[{"data":1,"prerenderedAt":3902},["ShallowReactive",2],{"header-counts":3,"playbook-onboarding\u002Ffastgpt-getting-started":6,"footer-counts":876,"playbook-tools-onboarding\u002Ffastgpt-getting-started":879},{"tools":4,"reviews":5},70,14,{"id":7,"title":8,"body":9,"category":856,"cover":857,"description":858,"extension":859,"meta":860,"navigation":596,"path":861,"published":862,"relatedTools":863,"seo":867,"stem":868,"tags":869,"updated":862,"__hash__":875},"playbook\u002Fplaybook\u002Fonboarding\u002Ffastgpt-getting-started.md","FastGPT 部署与知识库搭建实战：docker 起步、向量库选型、RAG 调参",{"type":10,"value":11,"toc":831},"minimark",[12,16,32,35,38,41,68,75,79,82,145,159,170,173,180,255,261,264,271,294,301,308,311,318,324,338,343,392,399,402,408,411,431,434,438,444,449,452,472,475,479,482,502,505,509,512,526,533,537,543,557,560,564,567,581,585,651,654,657,702,705,725,728,748,751,771,775,781,789,792,795,827],[13,14,15],"h2",{"id":15},"适用人群",[17,18,19,20,24,25,24,28,31],"p",{},"这份指南适合三类用户：",[21,22,23],"strong",{},"准备用 FastGPT 搭企业知识库的开发者","、",[21,26,27],{},"已经装了 FastGPT 但 RAG 精度不满意的团队",[21,29,30],{},"在 FastGPT \u002F Dify \u002F Coze 之间做技术选型的架构师","。",[17,33,34],{},"它不是抄命令行的操作手册，而是把\"部署 → 建库 → 调参 → 上线\"这条链路上真正会踩的坑写清楚，让你少走弯路。",[13,36,37],{"id":37},"前置条件",[17,39,40],{},"在开始之前，确认三件事：",[42,43,44,51,61],"ul",{},[45,46,47,48,31],"li",{},"一台 Linux 服务器（阿里云 \u002F 腾讯云 \u002F 自建都行），最低 ",[21,49,50],{},"2C4G + 20GB 硬盘",[45,52,53,54,57,58,31],{},"已装 ",[21,55,56],{},"Docker 28+"," 和 ",[21,59,60],{},"Docker Compose 2.20+",[45,62,63,64,67],{},"至少一个可用的 ",[21,65,66],{},"LLM API Key","（OpenAI 兼容即可，DeepSeek、豆包、Qwen、Kimi 都行）。",[17,69,70,71,74],{},"如果打算做企业级、日均 10 万次问答规模，起步就上 ",[21,72,73],{},"8C16G + 200GB SSD + Milvus 向量库","，别用 pgvector。",[13,76,78],{"id":77},"部署10-分钟","部署（10 分钟）",[17,80,81],{},"标准命令：",[83,84,89],"pre",{"className":85,"code":86,"language":87,"meta":88,"style":88},"language-bash shiki shiki-themes github-light github-dark","git clone https:\u002F\u002Fgithub.com\u002Flabring\u002FFastGPT.git\ncd FastGPT\u002Fdeploy\u002Fdocker\u002Fcn\nwget https:\u002F\u002Fdoc.fastgpt.cn\u002Fdeploy\u002Fconfig\u002Fconfig.json\ndocker-compose -f docker-compose.pg.yml up -d\n","bash","",[90,91,92,108,118,127],"code",{"__ignoreMap":88},[93,94,97,101,105],"span",{"class":95,"line":96},"line",1,[93,98,100],{"class":99},"sScJk","git",[93,102,104],{"class":103},"sZZnC"," clone",[93,106,107],{"class":103}," https:\u002F\u002Fgithub.com\u002Flabring\u002FFastGPT.git\n",[93,109,111,115],{"class":95,"line":110},2,[93,112,114],{"class":113},"sj4cs","cd",[93,116,117],{"class":103}," FastGPT\u002Fdeploy\u002Fdocker\u002Fcn\n",[93,119,121,124],{"class":95,"line":120},3,[93,122,123],{"class":99},"wget",[93,125,126],{"class":103}," https:\u002F\u002Fdoc.fastgpt.cn\u002Fdeploy\u002Fconfig\u002Fconfig.json\n",[93,128,130,133,136,139,142],{"class":95,"line":129},4,[93,131,132],{"class":99},"docker-compose",[93,134,135],{"class":113}," -f",[93,137,138],{"class":103}," docker-compose.pg.yml",[93,140,141],{"class":103}," up",[93,143,144],{"class":113}," -d\n",[17,146,147,148,151,152,31,155,158],{},"启动后访问 ",[90,149,150],{},"http:\u002F\u002F\u003C服务器IP>:3000","，默认账号 ",[90,153,154],{},"root \u002F 1234",[21,156,157],{},"登录后第一件事","：改 root 密码 + 改 MinIO 默认密码。",[17,160,161,162,165,166,169],{},"如果服务器有其他服务占用 3000\u002F9000\u002F3005 端口，编辑 ",[90,163,164],{},"docker-compose.pg.yml"," 里对应服务的 ",[90,167,168],{},"ports:","，把宿主机端口改成没冲突的。",[13,171,172],{"id":172},"向量库怎么选",[17,174,175,176,179],{},"这是",[21,177,178],{},"最容易选错","的地方。选错了后续要重新 embedding 整个知识库，成本很高。",[181,182,183,199],"table",{},[184,185,186],"thead",{},[187,188,189,193,196],"tr",{},[190,191,192],"th",{},"索引量",[190,194,195],{},"推荐向量库",[190,197,198],{},"说明",[200,201,202,216,229,242],"tbody",{},[187,203,204,208,213],{},[205,206,207],"td",{},"\u003C 500 万",[205,209,210],{},[21,211,212],{},"pgvector",[205,214,215],{},"docker-compose.pg.yml，最简单",[187,217,218,221,226],{},[205,219,220],{},"500 万–5000 万",[205,222,223],{},[21,224,225],{},"pgvector 或 Milvus",[205,227,228],{},"看查询延迟需求",[187,230,231,234,239],{},[205,232,233],{},"> 5000 万",[205,235,236],{},[21,237,238],{},"Milvus",[205,240,241],{},"亿级索引首选",[187,243,244,247,252],{},[205,245,246],{},"企业国产化",[205,248,249],{},[21,250,251],{},"OceanBase \u002F SeekDB",[205,253,254],{},"有信创要求走这条",[17,256,257,260],{},[21,258,259],{},"判断索引量","：文档数 × 每份平均分块数（通常 20–100 段）。100 份 100 页文档 ≈ 100 × 200 = 2 万段，pgvector 完全够。1 万份 100 页 ≈ 200 万段，仍在 pgvector 范围内。5 万份以上再考虑 Milvus。",[13,262,263],{"id":263},"配模型",[17,265,266,267,270],{},"进入后台 → ",[21,268,269],{},"账号 → 模型提供商","，必须同时配置两类模型：",[272,273,274,280],"ol",{},[45,275,276,279],{},[21,277,278],{},"对话模型 LLM","：DeepSeek-V3 \u002F GPT-5 \u002F Claude Sonnet 4 \u002F 豆包-Pro \u002F Kimi K2 都行",[45,281,282,285,286,289,290,293],{},[21,283,284],{},"嵌入模型 Embedding","：",[90,287,288],{},"bge-large-zh-v1.5","（中文首选）、",[90,291,292],{},"text-embedding-3-large","（OpenAI 兼容）",[17,295,296,297,300],{},"常见坑：",[21,298,299],{},"只配对话模型没配嵌入模型","，知识库上传后无法索引。请务必两类都配。",[17,302,303,304,307],{},"如果预算敏感，推荐组合：",[21,305,306],{},"DeepSeek-V3（对话）+ bge-large-zh（嵌入 + 本地）","，token 便宜、中文强。",[13,309,310],{"id":310},"建第一个知识库",[17,312,313,314,317],{},"进入 ",[21,315,316],{},"知识库"," → 新建 → 输入名称、简介、选择嵌入模型。",[17,319,320,323],{},[21,321,322],{},"上传文档","：拖入 PDF \u002F Word \u002F Markdown \u002F TXT。上传后 FastGPT 会做：",[272,325,326,329,332,335],{},[45,327,328],{},"文本提取（PDF 会走 pdfjs \u002F minerU）",[45,330,331],{},"切片（默认按段落 + 长度）",[45,333,334],{},"向量化（调 embedding 模型）",[45,336,337],{},"入库（PG\u002FMilvus）",[17,339,340,285],{},[21,341,342],{},"关键参数",[181,344,345,357],{},[184,346,347],{},[187,348,349,352,355],{},[190,350,351],{},"参数",[190,353,354],{},"默认",[190,356,198],{},[200,358,359,370,381],{},[187,360,361,364,367],{},[205,362,363],{},"chunk_size",[205,365,366],{},"512",[205,368,369],{},"每块字符数，长文档调高到 800–1000",[187,371,372,375,378],{},[205,373,374],{},"chunk_overlap",[205,376,377],{},"50",[205,379,380],{},"块之间重叠，避免语义断裂",[187,382,383,386,389],{},[205,384,385],{},"分段策略",[205,387,388],{},"按段落",[205,390,391],{},"长法律 \u002F 医疗文档改\"按章节\"",[17,393,394,395,398],{},"上传后可以在",[21,396,397],{},"详情页看每一段的切片结果","——这一步一定要抽样看，很多知识库 RAG 效果差是因为切片乱。",[13,400,401],{"id":401},"建第一个应用",[17,403,404,407],{},[21,405,406],{},"应用 → 新建 → 简易应用（对话）","，绑定刚才建的知识库。",[17,409,410],{},"进入调试页，问一个知识库里明确有答案的问题，看：",[272,412,413,419,425],{},[45,414,415,418],{},[21,416,417],{},"是否召回","了相关 chunk（右侧会显示）",[45,420,421,424],{},[21,422,423],{},"答案是否引用","了正确的段落",[45,426,427,430],{},[21,428,429],{},"答案是否幻觉","（没引用凭空说话）",[17,432,433],{},"如果召回不对，回到知识库调 chunk_size \u002F overlap；如果召回对但答案不对，调 prompt 或换模型。",[13,435,437],{"id":436},"rag-调参五步都能调","RAG 调参：五步都能调",[17,439,440,441,31],{},"FastGPT 相比 Coze\u002FDify 最大优势是 ",[21,442,443],{},"RAG 五步都能可视化调",[445,446,448],"h3",{"id":447},"_1-问题预处理","1. 问题预处理",[17,450,451],{},"在应用节点里打开\"问题优化\"：",[42,453,454,460,466],{},[45,455,456,459],{},[21,457,458],{},"改写","：把口语化问题改成结构化查询",[45,461,462,465],{},[21,463,464],{},"扩展","：一个问题拆成多个子问题，分别检索再合并",[45,467,468,471],{},[21,469,470],{},"纠错","：修正错别字",[17,473,474],{},"对不熟悉领域术语的用户，问题预处理能让召回率提升 10%–20%。",[445,476,478],{"id":477},"_2-检索策略","2. 检索策略",[17,480,481],{},"三种检索：",[42,483,484,490,496],{},[45,485,486,489],{},[21,487,488],{},"语义检索","：向量相似度，适合意图理解",[45,491,492,495],{},[21,493,494],{},"关键词 BM25","：字面匹配，适合专业术语、编号",[45,497,498,501],{},[21,499,500],{},"混合检索","：两者加权，通常最好",[17,503,504],{},"对政策、条款、代号密集的场景（如法律条文、药品编号），一定要开混合检索。",[445,506,508],{"id":507},"_3-重排序-rerank","3. 重排序 Rerank",[17,510,511],{},"在检索节点加一个 reranker，可以用：",[42,513,514,520],{},[45,515,516,519],{},[21,517,518],{},"BGE Reranker","（本地，中文好）",[45,521,522,525],{},[21,523,524],{},"Cohere Rerank","（云端，多语言）",[17,527,528,529,532],{},"Rerank 会把初步召回的前 20 段进一步排序，选出最相关的 5–10 段。",[21,530,531],{},"对精度提升明显","，但每次调用会增加 200–500ms 延迟。",[445,534,536],{"id":535},"_4-上下文组装","4. 上下文组装",[17,538,539,540,285],{},"控制",[21,541,542],{},"送给 LLM 的 chunk 数量",[42,544,545,551],{},[45,546,547,550],{},[21,548,549],{},"少（3–5 段）","：答案精准但可能漏信息",[45,552,553,556],{},[21,554,555],{},"多（10–20 段）","：召回全但 token 贵、幻觉风险高",[17,558,559],{},"通用推荐 5–8 段起步，看真实业务再调。",[445,561,563],{"id":562},"_5-答案生成","5. 答案生成",[17,565,566],{},"调 prompt 的 4 个关键点：",[42,568,569,572,575,578],{},[45,570,571],{},"明确\"只根据参考资料回答\"",[45,573,574],{},"强制\"引用来源段号\"",[45,576,577],{},"拒答\"参考资料里没有的问题\"",[45,579,580],{},"输出格式（Markdown \u002F JSON \u002F 表格）",[13,582,584],{"id":583},"上线前-checklist","上线前 checklist",[42,586,589,603,609,615,621,627,633,639,645],{"className":587},[588],"contains-task-list",[45,590,593,598,599,602],{"className":591},[592],"task-list-item",[594,595],"input",{"disabled":596,"type":597},true,"checkbox"," ",[90,600,601],{},"root"," 密码已改，MinIO 默认密码已改",[45,604,606,608],{"className":605},[592],[594,607],{"disabled":596,"type":597}," 服务器只开必要端口，3000 走 Nginx 反代 + HTTPS",[45,610,612,614],{"className":611},[592],[594,613],{"disabled":596,"type":597}," 数据库（PG\u002FMilvus）已定期备份",[45,616,618,620],{"className":617},[592],[594,619],{"disabled":596,"type":597}," LLM API Key 走 secret 管理，不写到代码里",[45,622,624,626],{"className":623},[592],[594,625],{"disabled":596,"type":597}," 至少 20 条测试问答已跑过，召回准确率 > 80%",[45,628,630,632],{"className":629},[592],[594,631],{"disabled":596,"type":597}," 幻觉抽查：对知识库里没有的问题应该拒答",[45,634,636,638],{"className":635},[592],[594,637],{"disabled":596,"type":597}," 监控：接了日志（LLM 调用量、错误率、延迟）",[45,640,642,644],{"className":641},[592],[594,643],{"disabled":596,"type":597}," 限流：单用户 QPM、单知识库 QPS 都有上限",[45,646,648,650],{"className":647},[592],[594,649],{"disabled":596,"type":597}," 内网 DNS 已配好，员工能通过友好域名访问",[13,652,653],{"id":653},"常见坑",[445,655,656],{"id":656},"部署阶段",[42,658,659,677,683,692],{},[45,660,661,664,665,672,673,676],{},[21,662,663],{},"镜像 tag 对不上","：编排文件里的 image tag 不一定是最新 release，",[666,667,671],"a",{"href":668,"rel":669},"https:\u002F\u002Fcloud.tencent.com\u002Fdeveloper\u002Farticle\u002F2632669",[670],"nofollow","腾讯云教程"," 提到过；把 ",[90,674,675],{},"image:"," 手动改成对应 release tag",[45,678,679,682],{},[21,680,681],{},"PG 数据卷权限错","：first run 有时 PG 会因 volume 权限报错，删卷重来一次即可",[45,684,685,691],{},[21,686,687,688],{},"MinIO 默认密码 ",[90,689,690],{},"minioadmin\u002Fminioadmin","：公网部署前必须改，否则文档会被爬走",[45,693,694,697,698,701],{},[21,695,696],{},"服务器时间不同步","：token 签发失败，登录不上；",[90,699,700],{},"ntpdate"," 同步一下",[445,703,704],{"id":704},"建库阶段",[42,706,707,713,719],{},[45,708,709,712],{},[21,710,711],{},"忘配 embedding 模型","：只配 GPT-4 对话模型，上传知识库后无法索引，界面显示\"处理中\"永远不结束",[45,714,715,718],{},[21,716,717],{},"切片太碎","：默认 512 字符对代码 \u002F 长条款不友好，改成\"按章节 + 800 字符 + overlap 100\"",[45,720,721,724],{},[21,722,723],{},"上传大 PDF 卡住","：单文件 > 50MB 时切分工作会占用大量内存，先本地拆成小 PDF 再上传",[445,726,727],{"id":727},"调参阶段",[42,729,730,736,742],{},[45,731,732,735],{},[21,733,734],{},"只调 prompt 不调切片","：答案不准 90% 是切片问题，不是 prompt 问题；先看召回段是否正确，再改 prompt",[45,737,738,741],{},[21,739,740],{},"rerank 全开不看延迟","：对话延迟从 2s 涨到 5s 会影响用户体验，只对精度敏感场景开 rerank",[45,743,744,747],{},[21,745,746],{},"上下文塞太多","：一次给 LLM 20 段参考资料，token 成本涨 4 倍，幻觉反而更多",[445,749,750],{"id":750},"上线阶段",[42,752,753,759,765],{},[45,754,755,758],{},[21,756,757],{},"没做限流","：员工问出恶意长文本 prompt，一次烧掉几十美元；后台一定要配 QPM 限制",[45,760,761,764],{},[21,762,763],{},"知识库没版本管理","：文档更新后没重建 embedding，答案继续用老版本内容；定期重建 + 记录版本号",[45,766,767,770],{},[21,768,769],{},"没日志","：出错误答案后无法回溯是哪段召回的哪个 chunk 出的问题",[13,772,774],{"id":773},"和-dify-coze-的衔接","和 Dify \u002F Coze 的衔接",[17,776,777,778,285],{},"如果你未来会加入 Dify 或 Coze，FastGPT 提供 ",[21,779,780],{},"OpenAI 兼容的对话 API 和知识库检索 API",[42,782,783,786],{},[45,784,785],{},"在 Dify 里把 FastGPT 应用地址配成\"OpenAI 兼容模型端点\"，可以直接在 Dify 工作流里用 FastGPT 的知识库能力。",[45,787,788],{},"在 Coze 里通过\"外部 API 插件\"调 FastGPT 检索 API，把召回内容拼进 Coze 的 Bot prompt。",[17,790,791],{},"这两个组合让\"FastGPT 做知识库底座 + Dify\u002FCoze 做应用层\"变得非常自然。",[13,793,794],{"id":794},"相关阅读",[42,796,797,803,809,815,821],{},[45,798,799],{},[666,800,802],{"href":801},"\u002Freview\u002Ffastgpt-deep-review.html","FastGPT 深度评测：国内企业知识库首选？",[45,804,805],{},[666,806,808],{"href":807},"\u002Fcompare\u002Ffastgpt-vs-dify.html","FastGPT vs Dify：国内企业级 RAG 与 Agent 平台怎么选",[45,810,811],{},[666,812,814],{"href":813},"\u002Fagent\u002Fplatform\u002Ffastgpt.html","FastGPT 工具卡",[45,816,817],{},[666,818,820],{"href":819},"\u002Freview\u002Fcoze-vs-dify.html","Coze vs Dify 对比评测",[45,822,823],{},[666,824,826],{"href":825},"\u002Freview\u002Fllm-gateway-comparison.html","LLM Gateway 与 API 网关横评",[828,829,830],"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 .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":88,"searchDepth":120,"depth":120,"links":832},[833,834,835,836,837,838,839,840,847,848,854,855],{"id":15,"depth":110,"text":15},{"id":37,"depth":110,"text":37},{"id":77,"depth":110,"text":78},{"id":172,"depth":110,"text":172},{"id":263,"depth":110,"text":263},{"id":310,"depth":110,"text":310},{"id":401,"depth":110,"text":401},{"id":436,"depth":110,"text":437,"children":841},[842,843,844,845,846],{"id":447,"depth":120,"text":448},{"id":477,"depth":120,"text":478},{"id":507,"depth":120,"text":508},{"id":535,"depth":120,"text":536},{"id":562,"depth":120,"text":563},{"id":583,"depth":110,"text":584},{"id":653,"depth":110,"text":653,"children":849},[850,851,852,853],{"id":656,"depth":120,"text":656},{"id":704,"depth":120,"text":704},{"id":727,"depth":120,"text":727},{"id":750,"depth":120,"text":750},{"id":773,"depth":110,"text":774},{"id":794,"depth":110,"text":794},"onboarding","\u002Fimg\u002Ftools\u002Ffastgpt.webp","FastGPT 从零上手教程：docker-compose 十分钟部署、模型和向量库选型、第一次导入知识库的正确流程、RAG 调参思路、常见坑和上线前 checklist。","md",{},"\u002Fplaybook\u002Fonboarding\u002Ffastgpt-getting-started","2026-07-02",[864,865,866],"agent\u002Fplatform\u002Ffastgpt","agent\u002Fplatform\u002Fdify","agent\u002Fplatform\u002Fcoze",{"title":8,"description":858},"playbook\u002Fonboarding\u002Ffastgpt-getting-started",[870,871,316,872,873,874],"FastGPT","RAG","Docker","Agent 平台","私有部署","y-tYGWEs4uNqL0F2Den59oiO4ibPv0IpgcJXxm8uTqk",{"tools":4,"reviews":5,"playbooks":877,"news":878},19,13,[880,1800,2836],{"id":881,"title":882,"alternatives":883,"api_compatible":885,"body":886,"category":1721,"chinese_friendly":1722,"cover":1723,"description":1724,"domestic":1725,"extension":859,"faq":1726,"free":1725,"github":1726,"languages":1727,"meta":1730,"models":1731,"navigation":596,"notSuitable":1737,"opensource":1725,"path":1741,"pillar":1742,"platforms":1743,"priceTable":1745,"pricing":1766,"published":1767,"relatedPlaybooks":1726,"relatedReviews":1726,"score":1768,"self_host":1725,"seo":1769,"seoTitle":1770,"slug":866,"sources":1771,"stem":1782,"suitable":1783,"tagline":1789,"tags":1790,"updated":1797,"verdict":1798,"website":910,"__hash__":1799},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Fcoze.md","Coze",[865,864,884],"agent\u002Fplatform\u002Fyuanqi",[],{"type":10,"value":887,"toc":1705},[888,892,933,939,942,946,949,963,972,1022,1031,1035,1038,1070,1078,1082,1085,1105,1109,1191,1200,1203,1206,1226,1235,1239,1275,1278,1281,1423,1438,1469,1472,1545,1549,1552,1569,1572,1602,1604,1668,1671,1698],[13,889,891],{"id":890},"tldr","TL;DR",[893,894,899,920],"div",{"className":895},[896,897,898],"card","p-5","my-4",[17,900,901,904,905,919],{},[21,902,903],{},"一句话："," 字节跳动出品的低代码 Agent 平台，",[21,906,907,908,913,914],{},"国内版 ",[666,909,912],{"href":910,"rel":911},"https:\u002F\u002Fwww.coze.cn",[670],"coze.cn","（中文叫\"扣子\"）+ 国际版 ",[666,915,918],{"href":916,"rel":917},"https:\u002F\u002Fwww.coze.com",[670],"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。",[17,921,922,923,926,927,932],{},"最大价值是 ",[21,924,925],{},"零代码、最快上手","——文档详细 + 中文社区活跃 + 模板多，根据 ",[666,928,931],{"href":929,"rel":930},"https:\u002F\u002Fwww.cnblogs.com\u002Fuulucias\u002Fp\u002F19449008",[670],"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\"。",[934,935,936],"blockquote",{},[17,937,938],{},"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。",[13,940,941],{"id":941},"核心特性",[445,943,945],{"id":944},"可视化工作流最大卖点","可视化工作流（最大卖点）",[17,947,948],{},"Coze 把 Agent 拆成两层：",[42,950,951,957],{},[45,952,953,956],{},[21,954,955],{},"Bot \u002F 智能体","：对话式 AI，配置 prompt + 知识库 + 插件",[45,958,959,962],{},[21,960,961],{},"工作流（Workflow）","：DAG 节点编排，可被 Bot 调用，也可独立部署",[17,964,965,966,971],{},"工作流节点类型（基于 ",[666,967,970],{"href":968,"rel":969},"https:\u002F\u002Fdeveloper.volcengine.com\u002Farticles\u002F7530117616687480851",[670],"火山引擎社区 2025 实战","）：",[42,973,974,980,986,992,998,1004,1010,1016],{},[45,975,976,979],{},[21,977,978],{},"开始 \u002F 结束节点","：输入输出",[45,981,982,985],{},[21,983,984],{},"大模型节点","：调豆包 \u002F GPT \u002F Claude 任一模型",[45,987,988,991],{},[21,989,990],{},"代码节点","：内嵌 Python \u002F JavaScript（飞书插件常需要数据格式转换）",[45,993,994,997],{},[21,995,996],{},"循环节点","：批量处理多条数据",[45,999,1000,1003],{},[21,1001,1002],{},"条件节点","：分支判断",[45,1005,1006,1009],{},[21,1007,1008],{},"插件节点","：调 Coze 插件市场的工具",[45,1011,1012,1015],{},[21,1013,1014],{},"知识库节点","：RAG 检索",[45,1017,1018,1021],{},[21,1019,1020],{},"HTTP 节点","：调外部 API",[17,1023,1024,1025,1030],{},"典型用例（参考 ",[666,1026,1029],{"href":1027,"rel":1028},"https:\u002F\u002Fwww.toutiao.com\u002Farticle\u002F7469986334686315017",[670],"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。",[445,1032,1034],{"id":1033},"插件市场200-官方-海量第三方","插件市场（200+ 官方 + 海量第三方）",[17,1036,1037],{},"Coze 的\"插件\"是封装好的 API 工具，比如：",[42,1039,1040,1046,1052,1058,1064],{},[45,1041,1042,1045],{},[21,1043,1044],{},"飞书多维表格","：增删改查记录（国内 toB 场景的杀手锏）",[45,1047,1048,1051],{},[21,1049,1050],{},"图像生成","：调豆包 \u002F SD \u002F DALL-E",[45,1053,1054,1057],{},[21,1055,1056],{},"联网搜索","：实时网页检索",[45,1059,1060,1063],{},[21,1061,1062],{},"代码执行","：在线运行 Python",[45,1065,1066,1069],{},[21,1067,1068],{},"第三方 SaaS","：微博、抖音、bilibili、Notion……",[17,1071,1072,1077],{},[666,1073,1076],{"href":1074,"rel":1075},"https:\u002F\u002Fkouziai.github.io\u002F",[670],"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。",[445,1079,1081],{"id":1080},"bot-商店-多平台一键发布","Bot 商店 + 多平台一键发布",[17,1083,1084],{},"发布渠道：",[42,1086,1087,1090,1093,1096,1099,1102],{},[45,1088,1089],{},"飞书机器人（一键绑）",[45,1091,1092],{},"抖音 \u002F 头条号",[45,1094,1095],{},"微信小程序 \u002F 公众号（部分需企业认证）",[45,1097,1098],{},"自定义网页嵌入",[45,1100,1101],{},"API 接口（提供 OpenAPI 风格 REST 调用）",[45,1103,1104],{},"Discord（国际版）",[445,1106,1108],{"id":1107},"国内版-vs-国际版","国内版 vs 国际版",[181,1110,1111,1124],{},[184,1112,1113],{},[187,1114,1115,1118,1121],{},[190,1116,1117],{},"维度",[190,1119,1120],{},"扣子（coze.cn）",[190,1122,1123],{},"Coze（coze.com）",[200,1125,1126,1137,1148,1158,1169,1180],{},[187,1127,1128,1131,1134],{},[205,1129,1130],{},"主力模型",[205,1132,1133],{},"豆包 Pro \u002F DeepSeek \u002F Qwen \u002F Kimi",[205,1135,1136],{},"OpenAI \u002F Claude \u002F Gemini \u002F Cohere",[187,1138,1139,1142,1145],{},[205,1140,1141],{},"飞书 \u002F 抖音 \u002F 微信集成",[205,1143,1144],{},"✅ 原生",[205,1146,1147],{},"❌",[187,1149,1150,1153,1156],{},[205,1151,1152],{},"Discord \u002F Slack 集成",[205,1154,1155],{},"⚠️ 有限",[205,1157,1144],{},[187,1159,1160,1163,1166],{},[205,1161,1162],{},"数据存储位置",[205,1164,1165],{},"国内",[205,1167,1168],{},"海外",[187,1170,1171,1174,1177],{},[205,1172,1173],{},"支付",[205,1175,1176],{},"微信 \u002F 支付宝",[205,1178,1179],{},"海外信用卡",[187,1181,1182,1185,1188],{},[205,1183,1184],{},"内容合规",[205,1186,1187],{},"严格审核",[205,1189,1190],{},"宽松",[17,1192,1193,1196,1197,31],{},[21,1194,1195],{},"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 ",[21,1198,1199],{},"不互通",[13,1201,1202],{"id":1202},"价格与运行成本",[17,1204,1205],{},"国内版（扣子）：",[42,1207,1208,1214,1220],{},[45,1209,1210,1213],{},[21,1211,1212],{},"免费版","：免费模型有日额度（豆包 lite 等），适合个人玩 \u002F Demo",[45,1215,1216,1219],{},[21,1217,1218],{},"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",[45,1221,1222,1225],{},[21,1223,1224],{},"企业版","：议价，含 VPC、私有化（限定场景）、SLA",[17,1227,1228,1229,1234],{},"国际版（coze.com）的定价模式据 ",[666,1230,1233],{"href":1231,"rel":1232},"https:\u002F\u002Fdocs.coze.com\u002F",[670],"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。",[13,1236,1238],{"id":1237},"上手-10-分钟","上手 10 分钟",[272,1240,1241,1254,1257,1260,1263,1266,1269,1272],{},[45,1242,1243,1244,1248,1249,1253],{},"打开 ",[666,1245,1247],{"href":910,"rel":1246},[670],"www.coze.cn","（国内）或 ",[666,1250,1252],{"href":916,"rel":1251},[670],"www.coze.com","（国际），用飞书 \u002F 抖音账号 \u002F Google 账号登录",[45,1255,1256],{},"左侧\"工作空间\" → \"+创建 Bot\"，起个名字",[45,1258,1259],{},"选模型（国内推荐豆包 Pro，国际推 Claude Sonnet 4）",[45,1261,1262],{},"写 Bot 角色 prompt",[45,1264,1265],{},"可选：上传 PDF \u002F 文档建知识库",[45,1267,1268],{},"测试一下对话效果",[45,1270,1271],{},"右上\"发布\" → 选渠道（飞书 \u002F 抖音 \u002F API \u002F Web）",[45,1273,1274],{},"拿到调用 URL \u002F 飞书机器人 webhook",[17,1276,1277],{},"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。",[13,1279,1280],{"id":1280},"与同类怎么选",[181,1282,1283,1307],{},[184,1284,1285],{},[187,1286,1287,1289,1291,1297,1301],{},[190,1288,1117],{},[190,1290,882],{},[190,1292,1293],{},[666,1294,1296],{"href":1295},"\u002Fagent\u002Fplatform\u002Fdify.html","Dify",[190,1298,1299],{},[666,1300,870],{"href":813},[190,1302,1303],{},[666,1304,1306],{"href":1305},"\u002Fagent\u002Fplatform\u002Fyuanqi.html","元器 yuanqi",[200,1308,1309,1323,1337,1353,1369,1382,1396,1410],{},[187,1310,1311,1314,1316,1319,1321],{},[205,1312,1313],{},"开源",[205,1315,1147],{},[205,1317,1318],{},"✅",[205,1320,1318],{},[205,1322,1147],{},[187,1324,1325,1327,1330,1332,1334],{},[205,1326,874],{},[205,1328,1329],{},"⚠️ 仅企业版",[205,1331,1318],{},[205,1333,1318],{},[205,1335,1336],{},"⚠️",[187,1338,1339,1342,1345,1348,1350],{},[205,1340,1341],{},"上手难度",[205,1343,1344],{},"★ 最简单",[205,1346,1347],{},"★★★",[205,1349,1347],{},[205,1351,1352],{},"★★",[187,1354,1355,1358,1361,1364,1367],{},[205,1356,1357],{},"工作流编排",[205,1359,1360],{},"★★★★☆",[205,1362,1363],{},"★★★★★",[205,1365,1366],{},"★★★☆☆",[205,1368,1366],{},[187,1370,1371,1374,1376,1378,1380],{},[205,1372,1373],{},"RAG 精度",[205,1375,1366],{},[205,1377,1360],{},[205,1379,1363],{},[205,1381,1366],{},[187,1383,1384,1387,1389,1391,1393],{},[205,1385,1386],{},"字节生态",[205,1388,1363],{},[205,1390,1147],{},[205,1392,1147],{},[205,1394,1395],{},"❌（腾讯系）",[187,1397,1398,1401,1404,1406,1408],{},[205,1399,1400],{},"插件市场",[205,1402,1403],{},"★★★★★ 200+",[205,1405,1366],{},[205,1407,1366],{},[205,1409,1366],{},[187,1411,1412,1415,1417,1419,1421],{},[205,1413,1414],{},"中文社区",[205,1416,1363],{},[205,1418,1360],{},[205,1420,1360],{},[205,1422,1360],{},[17,1424,1425,1428,1429,57,1434,971],{},[21,1426,1427],{},"怎么选","（综合 ",[666,1430,1433],{"href":1431,"rel":1432},"https:\u002F\u002Fwww.besthub.dev\u002Farticles\u002Fcoze-vs-dify-vs-fastgpt-which-ai-agent-platform-fits-your-needs-fa59cf97b798",[670],"BestHub 2025-07",[666,1435,1437],{"href":929,"rel":1436},[670],"博客园 2026-01",[42,1439,1440,1446,1454,1461],{},[45,1441,1442,1445],{},[21,1443,1444],{},"快速验证 \u002F 不懂代码 \u002F 1-2 天出原型"," → Coze",[45,1447,1448,1451,1452],{},[21,1449,1450],{},"数据安全要求高 \u002F 复杂业务流程 \u002F 有技术团队"," → ",[666,1453,1296],{"href":1295},[45,1455,1456,1451,1459],{},[21,1457,1458],{},"核心场景就是企业知识库 QA",[666,1460,870],{"href":813},[45,1462,1463,1451,1466],{},[21,1464,1465],{},"QQ \u002F 微信生态 + 腾讯系",[666,1467,1468],{"href":1305},"元器",[13,1470,1471],{"id":1471},"避坑清单",[42,1473,1474,1484,1490,1505,1515,1521,1533,1539],{},[45,1475,1476,1479,1480,1483],{},[21,1477,1478],{},"国内版 vs 国际版的\"双账号陷阱\"","：扣子（coze.cn）和 Coze（coze.com）是",[21,1481,1482],{},"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",[45,1485,1486,1489],{},[21,1487,1488],{},"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",[45,1491,1492,1495,1496,1499,1500,1504],{},[21,1493,1494],{},"飞书多维表格插件数据格式坑","：写入多维表格需要 ",[90,1497,1498],{},"Array\u003CObject>"," 格式，代码节点要做转换（参考 ",[666,1501,1503],{"href":968,"rel":1502},[670],"火山引擎 2025 教程","）",[45,1506,1507,1510,1511,1514],{},[21,1508,1509],{},"工作流读取飞书表格默认 20 条","：要改 ",[90,1512,1513],{},"page_size","，最大 500 条；超过 500 要分页或循环",[45,1516,1517,1520],{},[21,1518,1519],{},"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",[45,1522,1523,1526,1527,1529,1530,1532],{},[21,1524,1525],{},"企业版\"私有化\"是有限的","：完全数据不出网仍建议 ",[666,1528,1296],{"href":1295}," \u002F ",[666,1531,870],{"href":813}," 自托管",[45,1534,1535,1538],{},[21,1536,1537],{},"国际版接 Claude \u002F GPT 需要 BYOK","：自己绑海外信用卡，平台不代付",[45,1540,1541,1544],{},[21,1542,1543],{},"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",[13,1546,1548],{"id":1547},"适合-不适合","适合 \u002F 不适合",[17,1550,1551],{},"✅ 适合：",[42,1553,1554,1557,1560,1563,1566],{},[45,1555,1556],{},"产品 \u002F 运营 \u002F 非技术人员快速做 Bot",[45,1558,1559],{},"在飞书 \u002F 抖音 \u002F 头条生态内做集成",[45,1561,1562],{},"个人副业（小红书账号批量内容生成等）",[45,1564,1565],{},"中小企业客服 Bot（3 天上线）",[45,1567,1568],{},"想用豆包 \u002F DeepSeek 国产模型的人",[17,1570,1571],{},"❌ 不适合：",[42,1573,1574,1577,1580,1585,1592],{},[45,1575,1576],{},"金融 \u002F 政府 \u002F 医疗（数据敏感，需自托管）",[45,1578,1579],{},"复杂业务系统深度集成（自由度不够）",[45,1581,1582,1583,1504],{},"反感字节生态（去 ",[666,1584,1296],{"href":1295},[45,1586,1587,1588,1529,1590,1504],{},"希望开源 \u002F 完全自主可控（去 ",[666,1589,1296],{"href":1295},[666,1591,870],{"href":813},[45,1593,1594,1595,1529,1597,1601],{},"海外 toB SaaS 产品后端（",[666,1596,1296],{"href":1295},[666,1598,1600],{"href":1599},"\u002Fagent\u002Fplatform\u002Flangflow.html","Langflow"," 更合适）",[13,1603,794],{"id":794},[42,1605,1606,1621,1638,1657],{},[45,1607,1608,1609,1529,1611,1529,1613,1529,1615,1529,1617],{},"同类对比：",[666,1610,1296],{"href":1295},[666,1612,870],{"href":813},[666,1614,1468],{"href":1305},[666,1616,1600],{"href":1599},[666,1618,1620],{"href":1619},"\u002Fagent\u002Fplatform\u002Fn8n.html","n8n",[45,1622,1623,1624,1529,1628,1529,1631,1529,1635],{},"概念：",[666,1625,1627],{"href":1626},"\u002Fwiki\u002Fai-agent.html","AI Agent",[666,1629,871],{"href":1630},"\u002Fwiki\u002Frag.html",[666,1632,1634],{"href":1633},"\u002Fwiki\u002Ffunction-calling.html","Function Calling",[666,1636,1637],{"href":1626},"Multi-Agent",[45,1639,1640,1641,1529,1645,1529,1649,1529,1653],{},"模型：",[666,1642,1644],{"href":1643},"\u002Fmodels\u002Fdoubao-1-5-pro.html","豆包 Doubao",[666,1646,1648],{"href":1647},"\u002Fmodels\u002Fdeepseek-v3.html","DeepSeek-V3",[666,1650,1652],{"href":1651},"\u002Fmodels\u002Fqwen-3.html","Qwen3",[666,1654,1656],{"href":1655},"\u002Fmodels\u002Fglm-5.2.html","GLM-5.2",[45,1658,1659,1660,1529,1664],{},"进阶：",[666,1661,1663],{"href":1662},"\u002Fwiki\u002Fprompt-engineering.html","Prompt Engineering",[666,1665,1667],{"href":1666},"\u002Fwiki\u002Fcontext-engineering.html","Context Engineering",[13,1669,1670],{"id":1670},"来源",[42,1672,1673,1679,1685,1692,1695],{},[45,1674,1675,1676],{},"国内版：",[666,1677,910],{"href":910,"rel":1678},[670],[45,1680,1681,1682],{},"国际版：",[666,1683,916],{"href":916,"rel":1684},[670],[45,1686,1687,1688],{},"官方文档：",[666,1689,1690],{"href":1690,"rel":1691},"https:\u002F\u002Fdocs.coze.com",[670],[45,1693,1694],{},"第三方选型评测：cnblogs.com \u002F besthub.dev \u002F aibotgo.net",[45,1696,1697],{},"实战教程：火山引擎社区、今日头条 涛哥讲AI",[17,1699,1700,1701,1704],{},"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 ",[666,1702,1703],{"href":1703},"\u002Fsubmit"," 反馈。",{"title":88,"searchDepth":120,"depth":120,"links":1706},[1707,1708,1714,1715,1716,1717,1718,1719,1720],{"id":890,"depth":110,"text":891},{"id":941,"depth":110,"text":941,"children":1709},[1710,1711,1712,1713],{"id":944,"depth":120,"text":945},{"id":1033,"depth":120,"text":1034},{"id":1080,"depth":120,"text":1081},{"id":1107,"depth":120,"text":1108},{"id":1202,"depth":110,"text":1202},{"id":1237,"depth":110,"text":1238},{"id":1280,"depth":110,"text":1280},{"id":1471,"depth":110,"text":1471},{"id":1547,"depth":110,"text":1548},{"id":794,"depth":110,"text":794},{"id":1670,"depth":110,"text":1670},"platform",5,"\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,[1728,1729],"zh","en",{},[1732,1733,1734,1735,1736],"doubao-pro","deepseek-v3","qwen-max","gpt-4o (国际版)","claude (国际版)",[1738,1739,1740],"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",[1744],"web",[1746,1752,1757,1761],{"plan":1747,"price":1748,"limit":1749,"cn_pay":1750,"note":1751},"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":1218,"price":1753,"limit":1754,"cn_pay":1755,"note":1756},"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":1224,"price":1758,"limit":1759,"cn_pay":1318,"note":1760},"议价","VPC、合规、SLA、私有化（部分）","B 端落地",{"plan":1762,"price":1763,"cn_pay":1764,"note":1765,"limit":1750},"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":129,"ux":1722,"price":129,"cn_support":1722,"stability":129},{"title":882,"description":1724},"Coze 扣子评测 2026：字节 AI Agent 平台，对比 Dify",[1772,1774,1776,1778,1780],{"title":1773,"url":910},"Coze 国内版（扣子）",{"title":1775,"url":916},"Coze 国际版",{"title":1777,"url":1690},"Coze 官方文档",{"title":1779,"url":929},"Coze vs Dify vs FastGPT 选型 2026",{"title":1781,"url":1431},"BestHub 三平台对比","tools\u002Fagent\u002Fplatform\u002Fcoze",[1784,1785,1786,1787,1788],"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",[1791,1792,1793,1794,1795,1796],"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":1801,"title":1296,"alternatives":1802,"api_compatible":1726,"body":1805,"category":1721,"chinese_friendly":129,"cover":2783,"description":2784,"domestic":1725,"extension":859,"faq":1726,"free":1725,"github":2089,"languages":2785,"meta":2787,"models":1726,"navigation":596,"notSuitable":1726,"opensource":596,"path":2788,"pillar":1742,"platforms":2789,"priceTable":2793,"pricing":2810,"published":1767,"relatedPlaybooks":1726,"relatedReviews":1726,"score":2811,"self_host":596,"seo":2812,"seoTitle":2813,"slug":865,"sources":2814,"stem":2826,"suitable":1726,"tagline":2827,"tags":2828,"updated":1797,"verdict":2834,"website":2729,"__hash__":2835},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Fdify.md",[866,864,1803,1804],"agent\u002Fplatform\u002Fn8n","agent\u002Fplatform\u002Flangflow",{"type":10,"value":1806,"toc":2761},[1807,1809,1827,1832,1834,1838,1841,1910,1913,1917,1920,1934,1950,1954,1957,1968,1972,1980,1991,1995,1998,2000,2004,2013,2074,2081,2085,2093,2143,2146,2166,2170,2173,2233,2239,2241,2333,2336,2365,2367,2495,2507,2542,2544,2616,2618,2620,2640,2642,2671,2673,2720,2722,2753,2758],[13,1808,891],{"id":890},[893,1810,1812,1817],{"className":1811},[896,897,898],[17,1813,1814,1816],{},[21,1815,903],{}," 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。",[17,1818,922,1819,1822,1823,1826],{},[21,1820,1821],{},"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 ",[666,1824,882],{"href":1825},"\u002Fagent\u002Fplatform\u002Fcoze.html"," 折腾，新手得读 1-2 小时文档。",[934,1828,1829],{},[17,1830,1831],{},"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。",[13,1833,941],{"id":941},[445,1835,1837],{"id":1836},"可视化工作流chatflow-workflow","可视化工作流（Chatflow + Workflow）",[17,1839,1840],{},"Dify 把 LLM 应用拆成两种\"应用类型\"：",[181,1842,1843,1856],{},[184,1844,1845],{},[187,1846,1847,1850,1853],{},[190,1848,1849],{},"类型",[190,1851,1852],{},"适合场景",[190,1854,1855],{},"编排范式",[200,1857,1858,1871,1884,1897],{},[187,1859,1860,1865,1868],{},[205,1861,1862],{},[21,1863,1864],{},"Chatbot",[205,1866,1867],{},"简单对话机器人",[205,1869,1870],{},"prompt + tools",[187,1872,1873,1878,1881],{},[205,1874,1875],{},[21,1876,1877],{},"Agent",[205,1879,1880],{},"自主多步任务",[205,1882,1883],{},"ReAct \u002F Function Calling",[187,1885,1886,1891,1894],{},[205,1887,1888],{},[21,1889,1890],{},"Chatflow",[205,1892,1893],{},"对话型工作流（多轮 + 分支）",[205,1895,1896],{},"节点 DAG，带聊天上下文",[187,1898,1899,1904,1907],{},[205,1900,1901],{},[21,1902,1903],{},"Workflow",[205,1905,1906],{},"单次输入→输出（API 模式）",[205,1908,1909],{},"节点 DAG，无对话状态",[17,1911,1912],{},"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\"。",[445,1914,1916],{"id":1915},"rag-知识库","RAG 知识库",[17,1918,1919],{},"内置完整 RAG 链路：",[272,1921,1922,1925,1928,1931],{},[45,1923,1924],{},"上传文档（PDF \u002F Word \u002F Markdown \u002F 网页）",[45,1926,1927],{},"自动分块 + embedding（可配置分段策略和 embedding 模型）",[45,1929,1930],{},"混合检索（向量 + 全文 + 重排）",[45,1932,1933],{},"引用溯源（回答末尾自动附原文片段）",[17,1935,1936,1937,1942,1943,1946,1947,1949],{},"注意：根据 ",[666,1938,1941],{"href":1939,"rel":1940},"https:\u002F\u002Fzhuanlan.zhihu.com\u002Fp\u002F1887141987838309480",[670],"知乎 LLM 实战笔记 2025-03 对比"," 的实测，Dify ",[21,1944,1945],{},"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 ",[666,1948,870],{"href":813},"（实测准确率高 10+ 个百分点），Dify 胜在工作流而非纯 RAG。",[445,1951,1953],{"id":1952},"模型生态40-提供商","模型生态：40+ 提供商",[17,1955,1956],{},"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。",[17,1958,1959,1960,1962,1963,1967],{},"国产模型原生支持（不像 ",[666,1961,870],{"href":813}," 需要 ",[666,1964,1966],{"href":1965},"\u002Fcoding\u002Fapi\u002Fone-api.html","OneAPI"," 中转），是 Dify 在国内 toB 场景流行的关键。",[445,1969,1971],{"id":1970},"mcp-协议支持","MCP 协议支持",[17,1973,1974,1975,1979],{},"Dify 较早接入了 ",[666,1976,1978],{"href":1977},"\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：",[42,1981,1982,1985,1988],{},[45,1983,1984],{},"通过 MCP 调本地 PostgreSQL \u002F SQLite",[45,1986,1987],{},"通过 MCP 调 GitHub \u002F Slack \u002F Linear",[45,1989,1990],{},"通过 MCP 调自家内部系统（写一个 MCP Server 即可）",[445,1992,1994],{"id":1993},"api-first","API-first",[17,1996,1997],{},"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。",[13,1999,1202],{"id":1202},[445,2001,2003],{"id":2002},"云版difyai","云版（dify.ai）",[17,2005,2006,2007,2012],{},"根据 ",[666,2008,2011],{"href":2009,"rel":2010},"https:\u002F\u002Fwww.tooljunction.io\u002Fai-tools\u002Fdify-ai",[670],"tooljunction.io 2026 评测"," 引用的官方定价：",[181,2014,2015,2028],{},[184,2016,2017],{},[187,2018,2019,2022,2025],{},[190,2020,2021],{},"套餐",[190,2023,2024],{},"价格",[190,2026,2027],{},"主要限制",[200,2029,2030,2041,2052,2063],{},[187,2031,2032,2035,2038],{},[205,2033,2034],{},"Sandbox",[205,2036,2037],{},"免费",[205,2039,2040],{},"200 次模型调用，1 app，5MB 知识库",[187,2042,2043,2046,2049],{},[205,2044,2045],{},"Professional",[205,2047,2048],{},"$59\u002F月起",[205,2050,2051],{},"5000 调用\u002F月，多 app，50MB 知识库",[187,2053,2054,2057,2060],{},[205,2055,2056],{},"Team",[205,2058,2059],{},"$159\u002F月起",[205,2061,2062],{},"团队协作、SSO",[187,2064,2065,2068,2071],{},[205,2066,2067],{},"Enterprise",[205,2069,2070],{},"联系销售",[205,2072,2073],{},"定制 SLA、私有云",[17,2075,2076,2077,2080],{},"注意：云版价格只是 Dify 平台费，",[21,2078,2079],{},"模型 API 费用另算","（自带 OpenAI \u002F Anthropic key）。",[445,2082,2084],{"id":2083},"自托管推荐","自托管（推荐）",[17,2086,2087,2092],{},[666,2088,2091],{"href":2089,"rel":2090},"https:\u002F\u002Fgithub.com\u002Flanggenius\u002Fdify",[670],"官方 GitHub 仓库"," 提供 Docker Compose 部署，社区版完全免费可商用：",[83,2094,2096],{"className":85,"code":2095,"language":87,"meta":88,"style":88},"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",[90,2097,2098,2107,2114,2125,2137],{"__ignoreMap":88},[93,2099,2100,2102,2104],{"class":95,"line":96},[93,2101,100],{"class":99},[93,2103,104],{"class":103},[93,2105,2106],{"class":103}," https:\u002F\u002Fgithub.com\u002Flanggenius\u002Fdify.git\n",[93,2108,2109,2111],{"class":95,"line":110},[93,2110,114],{"class":113},[93,2112,2113],{"class":103}," dify\u002Fdocker\n",[93,2115,2116,2119,2122],{"class":95,"line":120},[93,2117,2118],{"class":99},"cp",[93,2120,2121],{"class":103}," .env.example",[93,2123,2124],{"class":103}," .env\n",[93,2126,2127,2130,2133,2135],{"class":95,"line":129},[93,2128,2129],{"class":99},"docker",[93,2131,2132],{"class":103}," compose",[93,2134,141],{"class":103},[93,2136,144],{"class":113},[93,2138,2139],{"class":95,"line":1722},[93,2140,2142],{"class":2141},"sJ8bj","# 默认 http:\u002F\u002Flocalhost \u002F 端口可在 .env 调整\n",[17,2144,2145],{},"硬件门槛（社区共识，非官方硬性要求）：",[42,2147,2148,2154,2160],{},[45,2149,2150,2153],{},[21,2151,2152],{},"最低","：2 核 4G，纯外接 API 模式",[45,2155,2156,2159],{},[21,2157,2158],{},"推荐","：4 核 8G + 至少 30GB 磁盘（向量数据 + 文件存储）",[45,2161,2162,2165],{},[21,2163,2164],{},"企业","：8 核 16G+，单机日活上千",[445,2167,2169],{"id":2168},"真实-tco","真实 TCO",[17,2171,2172],{},"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）：",[181,2174,2175,2188],{},[184,2176,2177],{},[187,2178,2179,2182,2185],{},[190,2180,2181],{},"成本项",[190,2183,2184],{},"云版 Professional",[190,2186,2187],{},"自托管",[200,2189,2190,2201,2211,2222],{},[187,2191,2192,2195,2198],{},[205,2193,2194],{},"平台费",[205,2196,2197],{},"~$2,100（3 年）",[205,2199,2200],{},"$0",[187,2202,2203,2206,2208],{},[205,2204,2205],{},"服务器",[205,2207,2200],{},[205,2209,2210],{},"~$50\u002F月 × 36 = $1,800",[187,2212,2213,2216,2219],{},[205,2214,2215],{},"模型 API",[205,2217,2218],{},"与下同",[205,2220,2221],{},"与上同",[187,2223,2224,2227,2230],{},[205,2225,2226],{},"运维人力",[205,2228,2229],{},"0",[205,2231,2232],{},"约 0.2 人月",[17,2234,2235,2238],{},[21,2236,2237],{},"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。",[13,2240,1238],{"id":1237},[83,2242,2244],{"className":85,"code":2243,"language":87,"meta":88,"style":88},"# 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",[90,2245,2246,2251,2259,2265,2273,2283,2289,2295,2301,2306,2312,2317,2323,2328],{"__ignoreMap":88},[93,2247,2248],{"class":95,"line":96},[93,2249,2250],{"class":2141},"# 1. 自托管（社区版）\n",[93,2252,2253,2255,2257],{"class":95,"line":110},[93,2254,100],{"class":99},[93,2256,104],{"class":103},[93,2258,2106],{"class":103},[93,2260,2261,2263],{"class":95,"line":120},[93,2262,114],{"class":113},[93,2264,2113],{"class":103},[93,2266,2267,2269,2271],{"class":95,"line":129},[93,2268,2118],{"class":99},[93,2270,2121],{"class":103},[93,2272,2124],{"class":103},[93,2274,2275,2277,2279,2281],{"class":95,"line":1722},[93,2276,2129],{"class":99},[93,2278,2132],{"class":103},[93,2280,141],{"class":103},[93,2282,144],{"class":113},[93,2284,2286],{"class":95,"line":2285},6,[93,2287,2288],{"emptyLinePlaceholder":596},"\n",[93,2290,2292],{"class":95,"line":2291},7,[93,2293,2294],{"class":2141},"# 2. 浏览器打开 http:\u002F\u002Flocalhost\n",[93,2296,2298],{"class":95,"line":2297},8,[93,2299,2300],{"class":2141},"#    首次会让你创建 admin 账号\n",[93,2302,2304],{"class":95,"line":2303},9,[93,2305,2288],{"emptyLinePlaceholder":596},[93,2307,2309],{"class":95,"line":2308},10,[93,2310,2311],{"class":2141},"# 3. 进入\"设置 → 模型供应商\"，配置 OpenAI \u002F 国产模型 API key\n",[93,2313,2315],{"class":95,"line":2314},11,[93,2316,2288],{"emptyLinePlaceholder":596},[93,2318,2320],{"class":95,"line":2319},12,[93,2321,2322],{"class":2141},"# 4. 在主界面\"创建空白应用\"，选 Chatflow 或 Workflow\n",[93,2324,2325],{"class":95,"line":878},[93,2326,2327],{"class":2141},"# 5. 拖入\"开始 → LLM → 结束\"节点试一下基础 prompt\n",[93,2329,2330],{"class":95,"line":5},[93,2331,2332],{"class":2141},"# 6. 满意了点右上\"发布\"，自动生成 API endpoint\n",[13,2334,2335],{"id":2335},"国内使用注意事项",[272,2337,2338,2344,2350,2356],{},[45,2339,2340,2343],{},[21,2341,2342],{},"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",[45,2345,2346,2349],{},[21,2347,2348],{},"自托管 + 国产模型"," = 完全国内闭环，是 Dify 在国内最大优势",[45,2351,2352,2355],{},[21,2353,2354],{},"Docker 镜像拉取","：国内可能慢，建议配 Docker registry 镜像（阿里云 \u002F 网易）",[45,2357,2358,2361,2362,2364],{},[21,2359,2360],{},"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 ",[666,2363,882],{"href":1825}," 迁到 Dify",[13,2366,1280],{"id":1280},[181,2368,2369,2389],{},[184,2370,2371],{},[187,2372,2373,2375,2377,2381,2385],{},[190,2374,1117],{},[190,2376,1296],{},[190,2378,2379],{},[666,2380,882],{"href":1825},[190,2382,2383],{},[666,2384,870],{"href":813},[190,2386,2387],{},[666,2388,1620],{"href":1619},[200,2390,2391,2404,2416,2429,2441,2454,2468,2481],{},[187,2392,2393,2395,2397,2399,2401],{},[205,2394,1313],{},[205,2396,1318],{},[205,2398,1147],{},[205,2400,1318],{},[205,2402,2403],{},"✅（fair-code）",[187,2405,2406,2408,2410,2412,2414],{},[205,2407,874],{},[205,2409,1318],{},[205,2411,1147],{},[205,2413,1318],{},[205,2415,1318],{},[187,2417,2418,2420,2422,2425,2427],{},[205,2419,1341],{},[205,2421,1366],{},[205,2423,2424],{},"★★☆☆☆ 最简单",[205,2426,1366],{},[205,2428,1360],{},[187,2430,2431,2433,2435,2437,2439],{},[205,2432,1357],{},[205,2434,1363],{},[205,2436,1360],{},[205,2438,1366],{},[205,2440,1363],{},[187,2442,2443,2445,2447,2449,2451],{},[205,2444,1373],{},[205,2446,1360],{},[205,2448,1366],{},[205,2450,1363],{},[205,2452,2453],{},"★★☆☆☆",[187,2455,2456,2459,2461,2463,2466],{},[205,2457,2458],{},"模型生态",[205,2460,1363],{},[205,2462,1360],{},[205,2464,2465],{},"★★★☆☆（OneAPI 中转）",[205,2467,1360],{},[187,2469,2470,2473,2475,2477,2479],{},[205,2471,2472],{},"中文场景",[205,2474,1360],{},[205,2476,1363],{},[205,2478,1360],{},[205,2480,1366],{},[187,2482,2483,2486,2488,2491,2493],{},[205,2484,2485],{},"字节生态绑定",[205,2487,1147],{},[205,2489,2490],{},"✅（飞书\u002F抖音深度集成）",[205,2492,1147],{},[205,2494,1147],{},[17,2496,2497,2499,2500,57,2503,2506],{},[21,2498,1427],{},"（基于 ",[666,2501,1433],{"href":1431,"rel":2502},[670],[666,2504,1437],{"href":929,"rel":2505},[670]," 两份选型指南综合）：",[42,2508,2509,2515,2522,2528,2535],{},[45,2510,2511,2514],{},[21,2512,2513],{},"数据必须不出内网 + 工作流复杂"," → Dify",[45,2516,2517,1451,2520],{},[21,2518,2519],{},"个人 \u002F 小团队 \u002F 快速原型 + 字节生态",[666,2521,882],{"href":1825},[45,2523,2524,1451,2526],{},[21,2525,1458],{},[666,2527,870],{"href":813},[45,2529,2530,1451,2533],{},[21,2531,2532],{},"重点是连接外部 SaaS（Slack \u002F Notion \u002F 数据库）",[666,2534,1620],{"href":1619},[45,2536,2537,1451,2540],{},[21,2538,2539],{},"要画图式表达 LangChain pipeline",[666,2541,1600],{"href":1599},[13,2543,1471],{"id":1471},[42,2545,2546,2552,2568,2579,2592,2598,2604,2610],{},[45,2547,2548,2551],{},[21,2549,2550],{},"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。",[45,2553,2554,285,2560,2563,2564,2567],{},[21,2555,2556,2559],{},[90,2557,2558],{},".env"," 文件改完忘 restart",[90,2561,2562],{},"docker compose down && up -d","，不是 ",[90,2565,2566],{},"restart","——后者不重新加载 env。",[45,2569,2570,285,2573,2578],{},[21,2571,2572],{},"大版本升级会破坏数据库 schema",[666,2574,2577],{"href":2575,"rel":2576},"https:\u002F\u002Fdocs.dify.ai\u002Fzh-hans",[670],"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。",[45,2580,2581,2584,2585,2587,2588,2591],{},[21,2582,2583],{},"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 ",[90,2586,2558],{}," 的 ",[90,2589,2590],{},"UPLOAD_FILE_SIZE_LIMIT"," 并重启容器。",[45,2593,2594,2597],{},[21,2595,2596],{},"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。",[45,2599,2600,2603],{},[21,2601,2602],{},"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。",[45,2605,2606,2609],{},[21,2607,2608],{},"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。",[45,2611,2612,2615],{},[21,2613,2614],{},"国内 Docker 拉取镜像慢","：先配国内 registry，否则首次 pull 可能要 30+ 分钟。",[13,2617,1548],{"id":1547},[17,2619,1551],{},[42,2621,2622,2625,2628,2631,2634,2637],{},[45,2623,2624],{},"中大型企业 LLM 中台建设",[45,2626,2627],{},"需要私有化部署（金融 \u002F 医疗 \u002F 政府）",[45,2629,2630],{},"想做\"AI 工作流即产品\"的开发团队",[45,2632,2633],{},"同时需要 RAG + Agent + Workflow 三件套",[45,2635,2636],{},"想用国产模型 + 国际模型混合编排",[45,2638,2639],{},"已经接受 Docker + 一定运维投入",[17,2641,1571],{},[42,2643,2644,2650,2656,2659,2665],{},[45,2645,2646,2647,2649],{},"纯个人玩家做对话机器人（",[666,2648,882],{"href":1825}," 更快）",[45,2651,2652,2653,2655],{},"只想做企业知识库 QA（",[666,2654,870],{"href":813}," RAG 更专）",[45,2657,2658],{},"团队完全没运维能力（云版还行，自托管会踩坑）",[45,2660,2661,2662,2664],{},"需要深度对接字节飞书 \u002F 抖音（",[666,2663,882],{"href":1825}," 原生）",[45,2666,2667,2668,2670],{},"工作流核心是连接 100+ SaaS（",[666,2669,1620],{"href":1619}," 节点更全）",[13,2672,794],{"id":794},[42,2674,2675,2685,2697,2712],{},[45,2676,1608,2677,1529,2679,1529,2681,1529,2683],{},[666,2678,882],{"href":1825},[666,2680,870],{"href":813},[666,2682,1620],{"href":1619},[666,2684,1600],{"href":1599},[45,2686,2687,2688,1529,2690,1529,2692,1529,2695],{},"概念基础：",[666,2689,1627],{"href":1626},[666,2691,871],{"href":1630},[666,2693,2694],{"href":1977},"MCP",[666,2696,1634],{"href":1633},[45,2698,2699,2700,1529,2704,1529,2708,1529,2710],{},"模型选型：",[666,2701,2703],{"href":2702},"\u002Fmodels\u002Fgpt-5.html","GPT-5",[666,2705,2707],{"href":2706},"\u002Fmodels\u002Fclaude-sonnet-4.html","Claude Sonnet 4",[666,2709,1648],{"href":1647},[666,2711,1656],{"href":1655},[45,2713,1659,2714,1529,2718],{},[666,2715,2717],{"href":2716},"\u002Fwiki\u002Ffine-tuning-vs-rag.html","Fine-tuning vs RAG",[666,2719,1667],{"href":1666},[13,2721,1670],{"id":1670},[42,2723,2724,2731,2737,2743,2750],{},[45,2725,2726,2727],{},"官网：",[666,2728,2729],{"href":2729,"rel":2730},"https:\u002F\u002Fdify.ai",[670],[45,2732,2733,2734],{},"中文文档：",[666,2735,2575],{"href":2575,"rel":2736},[670],[45,2738,2739,2740],{},"GitHub：",[666,2741,2089],{"href":2089,"rel":2742},[670],[45,2744,2745,2746],{},"官方定价：",[666,2747,2748],{"href":2748,"rel":2749},"https:\u002F\u002Fdify.ai\u002Fpricing",[670],[45,2751,2752],{},"第三方评测：tooljunction.io \u002F chatforest.com \u002F besthub.dev \u002F joshuaopolko.com \u002F 知乎 LLM 实战笔记",[17,2754,2755,2756,1704],{},"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 ",[666,2757,1703],{"href":1703},[828,2759,2760],{},"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":88,"searchDepth":120,"depth":120,"links":2762},[2763,2764,2771,2776,2777,2778,2779,2780,2781,2782],{"id":890,"depth":110,"text":891},{"id":941,"depth":110,"text":941,"children":2765},[2766,2767,2768,2769,2770],{"id":1836,"depth":120,"text":1837},{"id":1915,"depth":120,"text":1916},{"id":1952,"depth":120,"text":1953},{"id":1970,"depth":120,"text":1971},{"id":1993,"depth":120,"text":1994},{"id":1202,"depth":110,"text":1202,"children":2772},[2773,2774,2775],{"id":2002,"depth":120,"text":2003},{"id":2083,"depth":120,"text":2084},{"id":2168,"depth":120,"text":2169},{"id":1237,"depth":110,"text":1238},{"id":2335,"depth":110,"text":2335},{"id":1280,"depth":110,"text":1280},{"id":1471,"depth":110,"text":1471},{"id":1547,"depth":110,"text":1548},{"id":794,"depth":110,"text":794},{"id":1670,"depth":110,"text":1670},"\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。",[1728,1729,2786],"ja",{},"\u002Ftools\u002Fagent\u002Fplatform\u002Fdify",[2790,2791,2792,2129],"windows","macos","linux",[2794,2798,2802,2806],{"plan":2795,"price":2200,"features":2796,"notes":2797},"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":2799,"price":2200,"features":2800,"notes":2801},"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":2803,"price":2048,"features":2804,"notes":2805},"Cloud Professional","更高调用额度 + 团队协作 + 商用支持","商用云首选",{"plan":2807,"price":2808,"features":2809,"notes":2070},"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":1722,"ux":129,"price":1722,"cn_support":129,"stability":129},{"title":1296,"description":2784},"Dify 评测 2026：开源 LLMOps 与 AI Agent 平台，自托管指南",[2815,2817,2819,2821,2823],{"title":2816,"url":2575},"Dify 官方文档（中文）",{"title":2818,"url":2089},"Dify GitHub",{"title":2820,"url":2748},"Dify 官方定价",{"title":2822,"url":1431},"Coze vs Dify vs FastGPT 选型",{"title":2824,"url":2825},"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",[1791,2829,2830,2831,1793,2832,2833],"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":2837,"title":870,"alternatives":2838,"api_compatible":2841,"body":2843,"category":1721,"chinese_friendly":1722,"cover":857,"description":3835,"domestic":1725,"extension":859,"faq":1726,"free":1725,"github":3798,"languages":3836,"meta":3837,"models":3838,"navigation":596,"notSuitable":3842,"opensource":596,"path":3846,"pillar":1742,"platforms":3847,"priceTable":3848,"pricing":3870,"published":1767,"relatedPlaybooks":3871,"relatedReviews":3873,"score":3875,"self_host":596,"seo":3876,"seoTitle":1726,"slug":864,"sources":3877,"stem":3888,"suitable":3889,"tagline":3895,"tags":3896,"updated":1797,"verdict":3900,"website":3792,"__hash__":3901},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Ffastgpt.md",[865,866,2839,2840],"agent\u002Fplatform\u002Fragflow","agent\u002Fplatform\u002Fanythingllm",[2842],"openai",{"type":10,"value":2844,"toc":3817},[2845,2847,2875,2886,2888,2892,2895,2909,2912,2938,2942,2949,2981,2988,2992,3045,3052,3055,3058,3061,3067,3112,3118,3122,3153,3157,3257,3260,3263,3267,3275,3395,3410,3412,3569,3577,3611,3617,3619,3691,3693,3695,3715,3717,3735,3737,3784,3786,3809,3814],[13,2846,891],{"id":890},[893,2848,2850,2864],{"className":2849},[896,897,898],[17,2851,2852,2854,2855,2859,2860,2863],{},[21,2853,903],{}," labring 团队开源的 LLM 知识库 RAG 平台，27k+ GitHub star（截至 2026-03 数据，",[666,2856,2858],{"href":668,"rel":2857},[670],"腾讯云 2026-03 教程"," 引用），Apache 2.0 许可证可商用。",[21,2861,2862],{},"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。",[17,2865,2866,2867,2870,2871,2874],{},"最大价值在 ",[21,2868,2869],{},"国内企业知识库 + 私有部署"," 场景。代价是 ",[21,2872,2873],{},"配置门槛","：docker 基础 + 网络知识 + 一定运维能力。",[934,2876,2877],{},[17,2878,2879,2880,2885],{},"来源说明：本文基于 fastgpt.io 官方页面、github.com\u002Flabring\u002FFastGPT 仓库、",[666,2881,2884],{"href":2882,"rel":2883},"https:\u002F\u002Fwww.nanhuantech.com\u002Fzh\u002Fai-reviews\u002Ffastgpt-2025-review",[670],"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。",[13,2887,941],{"id":941},[445,2889,2891],{"id":2890},"知识库管理核心能力","知识库管理（核心能力）",[17,2893,2894],{},"支持文件类型：",[42,2896,2897,2900,2903,2906],{},[45,2898,2899],{},"文档：PDF \u002F Word \u002F Markdown \u002F TXT \u002F HTML",[45,2901,2902],{},"表格：Excel \u002F CSV",[45,2904,2905],{},"网页：URL 抓取 + 定时同步",[45,2907,2908],{},"API：通过接口推送内容",[17,2910,2911],{},"处理流程：上传 → 文本切分 → 向量化 → 存储 → 可用于问答。支持：",[42,2913,2914,2920,2926,2932],{},[45,2915,2916,2919],{},[21,2917,2918],{},"文件夹分组","：不同主题 \u002F 部门分类",[45,2921,2922,2925],{},[21,2923,2924],{},"多种分块策略","：默认按段落 \u002F 按 token 数 \u002F 自定义",[45,2927,2928,2931],{},[21,2929,2930],{},"批量导入","：脚本化大批量同步",[45,2933,2934,2937],{},[21,2935,2936],{},"定时同步","：网页源自动更新",[445,2939,2941],{"id":2940},"rag-流程编排最强卖点","RAG 流程编排（最强卖点）",[17,2943,2944,2948],{},[666,2945,2947],{"href":2882,"rel":2946},[670],"南环 AI 2026 评测"," 总结的 FastGPT RAG 链路：",[272,2950,2951,2957,2963,2969,2975],{},[45,2952,2953,2956],{},[21,2954,2955],{},"问题预处理","：改写 \u002F 扩展 \u002F 错词纠正（提升召回率）",[45,2958,2959,2962],{},[21,2960,2961],{},"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",[45,2964,2965,2968],{},[21,2966,2967],{},"重排序（Rerank）","：对初步检索结果二次排序，提升相关性",[45,2970,2971,2974],{},[21,2972,2973],{},"上下文组装","：最优 chunk + 问题 → prompt",[45,2976,2977,2980],{},[21,2978,2979],{},"答案生成","：调大模型基于检索结果回答 + 引用标注",[17,2982,2983,2984,2987],{},"每一步都可视化调参，这是 FastGPT 比 Coze \u002F Dify 在 ",[21,2985,2986],{},"纯知识库 QA 精度","上更高的原因。",[445,2989,2991],{"id":2990},"多模型支持不绑定厂商","多模型支持（不绑定厂商）",[181,2993,2994,3004],{},[184,2995,2996],{},[187,2997,2998,3001],{},[190,2999,3000],{},"模型类别",[190,3002,3003],{},"支持",[200,3005,3006,3014,3021,3029,3037],{},[187,3007,3008,3011],{},[205,3009,3010],{},"国产闭源",[205,3012,3013],{},"豆包 \u002F 通义千问 \u002F 文心一言 \u002F 智谱 GLM \u002F Moonshot Kimi \u002F MiniMax",[187,3015,3016,3018],{},[205,3017,1313],{},[205,3019,3020],{},"LLaMA \u002F Qwen \u002F ChatGLM \u002F DeepSeek 等可自部署",[187,3022,3023,3026],{},[205,3024,3025],{},"OpenAI 系",[205,3027,3028],{},"GPT-5 \u002F GPT-5 mini \u002F o3",[187,3030,3031,3034],{},[205,3032,3033],{},"Claude 系",[205,3035,3036],{},"Sonnet 4 \u002F Opus 4 \u002F Haiku",[187,3038,3039,3042],{},[205,3040,3041],{},"嵌入 \u002F 重排",[205,3043,3044],{},"BGE \u002F m3e \u002F OpenAI text-embedding-3",[17,3046,3047,3048,3051],{},"可以在 ",[21,3049,3050],{},"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。",[445,3053,3054],{"id":3054},"工作流与高级编排",[17,3056,3057],{},"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。",[445,3059,3060],{"id":3060},"多向量库选择",[17,3062,3063,3066],{},[666,3064,671],{"href":668,"rel":3065},[670]," 公开的 4 种向量后端：",[181,3068,3069,3079],{},[184,3070,3071],{},[187,3072,3073,3076],{},[190,3074,3075],{},"后端",[190,3077,3078],{},"适用",[200,3080,3081,3089,3096,3104],{},[187,3082,3083,3086],{},[205,3084,3085],{},"PgVector",[205,3087,3088],{},"5000 万索引以下，新手 \u002F 小规模",[187,3090,3091,3093],{},[205,3092,238],{},[205,3094,3095],{},"亿级以上，高性能",[187,3097,3098,3101],{},[205,3099,3100],{},"Zilliz Cloud",[205,3102,3103],{},"Milvus 全托管 SaaS",[187,3105,3106,3109],{},[205,3107,3108],{},"SeekDB \u002F OceanBase",[205,3110,3111],{},"企业级国产化",[17,3113,3114,3115,31],{},"部署时选对应 ",[90,3116,3117],{},"docker-compose.{pgvector|milvus|...}.yml",[445,3119,3121],{"id":3120},"api-与-mcp","API 与 MCP",[42,3123,3124,3130,3136,3147],{},[45,3125,3126,3129],{},[21,3127,3128],{},"对话 API","：流式 \u002F 非流式 HTTP，OpenAI 兼容",[45,3131,3132,3135],{},[21,3133,3134],{},"知识库检索 API","：单独调检索（不走生成）做 hybrid pipeline",[45,3137,3138,3141,3142,3146],{},[21,3139,3140],{},"MCP Server","：3005 端口暴露 MCP SSE 服务，可被 ",[666,3143,3145],{"href":3144},"\u002Fcoding\u002Fcli\u002Fclaude-code.html","Claude Code"," 等客户端直接接入",[45,3148,3149,3152],{},[21,3150,3151],{},"Webhook","：回调通知",[13,3154,3156],{"id":3155},"部署-10-分钟docker","部署 10 分钟（Docker）",[83,3158,3160],{"className":85,"code":3159,"language":87,"meta":88,"style":88},"# 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",[90,3161,3162,3167,3175,3182,3186,3191,3204,3208,3213,3220,3226,3230,3235,3247,3251],{"__ignoreMap":88},[93,3163,3164],{"class":95,"line":96},[93,3165,3166],{"class":2141},"# 克隆代码\n",[93,3168,3169,3171,3173],{"class":95,"line":110},[93,3170,100],{"class":99},[93,3172,104],{"class":103},[93,3174,107],{"class":103},[93,3176,3177,3179],{"class":95,"line":120},[93,3178,114],{"class":113},[93,3180,3181],{"class":103}," FastGPT\n",[93,3183,3184],{"class":95,"line":129},[93,3185,2288],{"emptyLinePlaceholder":596},[93,3187,3188],{"class":95,"line":1722},[93,3189,3190],{"class":2141},"# 切到最新稳定版（参考 GitHub releases）\n",[93,3192,3193,3195,3198,3201],{"class":95,"line":2285},[93,3194,100],{"class":99},[93,3196,3197],{"class":103}," switch",[93,3199,3200],{"class":113}," -c",[93,3202,3203],{"class":113}," 4.14.7.2\n",[93,3205,3206],{"class":95,"line":2291},[93,3207,2288],{"emptyLinePlaceholder":596},[93,3209,3210],{"class":95,"line":2297},[93,3211,3212],{"class":2141},"# 选向量库版本（个人 \u002F 小规模选 pg）\n",[93,3214,3215,3217],{"class":95,"line":2303},[93,3216,114],{"class":113},[93,3218,3219],{"class":103}," deploy\u002Fdocker\u002Fcn\n",[93,3221,3222,3224],{"class":95,"line":2308},[93,3223,123],{"class":99},[93,3225,126],{"class":103},[93,3227,3228],{"class":95,"line":2314},[93,3229,2288],{"emptyLinePlaceholder":596},[93,3231,3232],{"class":95,"line":2319},[93,3233,3234],{"class":2141},"# 启动\n",[93,3236,3237,3239,3241,3243,3245],{"class":95,"line":878},[93,3238,132],{"class":99},[93,3240,135],{"class":113},[93,3242,138],{"class":103},[93,3244,141],{"class":103},[93,3246,144],{"class":113},[93,3248,3249],{"class":95,"line":5},[93,3250,2288],{"emptyLinePlaceholder":596},[93,3252,3254],{"class":95,"line":3253},15,[93,3255,3256],{"class":2141},"# 访问 http:\u002F\u002F\u003Cip>:3000，默认账号 root \u002F 1234\n",[17,3258,3259],{},"最低配置：2C4G + 20GB 硬盘 + Docker 28+ + Docker Compose 2.20+。",[17,3261,3262],{},"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。",[13,3264,3266],{"id":3265},"云版-vs-自托管对比","云版 vs 自托管对比",[17,3268,3269,3274],{},[666,3270,3273],{"href":3271,"rel":3272},"https:\u002F\u002Ffastgpt.io\u002Fzh\u002Fprice",[670],"fastgpt.io 官方定价"," 公开数据：",[181,3276,3277,3301],{},[184,3278,3279],{},[187,3280,3281,3283,3285,3288,3291,3294,3296,3298],{},[190,3282,2021],{},[190,3284,2024],{},[190,3286,3287],{},"AI 积分",[190,3289,3290],{},"知识库索引",[190,3292,3293],{},"团队",[190,3295,1877],{},[190,3297,316],{},[190,3299,3300],{},"QPM",[200,3302,3303,3327,3351,3375],{},[187,3304,3305,3307,3309,3312,3315,3318,3321,3324],{},[205,3306,2037],{},[205,3308,1748],{},[205,3310,3311],{},"100",[205,3313,3314],{},"600",[205,3316,3317],{},"1",[205,3319,3320],{},"10",[205,3322,3323],{},"3",[205,3325,3326],{},"30",[187,3328,3329,3332,3335,3338,3341,3344,3346,3348],{},[205,3330,3331],{},"基础",[205,3333,3334],{},"¥99\u002F月",[205,3336,3337],{},"4000",[205,3339,3340],{},"6000",[205,3342,3343],{},"5",[205,3345,377],{},[205,3347,3326],{},[205,3349,3350],{},"300",[187,3352,3353,3356,3359,3362,3365,3367,3370,3372],{},[205,3354,3355],{},"高级",[205,3357,3358],{},"¥599\u002F月",[205,3360,3361],{},"25000",[205,3363,3364],{},"36000",[205,3366,377],{},[205,3368,3369],{},"200",[205,3371,3311],{},[205,3373,3374],{},"1500",[187,3376,3377,3380,3382,3385,3387,3389,3391,3393],{},[205,3378,3379],{},"定制",[205,3381,1758],{},[205,3383,3384],{},"弹性",[205,3386,3384],{},[205,3388,3384],{},[205,3390,3384],{},[205,3392,3384],{},[205,3394,3384],{},[17,3396,3397,3400,3401,3404,3405,3409],{},[21,3398,3399],{},"云版适合","：不想运维、量小、要快速上线\n",[21,3402,3403],{},"自托管适合","：量大（10 万+ 日问答）、数据敏感、要深度定制——按 ",[666,3406,3408],{"href":2882,"rel":3407},[670],"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\"",[13,3411,1280],{"id":1280},[181,3413,3414,3436],{},[184,3415,3416],{},[187,3417,3418,3420,3422,3426,3430,3433],{},[190,3419,1117],{},[190,3421,870],{},[190,3423,3424],{},[666,3425,1296],{"href":1295},[190,3427,3428],{},[666,3429,882],{"href":1825},[190,3431,3432],{},"RAGFlow",[190,3434,3435],{},"AnythingLLM",[200,3437,3438,3458,3474,3489,3506,3521,3538,3553],{},[187,3439,3440,3443,3446,3449,3452,3455],{},[205,3441,3442],{},"核心定位",[205,3444,3445],{},"知识库 QA",[205,3447,3448],{},"综合 LLMOps",[205,3450,3451],{},"Bot + 工作流",[205,3453,3454],{},"文档解析+RAG",[205,3456,3457],{},"桌面级 KB",[187,3459,3460,3462,3465,3467,3469,3471],{},[205,3461,1313],{},[205,3463,3464],{},"✅ Apache 2.0",[205,3466,3464],{},[205,3468,1147],{},[205,3470,3464],{},[205,3472,3473],{},"✅ MIT",[187,3475,3476,3478,3481,3483,3485,3487],{},[205,3477,874],{},[205,3479,3480],{},"★★★★★ docker",[205,3482,1363],{},[205,3484,1329],{},[205,3486,1360],{},[205,3488,1363],{},[187,3490,3491,3494,3497,3499,3501,3504],{},[205,3492,3493],{},"RAG 深度",[205,3495,3496],{},"★★★★★ 最细",[205,3498,1360],{},[205,3500,1366],{},[205,3502,3503],{},"★★★★★ 文档解析最强",[205,3505,1366],{},[187,3507,3508,3511,3513,3515,3517,3519],{},[205,3509,3510],{},"工作流",[205,3512,1360],{},[205,3514,1363],{},[205,3516,1360],{},[205,3518,1366],{},[205,3520,2453],{},[187,3522,3523,3526,3529,3531,3534,3536],{},[205,3524,3525],{},"上手",[205,3527,3528],{},"★★★☆☆ 需 docker",[205,3530,1360],{},[205,3532,3533],{},"★★★★★ 最简单",[205,3535,1366],{},[205,3537,1360],{},[187,3539,3540,3543,3545,3547,3549,3551],{},[205,3541,3542],{},"中文优化",[205,3544,1363],{},[205,3546,1360],{},[205,3548,1363],{},[205,3550,1360],{},[205,3552,1366],{},[187,3554,3555,3558,3561,3563,3565,3567],{},[205,3556,3557],{},"多平台发布",[205,3559,3560],{},"⚠️ API 为主",[205,3562,1360],{},[205,3564,1363],{},[205,3566,1336],{},[205,3568,1336],{},[17,3570,3571,1428,3573,971],{},[21,3572,1427],{},[666,3574,3576],{"href":2882,"rel":3575},[670],"南环 AI 评测",[42,3578,3579,3585,3592,3599,3605],{},[45,3580,3581,3584],{},[21,3582,3583],{},"核心需求是 RAG 精度"," → FastGPT",[45,3586,3587,1451,3590],{},[21,3588,3589],{},"需要丰富插件 + 复杂工作流 + 多平台发布",[666,3591,1296],{"href":1295},[45,3593,3594,1451,3597],{},[21,3595,3596],{},"零代码、快速发布到飞书 \u002F 微信",[666,3598,882],{"href":1825},[45,3600,3601,3604],{},[21,3602,3603],{},"文档解析（含 OCR \u002F 表格 \u002F 公式）是瓶颈"," → RAGFlow",[45,3606,3607,3610],{},[21,3608,3609],{},"桌面 \u002F 单机使用"," → AnythingLLM",[17,3612,3613,3616],{},[21,3614,3615],{},"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。",[13,3618,1471],{"id":1471},[42,3620,3621,3633,3642,3648,3658,3669,3675,3681],{},[45,3622,3623,285,3626,3629,3630,3632],{},[21,3624,3625],{},"docker-compose 镜像 tag 不一致",[666,3627,671],{"href":668,"rel":3628},[670]," 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 ",[90,3631,675],{}," 行为正确版本即可",[45,3634,3635,3638,3639,3641],{},[21,3636,3637],{},"3000 端口冲突","：默认占用 3000（主服务）\u002F 9000（S3 \u002F MinIO）\u002F 3005（MCP）；改 ",[90,3640,132],{}," 的 ports 映射端口",[45,3643,3644,3647],{},[21,3645,3646],{},"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",[45,3649,3650,3653,3654,3657],{},[21,3651,3652],{},"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 ",[666,3655,3408],{"href":2882,"rel":3656},[670],"：\"新手 \u002F 小规模 PgVector，中大规模 Milvus，企业 \u002F 国产 OceanBase\"",[45,3659,3660,285,3663,3665,3666],{},[21,3661,3662],{},"MinIO 默认密码",[90,3664,690],{},"，",[21,3667,3668],{},"部署到公网前必须改",[45,3670,3671,3674],{},[21,3672,3673],{},"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\"",[45,3676,3677,3680],{},[21,3678,3679],{},"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",[45,3682,3683,3686,3687,1504],{},[21,3684,3685],{},"云版 AI 积分会过期","：未用完不能跨月累积（按 ",[666,3688,3690],{"href":3271,"rel":3689},[670],"fastgpt.io 定价 FAQ",[13,3692,1548],{"id":1547},[17,3694,1551],{},[42,3696,3697,3700,3703,3706,3709,3712],{},[45,3698,3699],{},"企业内部知识库（员工手册 \u002F 制度 \u002F 流程）",[45,3701,3702],{},"产品 FAQ \u002F 用户手册问答",[45,3704,3705],{},"医疗 \u002F 法律 \u002F 金融垂直领域知识系统",[45,3707,3708],{},"数据严格不出网 + Apache 2.0 商用",[45,3710,3711],{},"有 docker 运维基础的技术团队",[45,3713,3714],{},"需要把 RAG 当后端服务的开发者（API 接入业务系统）",[17,3716,1571],{},[42,3718,3719,3724,3729,3732],{},[45,3720,3721,3722,1504],{},"完全非技术用户（去 ",[666,3723,882],{"href":1825},[45,3725,3726,3727,1504],{},"主要需求是工作流 + 插件集成（去 ",[666,3728,1296],{"href":1295},[45,3730,3731],{},"文档解析 \u002F OCR 是首要痛点（RAGFlow）",[45,3733,3734],{},"不想自己运维 + 量很小（FastGPT 云免费版起步即可）",[13,3736,794],{"id":794},[42,3738,3739,3748,3764,3778],{},[45,3740,1608,3741,1529,3743,3745,3746],{},[666,3742,1296],{"href":1295},[666,3744,882],{"href":1825}," \u002F RAGFlow \u002F AnythingLLM \u002F ",[666,3747,1620],{"href":1619},[45,3749,1623,3750,1529,3752,1529,3756,1529,3759,1529,3762],{},[666,3751,871],{"href":1630},[666,3753,3755],{"href":3754},"\u002Fwiki\u002Fembedding.html","Embedding",[666,3757,3758],{"href":3754},"Vector Database",[666,3760,3761],{"href":1630},"Reranker",[666,3763,1627],{"href":1626},[45,3765,1640,3766,1529,3768,1529,3770,1529,3772,1529,3776],{},[666,3767,1648],{"href":1647},[666,3769,1652],{"href":1651},[666,3771,1656],{"href":1655},[666,3773,3775],{"href":3774},"\u002Fmodels\u002Fkimi-k2.html","Kimi K2",[666,3777,1644],{"href":1643},[45,3779,1659,3780,1529,3782],{},[666,3781,1667],{"href":1666},[666,3783,1663],{"href":1662},[13,3785,1670],{"id":1670},[42,3787,3788,3794,3800,3806],{},[45,3789,2726,3790],{},[666,3791,3792],{"href":3792,"rel":3793},"https:\u002F\u002Ffastgpt.io",[670],[45,3795,2739,3796],{},[666,3797,3798],{"href":3798,"rel":3799},"https:\u002F\u002Fgithub.com\u002Flabring\u002FFastGPT",[670],[45,3801,3802,3803],{},"定价：",[666,3804,3271],{"href":3271,"rel":3805},[670],[45,3807,3808],{},"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",[17,3810,3811,3812,1704],{},"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 ",[666,3813,1703],{"href":1703},[828,3815,3816],{},"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":88,"searchDepth":120,"depth":120,"links":3818},[3819,3820,3828,3829,3830,3831,3832,3833,3834],{"id":890,"depth":110,"text":891},{"id":941,"depth":110,"text":941,"children":3821},[3822,3823,3824,3825,3826,3827],{"id":2890,"depth":120,"text":2891},{"id":2940,"depth":120,"text":2941},{"id":2990,"depth":120,"text":2991},{"id":3054,"depth":120,"text":3054},{"id":3060,"depth":120,"text":3060},{"id":3120,"depth":120,"text":3121},{"id":3155,"depth":110,"text":3156},{"id":3265,"depth":110,"text":3266},{"id":1280,"depth":110,"text":1280},{"id":1471,"depth":110,"text":1471},{"id":1547,"depth":110,"text":1548},{"id":794,"depth":110,"text":794},{"id":1670,"depth":110,"text":1670},"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。",[1728,1729],{},[1733,1734,1732,3839,3840,3841],"gpt-4o","claude-sonnet-4","kimi",[3843,3844,3845],"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",[2790,2791,2792],[3849,3853,3858,3862,3866],{"plan":3850,"price":2037,"limit":3851,"cn_pay":1750,"note":3852},"Self-host 开源","全功能 + 全数据本地","Apache 2.0 可商用",{"plan":3854,"price":3855,"limit":3856,"cn_pay":1750,"note":3857},"云免费版","¥0\u002F月","100 AI 积分 + 600 索引 + 3 知识库","试水",{"plan":3859,"price":3334,"limit":3860,"cn_pay":1755,"note":3861},"云基础版","4000 积分 + 6000 索引 + 50 Agent","中小团队 SaaS",{"plan":3863,"price":3358,"limit":3864,"cn_pay":1318,"note":3865},"云高级版","25000 积分 + 36000 索引 + 50 成员 + 200 Agent + 1500 QPM","企业级生产",{"plan":3867,"price":1758,"limit":3868,"cn_pay":1318,"note":3869},"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",[3872],"onboarding\u002Ffastgpt-getting-started",[3874],"fastgpt-deep-review",{"power":129,"ux":129,"price":1722,"cn_support":1722,"stability":129},{"title":870,"description":3835},[3878,3880,3882,3884,3886],{"title":3879,"url":3792},"FastGPT 官网",{"title":3881,"url":3798},"FastGPT GitHub",{"title":3883,"url":3271},"FastGPT 定价页",{"title":3885,"url":2882},"FastGPT 2025 测评（南环 AI）",{"title":3887,"url":668},"FastGPT 部署教程（腾讯云）","tools\u002Fagent\u002Fplatform\u002Ffastgpt",[3890,3891,3892,3893,3894],"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",[1791,2829,2830,2831,3897,3898,3899],"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",1783006599858]