":311,"text":1124},{"id":2159,"depth":311,"text":2159},{"id":1167,"depth":311,"text":1167},{"id":1355,"depth":311,"text":1355},{"id":1431,"depth":311,"text":1432},{"id":693,"depth":311,"text":693},{"id":1551,"depth":311,"text":1551},"\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。",[1608,1609,2601],"ja",{},"\u002Ftools\u002Fagent\u002Fplatform\u002Fdify",[2605,2606,2607,336],"windows","macos","linux",[2609,2612,2616,2619],{"plan":500,"price":392,"features":2610,"notes":2611},"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":2613,"price":392,"features":2614,"notes":2615},"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":521,"price":524,"features":2617,"notes":2618},"更高调用额度 + 团队协作 + 商用支持","商用云首选",{"plan":2620,"price":2621,"features":2622,"notes":546},"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":348,"ux":333,"price":348,"cn_support":333,"stability":333},{"title":763,"description":2599},"Dify 评测 2026：开源 LLMOps 与 AI Agent 平台，自托管指南",[2628,2630,2632,2634,2636],{"title":2629,"url":2395},"Dify 官方文档（中文）",{"title":2631,"url":280},"Dify GitHub",{"title":2633,"url":2564},"Dify 官方定价",{"title":2635,"url":1315},"Coze vs Dify vs FastGPT 选型",{"title":2637,"url":2638},"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",[1671,2642,2643,2644,1673,2645,2646],"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":2650,"title":102,"alternatives":2651,"api_compatible":2654,"body":2656,"category":1602,"chinese_friendly":348,"cover":3659,"description":3660,"domestic":1605,"extension":751,"faq":1606,"free":1605,"github":3622,"languages":3661,"meta":3662,"models":3663,"navigation":753,"notSuitable":3667,"opensource":753,"path":3671,"pillar":1622,"platforms":3672,"priceTable":3673,"pricing":3695,"published":1647,"relatedPlaybooks":3696,"relatedReviews":3698,"score":3700,"self_host":753,"seo":3701,"seoTitle":1606,"slug":758,"sources":3702,"stem":3713,"suitable":3714,"tagline":3720,"tags":3721,"updated":1677,"verdict":3725,"website":3616,"__hash__":3726},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Ffastgpt.md",[757,759,2652,2653],"agent\u002Fplatform\u002Fragflow","agent\u002Fplatform\u002Fanythingllm",[2655],"openai",{"type":13,"value":2657,"toc":3641},[2658,2660,2689,2700,2702,2706,2709,2723,2726,2752,2756,2763,2795,2802,2806,2859,2866,2869,2872,2875,2882,2928,2934,2938,2967,2971,3077,3080,3083,3087,3095,3217,3232,3234,3391,3399,3433,3439,3441,3515,3517,3519,3539,3541,3559,3561,3608,3610,3633,3638],[16,2659,781],{"id":780},[46,2661,2663,2678],{"className":2662},[49,50,51],[20,2664,2665,2667,2668,2673,2674,2677],{},[24,2666,789],{}," labring 团队开源的 LLM 知识库 RAG 平台，27k+ GitHub star（截至 2026-03 数据，",[40,2669,2672],{"href":2670,"rel":2671},"https:\u002F\u002Fcloud.tencent.com\u002Fdeveloper\u002Farticle\u002F2632669",[233],"腾讯云 2026-03 教程"," 引用），Apache 2.0 许可证可商用。",[24,2675,2676],{},"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。",[20,2679,2680,2681,2684,2685,2688],{},"最大价值在 ",[24,2682,2683],{},"国内企业知识库 + 私有部署"," 场景。代价是 ",[24,2686,2687],{},"配置门槛","：docker 基础 + 网络知识 + 一定运维能力。",[820,2690,2691],{},[20,2692,2693,2694,2699],{},"来源说明：本文基于 fastgpt.io 官方页面、github.com\u002Flabring\u002FFastGPT 仓库、",[40,2695,2698],{"href":2696,"rel":2697},"https:\u002F\u002Fwww.nanhuantech.com\u002Fzh\u002Fai-reviews\u002Ffastgpt-2025-review",[233],"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。",[16,2701,827],{"id":827},[829,2703,2705],{"id":2704},"知识库管理核心能力","知识库管理（核心能力）",[20,2707,2708],{},"支持文件类型：",[573,2710,2711,2714,2717,2720],{},[576,2712,2713],{},"文档：PDF \u002F Word \u002F Markdown \u002F TXT \u002F HTML",[576,2715,2716],{},"表格：Excel \u002F CSV",[576,2718,2719],{},"网页：URL 抓取 + 定时同步",[576,2721,2722],{},"API：通过接口推送内容",[20,2724,2725],{},"处理流程：上传 → 文本切分 → 向量化 → 存储 → 可用于问答。支持：",[573,2727,2728,2734,2740,2746],{},[576,2729,2730,2733],{},[24,2731,2732],{},"文件夹分组","：不同主题 \u002F 部门分类",[576,2735,2736,2739],{},[24,2737,2738],{},"多种分块策略","：默认按段落 \u002F 按 token 数 \u002F 自定义",[576,2741,2742,2745],{},[24,2743,2744],{},"批量导入","：脚本化大批量同步",[576,2747,2748,2751],{},[24,2749,2750],{},"定时同步","：网页源自动更新",[829,2753,2755],{"id":2754},"rag-流程编排最强卖点","RAG 流程编排（最强卖点）",[20,2757,2758,2762],{},[40,2759,2761],{"href":2696,"rel":2760},[233],"南环 AI 2026 评测"," 总结的 FastGPT RAG 链路：",[1126,2764,2765,2771,2777,2783,2789],{},[576,2766,2767,2770],{},[24,2768,2769],{},"问题预处理","：改写 \u002F 扩展 \u002F 错词纠正（提升召回率）",[576,2772,2773,2776],{},[24,2774,2775],{},"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",[576,2778,2779,2782],{},[24,2780,2781],{},"重排序（Rerank）","：对初步检索结果二次排序，提升相关性",[576,2784,2785,2788],{},[24,2786,2787],{},"上下文组装","：最优 chunk + 问题 → prompt",[576,2790,2791,2794],{},[24,2792,2793],{},"答案生成","：调大模型基于检索结果回答 + 引用标注",[20,2796,2797,2798,2801],{},"每一步都可视化调参，这是 FastGPT 比 Coze \u002F Dify 在 ",[24,2799,2800],{},"纯知识库 QA 精度","上更高的原因。",[829,2803,2805],{"id":2804},"多模型支持不绑定厂商","多模型支持（不绑定厂商）",[105,2807,2808,2818],{},[108,2809,2810],{},[111,2811,2812,2815],{},[114,2813,2814],{},"模型类别",[114,2816,2817],{},"支持",[124,2819,2820,2828,2835,2843,2851],{},[111,2821,2822,2825],{},[129,2823,2824],{},"国产闭源",[129,2826,2827],{},"豆包 \u002F 通义千问 \u002F 文心一言 \u002F 智谱 GLM \u002F Moonshot Kimi \u002F MiniMax",[111,2829,2830,2832],{},[129,2831,765],{},[129,2833,2834],{},"LLaMA \u002F Qwen \u002F ChatGLM \u002F DeepSeek 等可自部署",[111,2836,2837,2840],{},[129,2838,2839],{},"OpenAI 系",[129,2841,2842],{},"GPT-5 \u002F GPT-5 mini \u002F o3",[111,2844,2845,2848],{},[129,2846,2847],{},"Claude 系",[129,2849,2850],{},"Sonnet 4 \u002F Opus 4 \u002F Haiku",[111,2852,2853,2856],{},[129,2854,2855],{},"嵌入 \u002F 重排",[129,2857,2858],{},"BGE \u002F m3e \u002F OpenAI text-embedding-3",[20,2860,2861,2862,2865],{},"可以在 ",[24,2863,2864],{},"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。",[829,2867,2868],{"id":2868},"工作流与高级编排",[20,2870,2871],{},"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。",[829,2873,2874],{"id":2874},"多向量库选择",[20,2876,2877,2881],{},[40,2878,2880],{"href":2670,"rel":2879},[233],"腾讯云教程"," 公开的 4 种向量后端：",[105,2883,2884,2894],{},[108,2885,2886],{},[111,2887,2888,2891],{},[114,2889,2890],{},"后端",[114,2892,2893],{},"适用",[124,2895,2896,2904,2912,2920],{},[111,2897,2898,2901],{},[129,2899,2900],{},"PgVector",[129,2902,2903],{},"5000 万索引以下，新手 \u002F 小规模",[111,2905,2906,2909],{},[129,2907,2908],{},"Milvus",[129,2910,2911],{},"亿级以上，高性能",[111,2913,2914,2917],{},[129,2915,2916],{},"Zilliz Cloud",[129,2918,2919],{},"Milvus 全托管 SaaS",[111,2921,2922,2925],{},[129,2923,2924],{},"SeekDB \u002F OceanBase",[129,2926,2927],{},"企业级国产化",[20,2929,2930,2931,191],{},"部署时选对应 ",[257,2932,2933],{},"docker-compose.{pgvector|milvus|...}.yml",[829,2935,2937],{"id":2936},"api-与-mcp","API 与 MCP",[573,2939,2940,2946,2952,2961],{},[576,2941,2942,2945],{},[24,2943,2944],{},"对话 API","：流式 \u002F 非流式 HTTP，OpenAI 兼容",[576,2947,2948,2951],{},[24,2949,2950],{},"知识库检索 API","：单独调检索（不走生成）做 hybrid pipeline",[576,2953,2954,2957,2958,2960],{},[24,2955,2956],{},"MCP Server","：3005 端口暴露 MCP SSE 服务，可被 ",[40,2959,201],{"href":200}," 等客户端直接接入",[576,2962,2963,2966],{},[24,2964,2965],{},"Webhook","：回调通知",[16,2968,2970],{"id":2969},"部署-10-分钟docker","部署 10 分钟（Docker）",[285,2972,2974],{"className":287,"code":2973,"language":289,"meta":290,"style":290},"# 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",[257,2975,2976,2981,2990,2997,3001,3006,3019,3023,3028,3035,3043,3047,3052,3067,3071],{"__ignoreMap":290},[294,2977,2978],{"class":296,"line":297},[294,2979,2980],{"class":351},"# 克隆代码\n",[294,2982,2983,2985,2987],{"class":296,"line":311},[294,2984,301],{"class":300},[294,2986,305],{"class":304},[294,2988,2989],{"class":304}," https:\u002F\u002Fgithub.com\u002Flabring\u002FFastGPT.git\n",[294,2991,2992,2994],{"class":296,"line":321},[294,2993,315],{"class":314},[294,2995,2996],{"class":304}," FastGPT\n",[294,2998,2999],{"class":296,"line":333},[294,3000,2111],{"emptyLinePlaceholder":753},[294,3002,3003],{"class":296,"line":348},[294,3004,3005],{"class":351},"# 切到最新稳定版（参考 GitHub releases）\n",[294,3007,3008,3010,3013,3016],{"class":296,"line":2108},[294,3009,301],{"class":300},[294,3011,3012],{"class":304}," switch",[294,3014,3015],{"class":314}," -c",[294,3017,3018],{"class":314}," 4.14.7.2\n",[294,3020,3021],{"class":296,"line":2114},[294,3022,2111],{"emptyLinePlaceholder":753},[294,3024,3025],{"class":296,"line":2120},[294,3026,3027],{"class":351},"# 选向量库版本（个人 \u002F 小规模选 pg）\n",[294,3029,3030,3032],{"class":296,"line":2126},[294,3031,315],{"class":314},[294,3033,3034],{"class":304}," deploy\u002Fdocker\u002Fcn\n",[294,3036,3037,3040],{"class":296,"line":2131},[294,3038,3039],{"class":300},"wget",[294,3041,3042],{"class":304}," https:\u002F\u002Fdoc.fastgpt.cn\u002Fdeploy\u002Fconfig\u002Fconfig.json\n",[294,3044,3045],{"class":296,"line":2137},[294,3046,2111],{"emptyLinePlaceholder":753},[294,3048,3049],{"class":296,"line":2142},[294,3050,3051],{"class":351},"# 启动\n",[294,3053,3054,3057,3060,3063,3065],{"class":296,"line":8},[294,3055,3056],{"class":300},"docker-compose",[294,3058,3059],{"class":314}," -f",[294,3061,3062],{"class":304}," docker-compose.pg.yml",[294,3064,342],{"class":304},[294,3066,345],{"class":314},[294,3068,3069],{"class":296,"line":2153},[294,3070,2111],{"emptyLinePlaceholder":753},[294,3072,3074],{"class":296,"line":3073},15,[294,3075,3076],{"class":351},"# 访问 http:\u002F\u002F\u003Cip>:3000，默认账号 root \u002F 1234\n",[20,3078,3079],{},"最低配置：2C4G + 20GB 硬盘 + Docker 28+ + Docker Compose 2.20+。",[20,3081,3082],{},"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。",[16,3084,3086],{"id":3085},"云版-vs-自托管对比","云版 vs 自托管对比",[20,3088,3089,3094],{},[40,3090,3093],{"href":3091,"rel":3092},"https:\u002F\u002Ffastgpt.io\u002Fzh\u002Fprice",[233],"fastgpt.io 官方定价"," 公开数据：",[105,3096,3097,3122],{},[108,3098,3099],{},[111,3100,3101,3103,3105,3108,3111,3114,3116,3119],{},[114,3102,487],{},[114,3104,490],{},[114,3106,3107],{},"AI 积分",[114,3109,3110],{},"知识库索引",[114,3112,3113],{},"团队",[114,3115,146],{},[114,3117,3118],{},"知识库",[114,3120,3121],{},"QPM",[124,3123,3124,3148,3173,3197],{},[111,3125,3126,3128,3130,3133,3136,3139,3142,3145],{},[129,3127,513],{},[129,3129,1628],{},[129,3131,3132],{},"100",[129,3134,3135],{},"600",[129,3137,3138],{},"1",[129,3140,3141],{},"10",[129,3143,3144],{},"3",[129,3146,3147],{},"30",[111,3149,3150,3153,3156,3159,3162,3165,3168,3170],{},[129,3151,3152],{},"基础",[129,3154,3155],{},"¥99\u002F月",[129,3157,3158],{},"4000",[129,3160,3161],{},"6000",[129,3163,3164],{},"5",[129,3166,3167],{},"50",[129,3169,3147],{},[129,3171,3172],{},"300",[111,3174,3175,3178,3181,3184,3187,3189,3192,3194],{},[129,3176,3177],{},"高级",[129,3179,3180],{},"¥599\u002F月",[129,3182,3183],{},"25000",[129,3185,3186],{},"36000",[129,3188,3167],{},[129,3190,3191],{},"200",[129,3193,3132],{},[129,3195,3196],{},"1500",[111,3198,3199,3202,3204,3207,3209,3211,3213,3215],{},[129,3200,3201],{},"定制",[129,3203,1638],{},[129,3205,3206],{},"弹性",[129,3208,3206],{},[129,3210,3206],{},[129,3212,3206],{},[129,3214,3206],{},[129,3216,3206],{},[20,3218,3219,3222,3223,3226,3227,3231],{},[24,3220,3221],{},"云版适合","：不想运维、量小、要快速上线\n",[24,3224,3225],{},"自托管适合","：量大（10 万+ 日问答）、数据敏感、要深度定制——按 ",[40,3228,3230],{"href":2696,"rel":3229},[233],"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\"",[16,3233,1167],{"id":1167},[105,3235,3236,3258],{},[108,3237,3238],{},[111,3239,3240,3242,3244,3248,3252,3255],{},[114,3241,1004],{},[114,3243,102],{},[114,3245,3246],{},[40,3247,763],{"href":700},[114,3249,3250],{},[40,3251,43],{"href":42},[114,3253,3254],{},"RAGFlow",[114,3256,3257],{},"AnythingLLM",[124,3259,3260,3280,3296,3311,3328,3343,3360,3375],{},[111,3261,3262,3265,3268,3271,3274,3277],{},[129,3263,3264],{},"核心定位",[129,3266,3267],{},"知识库 QA",[129,3269,3270],{},"综合 LLMOps",[129,3272,3273],{},"Bot + 工作流",[129,3275,3276],{},"文档解析+RAG",[129,3278,3279],{},"桌面级 KB",[111,3281,3282,3284,3287,3289,3291,3293],{},[129,3283,765],{},[129,3285,3286],{},"✅ Apache 2.0",[129,3288,3286],{},[129,3290,1034],{},[129,3292,3286],{},[129,3294,3295],{},"✅ MIT",[111,3297,3298,3300,3303,3305,3307,3309],{},[129,3299,766],{},[129,3301,3302],{},"★★★★★ docker",[129,3304,1247],{},[129,3306,1213],{},[129,3308,1244],{},[129,3310,1247],{},[111,3312,3313,3316,3319,3321,3323,3326],{},[129,3314,3315],{},"RAG 深度",[129,3317,3318],{},"★★★★★ 最细",[129,3320,1244],{},[129,3322,1250],{},[129,3324,3325],{},"★★★★★ 文档解析最强",[129,3327,1250],{},[111,3329,3330,3333,3335,3337,3339,3341],{},[129,3331,3332],{},"工作流",[129,3334,1244],{},[129,3336,1247],{},[129,3338,1244],{},[129,3340,1250],{},[129,3342,2277],{},[111,3344,3345,3348,3351,3353,3356,3358],{},[129,3346,3347],{},"上手",[129,3349,3350],{},"★★★☆☆ 需 docker",[129,3352,1244],{},[129,3354,3355],{},"★★★★★ 最简单",[129,3357,1250],{},[129,3359,1244],{},[111,3361,3362,3365,3367,3369,3371,3373],{},[129,3363,3364],{},"中文优化",[129,3366,1247],{},[129,3368,1244],{},[129,3370,1247],{},[129,3372,1244],{},[129,3374,1250],{},[111,3376,3377,3380,3383,3385,3387,3389],{},[129,3378,3379],{},"多平台发布",[129,3381,3382],{},"⚠️ API 为主",[129,3384,1244],{},[129,3386,1247],{},[129,3388,1220],{},[129,3390,1220],{},[20,3392,3393,1312,3395,858],{},[24,3394,1311],{},[40,3396,3398],{"href":2696,"rel":3397},[233],"南环 AI 评测",[573,3400,3401,3407,3414,3421,3427],{},[576,3402,3403,3406],{},[24,3404,3405],{},"核心需求是 RAG 精度"," → FastGPT",[576,3408,3409,1336,3412],{},[24,3410,3411],{},"需要丰富插件 + 复杂工作流 + 多平台发布",[40,3413,763],{"href":700},[576,3415,3416,1336,3419],{},[24,3417,3418],{},"零代码、快速发布到飞书 \u002F 微信",[40,3420,43],{"href":42},[576,3422,3423,3426],{},[24,3424,3425],{},"文档解析（含 OCR \u002F 表格 \u002F 公式）是瓶颈"," → RAGFlow",[576,3428,3429,3432],{},[24,3430,3431],{},"桌面 \u002F 单机使用"," → AnythingLLM",[20,3434,3435,3438],{},[24,3436,3437],{},"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。",[16,3440,1355],{"id":1355},[573,3442,3443,3456,3465,3471,3481,3493,3499,3505],{},[576,3444,3445,65,3448,3451,3452,3455],{},[24,3446,3447],{},"docker-compose 镜像 tag 不一致",[40,3449,2880],{"href":2670,"rel":3450},[233]," 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 ",[257,3453,3454],{},"image:"," 行为正确版本即可",[576,3457,3458,3461,3462,3464],{},[24,3459,3460],{},"3000 端口冲突","：默认占用 3000（主服务）\u002F 9000（S3 \u002F MinIO）\u002F 3005（MCP）；改 ",[257,3463,3056],{}," 的 ports 映射端口",[576,3466,3467,3470],{},[24,3468,3469],{},"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",[576,3472,3473,3476,3477,3480],{},[24,3474,3475],{},"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 ",[40,3478,3230],{"href":2696,"rel":3479},[233],"：\"新手 \u002F 小规模 PgVector，中大规模 Milvus，企业 \u002F 国产 OceanBase\"",[576,3482,3483,65,3486,3489,3490],{},[24,3484,3485],{},"MinIO 默认密码",[257,3487,3488],{},"minioadmin\u002Fminioadmin","，",[24,3491,3492],{},"部署到公网前必须改",[576,3494,3495,3498],{},[24,3496,3497],{},"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\"",[576,3500,3501,3504],{},[24,3502,3503],{},"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",[576,3506,3507,3510,3511,1388],{},[24,3508,3509],{},"云版 AI 积分会过期","：未用完不能跨月累积（按 ",[40,3512,3514],{"href":3091,"rel":3513},[233],"fastgpt.io 定价 FAQ",[16,3516,1432],{"id":1431},[20,3518,1435],{},[573,3520,3521,3524,3527,3530,3533,3536],{},[576,3522,3523],{},"企业内部知识库（员工手册 \u002F 制度 \u002F 流程）",[576,3525,3526],{},"产品 FAQ \u002F 用户手册问答",[576,3528,3529],{},"医疗 \u002F 法律 \u002F 金融垂直领域知识系统",[576,3531,3532],{},"数据严格不出网 + Apache 2.0 商用",[576,3534,3535],{},"有 docker 运维基础的技术团队",[576,3537,3538],{},"需要把 RAG 当后端服务的开发者（API 接入业务系统）",[20,3540,1455],{},[573,3542,3543,3548,3553,3556],{},[576,3544,3545,3546,1388],{},"完全非技术用户（去 ",[40,3547,43],{"href":42},[576,3549,3550,3551,1388],{},"主要需求是工作流 + 插件集成（去 ",[40,3552,763],{"href":700},[576,3554,3555],{},"文档解析 \u002F OCR 是首要痛点（RAGFlow）",[576,3557,3558],{},"不想自己运维 + 量很小（FastGPT 云免费版起步即可）",[16,3560,693],{"id":693},[573,3562,3563,3572,3588,3602],{},[576,3564,1492,3565,1413,3567,3569,3570],{},[40,3566,763],{"href":700},[40,3568,43],{"href":42}," \u002F RAGFlow \u002F AnythingLLM \u002F ",[40,3571,651],{"href":650},[576,3573,1505,3574,1413,3576,1413,3580,1413,3583,1413,3586],{},[40,3575,1511],{"href":726},[40,3577,3579],{"href":3578},"\u002Fwiki\u002Fembedding.html","Embedding",[40,3581,3582],{"href":3578},"Vector Database",[40,3584,3585],{"href":726},"Reranker",[40,3587,764],{"href":1508},[576,3589,1521,3590,1413,3592,1413,3594,1413,3596,1413,3600],{},[40,3591,1529],{"href":1528},[40,3593,1533],{"href":1532},[40,3595,1537],{"href":1536},[40,3597,3599],{"href":3598},"\u002Fmodels\u002Fkimi-k2.html","Kimi K2",[40,3601,1525],{"href":1524},[576,3603,1540,3604,1413,3606],{},[40,3605,1548],{"href":1547},[40,3607,1544],{"href":1543},[16,3609,1551],{"id":1551},[573,3611,3612,3618,3624,3630],{},[576,3613,2542,3614],{},[40,3615,3616],{"href":3616,"rel":3617},"https:\u002F\u002Ffastgpt.io",[233],[576,3619,2555,3620],{},[40,3621,3622],{"href":3622,"rel":3623},"https:\u002F\u002Fgithub.com\u002Flabring\u002FFastGPT",[233],[576,3625,3626,3627],{},"定价：",[40,3628,3091],{"href":3091,"rel":3629},[233],[576,3631,3632],{},"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",[20,3634,3635,3636,1585],{},"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 ",[40,3637,1584],{"href":1584},[733,3639,3640],{},"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":290,"searchDepth":321,"depth":321,"links":3642},[3643,3644,3652,3653,3654,3655,3656,3657,3658],{"id":780,"depth":311,"text":781},{"id":827,"depth":311,"text":827,"children":3645},[3646,3647,3648,3649,3650,3651],{"id":2704,"depth":321,"text":2705},{"id":2754,"depth":321,"text":2755},{"id":2804,"depth":321,"text":2805},{"id":2868,"depth":321,"text":2868},{"id":2874,"depth":321,"text":2874},{"id":2936,"depth":321,"text":2937},{"id":2969,"depth":311,"text":2970},{"id":3085,"depth":311,"text":3086},{"id":1167,"depth":311,"text":1167},{"id":1355,"depth":311,"text":1355},{"id":1431,"depth":311,"text":1432},{"id":693,"depth":311,"text":693},{"id":1551,"depth":311,"text":1551},"\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。",[1608,1609],{},[1613,1614,1612,3664,3665,3666],"gpt-4o","claude-sonnet-4","kimi",[3668,3669,3670],"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",[2605,2606,2607],[3674,3678,3683,3687,3691],{"plan":3675,"price":513,"limit":3676,"cn_pay":1630,"note":3677},"Self-host 开源","全功能 + 全数据本地","Apache 2.0 可商用",{"plan":3679,"price":3680,"limit":3681,"cn_pay":1630,"note":3682},"云免费版","¥0\u002F月","100 AI 积分 + 600 索引 + 3 知识库","试水",{"plan":3684,"price":3155,"limit":3685,"cn_pay":1635,"note":3686},"云基础版","4000 积分 + 6000 索引 + 50 Agent","中小团队 SaaS",{"plan":3688,"price":3180,"limit":3689,"cn_pay":1202,"note":3690},"云高级版","25000 积分 + 36000 索引 + 50 成员 + 200 Agent + 1500 QPM","企业级生产",{"plan":3692,"price":1638,"limit":3693,"cn_pay":1202,"note":3694},"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",[3697],"onboarding\u002Ffastgpt-getting-started",[3699],"fastgpt-deep-review",{"power":333,"ux":333,"price":348,"cn_support":348,"stability":333},{"title":102,"description":3660},[3703,3705,3707,3709,3711],{"title":3704,"url":3616},"FastGPT 官网",{"title":3706,"url":3622},"FastGPT GitHub",{"title":3708,"url":3091},"FastGPT 定价页",{"title":3710,"url":2696},"FastGPT 2025 测评（南环 AI）",{"title":3712,"url":2670},"FastGPT 部署教程（腾讯云）","tools\u002Fagent\u002Fplatform\u002Ffastgpt",[3715,3716,3717,3718,3719],"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",[1671,2642,2643,2644,3722,3723,3724],"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",1783173060765]