[{"data":1,"prerenderedAt":3706},["ShallowReactive",2],{"header-counts":3,"footer-counts":6,"review-fastgpt-deep-review":9,"review-related-fastgpt-deep-review":744},{"tools":4,"reviews":5},70,14,{"tools":4,"reviews":5,"playbooks":7,"news":8},19,13,{"id":10,"title":11,"body":12,"cover":724,"description":725,"extension":726,"meta":727,"navigation":728,"path":729,"published":730,"relatedTools":731,"seo":735,"stem":736,"tags":737,"updated":730,"verdict":742,"__hash__":743},"review\u002Freview\u002Ffastgpt-deep-review.md","FastGPT 深度评测：国内企业知识库首选？和 Dify、Coze 到底怎么选",{"type":13,"value":14,"toc":713},"minimark",[15,19,32,35,49,53,60,70,80,86,92,96,107,141,151,155,161,164,227,238,245,281,285,288,294,305,311,317,321,328,462,472,475,619,624,655,659,665,674,677,709],[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",{},"企业内部知识库","——员工手册、产品文档、垂直领域 QA——并且数据必须留在自己机器上，",[24,29,30],{},"FastGPT 是 2026 年国内首选","。它把 RAG 链路的每一步都做成了可视化节点：问题预处理、混合检索、重排序、上下文组装、答案生成，一步步都能调、都能看。",[20,33,34],{},"但它不是\"给非技术用户用的\"。你至少要会 docker-compose、能看懂 Nginx 反代、能定位 postgres\u002Fmilvus 连接问题。如果你要的是\"零代码、拖拽出一个飞书机器人\"，去 Coze；如果你要的是\"复杂插件工作流 + 多平台发布\"，去 Dify。",[36,37,42],"div",{"className":38},[39,40,41],"card","p-5","my-4",[20,43,44,45,48],{},"搭配建议：",[24,46,47],{},"FastGPT 做知识库底座 + Coze\u002FDify 做前端 Bot 发布"," 是国内不少企业的实际选型，两个平台通过 API 打通即可。",[16,50,52],{"id":51},"fastgpt-真正在解决的问题","FastGPT 真正在解决的问题",[20,54,55,56,59],{},"社区讨论\"为什么用 FastGPT\"经常聚焦在\"27k star \u002F 开源 \u002F 免费\"，但这些只是表面。深一层看，FastGPT 在解决",[24,57,58],{},"国内做企业知识库这条路上的三堵墙","：",[20,61,62,65,66,69],{},[24,63,64],{},"第一堵墙：数据合规。"," 医疗、金融、法律、政务、大企业内部知识库——都不能把数据丢给 SaaS。Coze 云版会把数据发到火山引擎，Dify 云版走海外，都很难通过合规评审。FastGPT 支持一键 docker-compose 本地部署，代码 Apache 2.0，向量库、大模型都可以放在自己内网，",[24,67,68],{},"数据一步都不出内网","。",[20,71,72,75,76,79],{},[24,73,74],{},"第二堵墙：RAG 精度天花板。"," 通用平台的 RAG 是\"上传文档 → 拆块 → embedding → 检索 → 拼 prompt\"这一条固定链路。企业真实场景里，每一步都要可以调：切片策略、embedding 模型、混合检索比例、重排序模型、上下文长度、引用格式。FastGPT 把这五步都做成了可视化节点，是国内开源里 RAG ",[24,77,78],{},"调参深度","最高的一档。",[20,81,82,85],{},[24,83,84],{},"第三堵墙：成本可控。"," 云版 SaaS 定价按积分或调用量算，日均 10 万次问答的企业跑一年是六位数。FastGPT 自托管 + 开源模型（比如 DeepSeek\u002FQwen 家族）可以把成本压到硬件采购 + 电费，长期账更划算。",[20,87,88,89,69],{},"三堵墙加在一起，FastGPT 的定位就清楚了：",[24,90,91],{},"它不是要跟 Coze\u002FDify 比\"谁功能更全\"，是给\"企业级、私有化、RAG 精度敏感\"这个细分场景一个可以落地的答案",[16,93,95],{"id":94},"rag-链路细到什么程度","RAG 链路：细到什么程度",[20,97,98,99,106],{},"按 ",[100,101,105],"a",{"href":102,"rel":103},"https:\u002F\u002Fwww.nanhuantech.com\u002Fzh\u002Fai-reviews\u002Ffastgpt-2025-review",[104],"nofollow","南环 AI 2026-05 评测"," 整理的 FastGPT RAG 链路：",[108,109,110,117,123,129,135],"ol",{},[111,112,113,116],"li",{},[24,114,115],{},"问题预处理","：改写 \u002F 扩展 \u002F 错词纠正，提升召回率",[111,118,119,122],{},[24,120,121],{},"检索策略","：语义检索 \u002F 关键词 BM25 \u002F 混合检索三选一，可调阈值",[111,124,125,128],{},[24,126,127],{},"重排序 Rerank","：初步检索结果二次排序，用 BGE reranker 或 Cohere Rerank",[111,130,131,134],{},[24,132,133],{},"上下文组装","：最优 chunk + 问题拼 prompt，包括引用格式",[111,136,137,140],{},[24,138,139],{},"答案生成","：LLM 基于检索结果回答 + 引用标注",[20,142,143,144,147,148,69],{},"每一步都是",[24,145,146],{},"可视化节点","，能单独测试、能开关、能换模型。举个真实场景：一份 300 页的产品手册，用默认切片策略 QA 精度只有 60%，把切片改成\"按章节 + overlap 100\"、加一层 BGE reranker 后能提到 85%——",[24,149,150],{},"这种可调空间就是 FastGPT 相对 Coze\u002FDify 最大的差异化",[16,152,154],{"id":153},"部署-10-分钟真的能-10-分钟吗","部署 10 分钟：真的能 10 分钟吗",[20,156,157,158,69],{},"答：",[24,159,160],{},"如果你会 docker，10 分钟。不会 docker，一整天",[20,162,163],{},"标准流程（选 pgvector 版本，小规模）：",[165,166,171],"pre",{"className":167,"code":168,"language":169,"meta":170,"style":170},"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","",[172,173,174,190,200,209],"code",{"__ignoreMap":170},[175,176,179,183,187],"span",{"class":177,"line":178},"line",1,[175,180,182],{"class":181},"sScJk","git",[175,184,186],{"class":185},"sZZnC"," clone",[175,188,189],{"class":185}," https:\u002F\u002Fgithub.com\u002Flabring\u002FFastGPT.git\n",[175,191,193,197],{"class":177,"line":192},2,[175,194,196],{"class":195},"sj4cs","cd",[175,198,199],{"class":185}," FastGPT\u002Fdeploy\u002Fdocker\u002Fcn\n",[175,201,203,206],{"class":177,"line":202},3,[175,204,205],{"class":181},"wget",[175,207,208],{"class":185}," https:\u002F\u002Fdoc.fastgpt.cn\u002Fdeploy\u002Fconfig\u002Fconfig.json\n",[175,210,212,215,218,221,224],{"class":177,"line":211},4,[175,213,214],{"class":181},"docker-compose",[175,216,217],{"class":195}," -f",[175,219,220],{"class":185}," docker-compose.pg.yml",[175,222,223],{"class":185}," up",[175,225,226],{"class":195}," -d\n",[20,228,229,230,233,234,237],{},"启动后访问 ",[172,231,232],{},"http:\u002F\u002F\u003Cip>:3000","，默认 ",[172,235,236],{},"root \u002F 1234","。最低配置 2C4G + 20GB 硬盘 + Docker 28+ + Docker Compose 2.20+。",[20,239,240,241,244],{},"看起来简单，但",[24,242,243],{},"新手常见坑","是：",[246,247,248,254,266,272],"ul",{},[111,249,250,253],{},[24,251,252],{},"端口冲突","：3000（主服务）\u002F 9000（MinIO）\u002F 3005（MCP）都要空",[111,255,256,259,260,265],{},[24,257,258],{},"镜像 tag 对不上","：编排文件里的 image tag 不一定是最新 release，",[100,261,264],{"href":262,"rel":263},"https:\u002F\u002Fcloud.tencent.com\u002Fdeveloper\u002Farticle\u002F2632669",[104],"腾讯云 2026-03 教程"," 提到过这个坑",[111,267,268,271],{},[24,269,270],{},"忘配 embedding 模型","：只配了 GPT-4 对话模型没配 embedding，知识库根本无法索引",[111,273,274,59,277,280],{},[24,275,276],{},"MinIO 默认密码没改",[172,278,279],{},"minioadmin\u002Fminioadmin","，公网部署前必须改",[16,282,284],{"id":283},"能力天花板哪些场景不推荐用-fastgpt","能力天花板：哪些场景不推荐用 FastGPT",[20,286,287],{},"诚实地说，FastGPT 有几个明显的弱项：",[20,289,290,293],{},[24,291,292],{},"Bot 多平台发布","：Coze 一键就能发到飞书、微信、抖音，FastGPT 只能通过 API 自己接。要在企业内网起一个飞书 Bot，还得自己写中间层。",[20,295,296,299,300,304],{},[24,297,298],{},"工作流复杂度","：新版 v4.14 起有工作流节点，但整体的可视化编排、插件生态、外部 HTTP 集成，仍然弱于 ",[100,301,303],{"href":302},"\u002Fagent\u002Fplatform\u002Fdify.html","Dify","。做\"分类路由 + 多 Agent 协同 + 复杂业务链路\"这种活，Dify 更顺手。",[20,306,307,310],{},[24,308,309],{},"面向非技术用户","：没有 Coze 那种\"完全零代码\"的用户友好度。管理员用起来舒服，业务用户上手门槛偏高。",[20,312,313,316],{},[24,314,315],{},"文档解析深度","：PDF\u002FWord\u002FMarkdown 都支持，但复杂表格、公式、扫描件的 OCR 是弱项，做医疗、法律这种\"极端复杂文档\"场景应该同时评估 RAGFlow。",[16,318,320],{"id":319},"云版-vs-自托管","云版 vs 自托管",[20,322,323,59],{},[100,324,327],{"href":325,"rel":326},"https:\u002F\u002Ffastgpt.io\u002Fzh\u002Fprice",[104],"fastgpt.io 官方定价",[329,330,331,362],"table",{},[332,333,334],"thead",{},[335,336,337,341,344,347,350,353,356,359],"tr",{},[338,339,340],"th",{},"套餐",[338,342,343],{},"价格",[338,345,346],{},"AI 积分",[338,348,349],{},"索引",[338,351,352],{},"团队",[338,354,355],{},"Agent",[338,357,358],{},"知识库",[338,360,361],{},"QPM",[363,364,365,392,417,441],"tbody",{},[335,366,367,371,374,377,380,383,386,389],{},[368,369,370],"td",{},"免费",[368,372,373],{},"¥0",[368,375,376],{},"100",[368,378,379],{},"600",[368,381,382],{},"1",[368,384,385],{},"10",[368,387,388],{},"3",[368,390,391],{},"30",[335,393,394,397,400,403,406,409,412,414],{},[368,395,396],{},"基础",[368,398,399],{},"¥99\u002F月",[368,401,402],{},"4000",[368,404,405],{},"6000",[368,407,408],{},"5",[368,410,411],{},"50",[368,413,391],{},[368,415,416],{},"300",[335,418,419,422,425,428,431,433,436,438],{},[368,420,421],{},"高级",[368,423,424],{},"¥599\u002F月",[368,426,427],{},"25000",[368,429,430],{},"36000",[368,432,411],{},[368,434,435],{},"200",[368,437,376],{},[368,439,440],{},"1500",[335,442,443,446,449,452,454,456,458,460],{},[368,444,445],{},"定制",[368,447,448],{},"议价",[368,450,451],{},"弹性",[368,453,451],{},[368,455,451],{},[368,457,451],{},[368,459,451],{},[368,461,451],{},[20,463,464,467,468,471],{},[24,465,466],{},"选云版","：不想运维、量小、要快速上线、能接受数据在 fastgpt.io。\n",[24,469,470],{},"选自托管","：数据敏感、量大、要深度定制。日均 10 万次问答规模，自托管的年度总成本通常只有云版的 20%–30%。",[16,473,474],{"id":474},"与同类怎么选",[329,476,477,500],{},[332,478,479],{},[335,480,481,484,487,491,497],{},[338,482,483],{},"维度",[338,485,486],{},"FastGPT",[338,488,489],{},[100,490,303],{"href":302},[338,492,493],{},[100,494,496],{"href":495},"\u002Fagent\u002Fplatform\u002Fcoze.html","Coze",[338,498,499],{},"RAGFlow",[363,501,502,519,534,550,564,577,591,604],{},[335,503,504,507,510,513,516],{},[368,505,506],{},"核心定位",[368,508,509],{},"知识库 QA",[368,511,512],{},"综合 LLMOps",[368,514,515],{},"Bot + 工作流",[368,517,518],{},"文档解析 + RAG",[335,520,521,524,527,529,532],{},[368,522,523],{},"开源许可",[368,525,526],{},"Apache 2.0",[368,528,526],{},[368,530,531],{},"❌",[368,533,526],{},[335,535,536,539,542,544,547],{},[368,537,538],{},"私有部署",[368,540,541],{},"★★★★★",[368,543,541],{},[368,545,546],{},"⚠️ 企业版",[368,548,549],{},"★★★★☆",[335,551,552,555,557,559,562],{},[368,553,554],{},"RAG 深度",[368,556,541],{},[368,558,549],{},[368,560,561],{},"★★★☆☆",[368,563,541],{},[335,565,566,569,571,573,575],{},[368,567,568],{},"工作流",[368,570,549],{},[368,572,541],{},[368,574,549],{},[368,576,561],{},[335,578,579,582,585,587,589],{},[368,580,581],{},"上手门槛",[368,583,584],{},"⚠️ 需 docker",[368,586,549],{},[368,588,541],{},[368,590,561],{},[335,592,593,596,598,600,602],{},[368,594,595],{},"中文优化",[368,597,541],{},[368,599,549],{},[368,601,541],{},[368,603,549],{},[335,605,606,609,612,614,616],{},[368,607,608],{},"Bot 发布",[368,610,611],{},"⚠️ API 为主",[368,613,549],{},[368,615,541],{},[368,617,618],{},"⚠️",[20,620,621,59],{},[24,622,623],{},"怎么选",[246,625,626,633,642,649],{},[111,627,628,629,632],{},"核心需求是",[24,630,631],{},"私有化 + RAG 精度"," → FastGPT",[111,634,635,636,639,640],{},"需要",[24,637,638],{},"丰富插件 + 复杂工作流 + 多平台发布"," → ",[100,641,303],{"href":302},[111,643,644,639,647],{},[24,645,646],{},"零代码、快速发布到飞书\u002F微信\u002F抖音",[100,648,496],{"href":495},[111,650,651,654],{},[24,652,653],{},"文档解析（OCR \u002F 表格 \u002F 公式）是瓶颈"," → RAGFlow",[16,656,658],{"id":657},"aiho-推荐结论","AIHO 推荐结论",[20,660,22,661,664],{},[24,662,663],{},"面向企业员工\u002F客户的知识库问答","，并且负责这个项目的人能搞定 docker 和 postgres——先用 FastGPT 自托管做 POC，用 2 周时间跑通一个真实知识库、调好 RAG 参数、看召回和答案质量。这一步的产出足够支持后续的选型决策。",[20,666,667,668,670,671,673],{},"如果 POC 阶段发现 Bot 多平台分发是刚需，或者要做的其实是\"通用 AI 助手\"而不是\"知识库 QA\"，再考虑把发布层交给 ",[100,669,496],{"href":495}," 或 ",[100,672,303],{"href":302},"、FastGPT 只做知识库后端。国内不少团队最终都走到了这个组合。",[16,675,676],{"id":676},"相关阅读",[246,678,679,685,691,697,703],{},[111,680,681],{},[100,682,684],{"href":683},"\u002Fagent\u002Fplatform\u002Ffastgpt.html","FastGPT 工具卡：国内企业知识库私有化 RAG 平台",[111,686,687],{},[100,688,690],{"href":689},"\u002Fcompare\u002Ffastgpt-vs-dify.html","FastGPT vs Dify：国内企业 RAG 与 Agent 平台怎么选",[111,692,693],{},[100,694,696],{"href":695},"\u002Fplaybook\u002Fonboarding\u002Ffastgpt-getting-started.html","FastGPT 部署与知识库搭建实战",[111,698,699],{},[100,700,702],{"href":701},"\u002Freview\u002Fcoze-vs-dify.html","Coze vs Dify：国内 Agent 平台对比",[111,704,705],{},[100,706,708],{"href":707},"\u002Freview\u002Fllm-gateway-comparison.html","LLM Gateway 与 API 网关横评",[710,711,712],"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":170,"searchDepth":202,"depth":202,"links":714},[715,716,717,718,719,720,721,722,723],{"id":18,"depth":192,"text":18},{"id":51,"depth":192,"text":52},{"id":94,"depth":192,"text":95},{"id":153,"depth":192,"text":154},{"id":283,"depth":192,"text":284},{"id":319,"depth":192,"text":320},{"id":474,"depth":192,"text":474},{"id":657,"depth":192,"text":658},{"id":676,"depth":192,"text":676},"\u002Fog\u002Freview\u002Ffastgpt.png","FastGPT 真实评测：labring 团队开源的 LLM 知识库 RAG 平台，27k+ GitHub star。本文写它真正解决的问题、RAG 精度是怎么打出来的、和 Dify\u002FCoze\u002FRAGFlow 的能力差距，以及不推荐用 FastGPT 的 4 类场景。AIHO 编辑部基于官方文档与第三方评测整理。","md",{},true,"\u002Freview\u002Ffastgpt-deep-review","2026-07-02",[732,733,734],"agent\u002Fplatform\u002Ffastgpt","agent\u002Fplatform\u002Fdify","agent\u002Fplatform\u002Fcoze",{"title":11,"description":725},"review\u002Ffastgpt-deep-review",[486,738,358,739,740,741],"RAG","开源","Agent 平台","深度评测","国内企业知识库私有化的默认答案：Apache 2.0 可商用、docker-compose 十分钟起、RAG 链路每一步都可视化调参。代价是需要有 docker 运维基础，Bot 多平台发布和插件生态弱于 Dify\u002FCoze。","o_yGhq9WwTgxKVQSUdmVn-7KMdUv0floK__F8k07KFQ",[745,1648,2679],{"id":746,"title":496,"alternatives":747,"api_compatible":749,"body":750,"category":1571,"chinese_friendly":1572,"cover":1573,"description":1574,"domestic":1575,"extension":726,"faq":1576,"free":1575,"github":1576,"languages":1577,"meta":1580,"models":1581,"navigation":728,"notSuitable":1587,"opensource":1575,"path":1591,"pillar":1592,"platforms":1593,"priceTable":1595,"pricing":1614,"published":1615,"relatedPlaybooks":1576,"relatedReviews":1576,"score":1616,"self_host":1575,"seo":1617,"seoTitle":1618,"slug":734,"sources":1619,"stem":1630,"suitable":1631,"tagline":1637,"tags":1638,"updated":1645,"verdict":1646,"website":770,"__hash__":1647},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Fcoze.md",[733,732,748],"agent\u002Fplatform\u002Fyuanqi",[],{"type":13,"value":751,"toc":1555},[752,756,793,799,802,807,810,824,833,883,892,896,899,931,939,943,946,966,970,1050,1059,1062,1065,1085,1094,1098,1134,1137,1139,1274,1289,1319,1322,1395,1399,1402,1419,1422,1452,1454,1518,1521,1548],[16,753,755],{"id":754},"tldr","TL;DR",[36,757,759,780],{"className":758},[39,40,41],[20,760,761,764,765,779],{},[24,762,763],{},"一句话："," 字节跳动出品的低代码 Agent 平台，",[24,766,767,768,773,774],{},"国内版 ",[100,769,772],{"href":770,"rel":771},"https:\u002F\u002Fwww.coze.cn",[104],"coze.cn","（中文叫\"扣子\"）+ 国际版 ",[100,775,778],{"href":776,"rel":777},"https:\u002F\u002Fwww.coze.com",[104],"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。",[20,781,782,783,786,787,792],{},"最大价值是 ",[24,784,785],{},"零代码、最快上手","——文档详细 + 中文社区活跃 + 模板多，根据 ",[100,788,791],{"href":789,"rel":790},"https:\u002F\u002Fwww.cnblogs.com\u002Fuulucias\u002Fp\u002F19449008",[104],"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\"。",[794,795,796],"blockquote",{},[20,797,798],{},"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。",[16,800,801],{"id":801},"核心特性",[803,804,806],"h3",{"id":805},"可视化工作流最大卖点","可视化工作流（最大卖点）",[20,808,809],{},"Coze 把 Agent 拆成两层：",[246,811,812,818],{},[111,813,814,817],{},[24,815,816],{},"Bot \u002F 智能体","：对话式 AI，配置 prompt + 知识库 + 插件",[111,819,820,823],{},[24,821,822],{},"工作流（Workflow）","：DAG 节点编排，可被 Bot 调用，也可独立部署",[20,825,826,827,832],{},"工作流节点类型（基于 ",[100,828,831],{"href":829,"rel":830},"https:\u002F\u002Fdeveloper.volcengine.com\u002Farticles\u002F7530117616687480851",[104],"火山引擎社区 2025 实战","）：",[246,834,835,841,847,853,859,865,871,877],{},[111,836,837,840],{},[24,838,839],{},"开始 \u002F 结束节点","：输入输出",[111,842,843,846],{},[24,844,845],{},"大模型节点","：调豆包 \u002F GPT \u002F Claude 任一模型",[111,848,849,852],{},[24,850,851],{},"代码节点","：内嵌 Python \u002F JavaScript（飞书插件常需要数据格式转换）",[111,854,855,858],{},[24,856,857],{},"循环节点","：批量处理多条数据",[111,860,861,864],{},[24,862,863],{},"条件节点","：分支判断",[111,866,867,870],{},[24,868,869],{},"插件节点","：调 Coze 插件市场的工具",[111,872,873,876],{},[24,874,875],{},"知识库节点","：RAG 检索",[111,878,879,882],{},[24,880,881],{},"HTTP 节点","：调外部 API",[20,884,885,886,891],{},"典型用例（参考 ",[100,887,890],{"href":888,"rel":889},"https:\u002F\u002Fwww.toutiao.com\u002Farticle\u002F7469986334686315017",[104],"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。",[803,893,895],{"id":894},"插件市场200-官方-海量第三方","插件市场（200+ 官方 + 海量第三方）",[20,897,898],{},"Coze 的\"插件\"是封装好的 API 工具，比如：",[246,900,901,907,913,919,925],{},[111,902,903,906],{},[24,904,905],{},"飞书多维表格","：增删改查记录（国内 toB 场景的杀手锏）",[111,908,909,912],{},[24,910,911],{},"图像生成","：调豆包 \u002F SD \u002F DALL-E",[111,914,915,918],{},[24,916,917],{},"联网搜索","：实时网页检索",[111,920,921,924],{},[24,922,923],{},"代码执行","：在线运行 Python",[111,926,927,930],{},[24,928,929],{},"第三方 SaaS","：微博、抖音、bilibili、Notion……",[20,932,933,938],{},[100,934,937],{"href":935,"rel":936},"https:\u002F\u002Fkouziai.github.io\u002F",[104],"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。",[803,940,942],{"id":941},"bot-商店-多平台一键发布","Bot 商店 + 多平台一键发布",[20,944,945],{},"发布渠道：",[246,947,948,951,954,957,960,963],{},[111,949,950],{},"飞书机器人（一键绑）",[111,952,953],{},"抖音 \u002F 头条号",[111,955,956],{},"微信小程序 \u002F 公众号（部分需企业认证）",[111,958,959],{},"自定义网页嵌入",[111,961,962],{},"API 接口（提供 OpenAPI 风格 REST 调用）",[111,964,965],{},"Discord（国际版）",[803,967,969],{"id":968},"国内版-vs-国际版","国内版 vs 国际版",[329,971,972,984],{},[332,973,974],{},[335,975,976,978,981],{},[338,977,483],{},[338,979,980],{},"扣子（coze.cn）",[338,982,983],{},"Coze（coze.com）",[363,985,986,997,1007,1017,1028,1039],{},[335,987,988,991,994],{},[368,989,990],{},"主力模型",[368,992,993],{},"豆包 Pro \u002F DeepSeek \u002F Qwen \u002F Kimi",[368,995,996],{},"OpenAI \u002F Claude \u002F Gemini \u002F Cohere",[335,998,999,1002,1005],{},[368,1000,1001],{},"飞书 \u002F 抖音 \u002F 微信集成",[368,1003,1004],{},"✅ 原生",[368,1006,531],{},[335,1008,1009,1012,1015],{},[368,1010,1011],{},"Discord \u002F Slack 集成",[368,1013,1014],{},"⚠️ 有限",[368,1016,1004],{},[335,1018,1019,1022,1025],{},[368,1020,1021],{},"数据存储位置",[368,1023,1024],{},"国内",[368,1026,1027],{},"海外",[335,1029,1030,1033,1036],{},[368,1031,1032],{},"支付",[368,1034,1035],{},"微信 \u002F 支付宝",[368,1037,1038],{},"海外信用卡",[335,1040,1041,1044,1047],{},[368,1042,1043],{},"内容合规",[368,1045,1046],{},"严格审核",[368,1048,1049],{},"宽松",[20,1051,1052,1055,1056,69],{},[24,1053,1054],{},"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 ",[24,1057,1058],{},"不互通",[16,1060,1061],{"id":1061},"价格与运行成本",[20,1063,1064],{},"国内版（扣子）：",[246,1066,1067,1073,1079],{},[111,1068,1069,1072],{},[24,1070,1071],{},"免费版","：免费模型有日额度（豆包 lite 等），适合个人玩 \u002F Demo",[111,1074,1075,1078],{},[24,1076,1077],{},"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",[111,1080,1081,1084],{},[24,1082,1083],{},"企业版","：议价，含 VPC、私有化（限定场景）、SLA",[20,1086,1087,1088,1093],{},"国际版（coze.com）的定价模式据 ",[100,1089,1092],{"href":1090,"rel":1091},"https:\u002F\u002Fdocs.coze.com\u002F",[104],"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。",[16,1095,1097],{"id":1096},"上手-10-分钟","上手 10 分钟",[108,1099,1100,1113,1116,1119,1122,1125,1128,1131],{},[111,1101,1102,1103,1107,1108,1112],{},"打开 ",[100,1104,1106],{"href":770,"rel":1105},[104],"www.coze.cn","（国内）或 ",[100,1109,1111],{"href":776,"rel":1110},[104],"www.coze.com","（国际），用飞书 \u002F 抖音账号 \u002F Google 账号登录",[111,1114,1115],{},"左侧\"工作空间\" → \"+创建 Bot\"，起个名字",[111,1117,1118],{},"选模型（国内推荐豆包 Pro，国际推 Claude Sonnet 4）",[111,1120,1121],{},"写 Bot 角色 prompt",[111,1123,1124],{},"可选：上传 PDF \u002F 文档建知识库",[111,1126,1127],{},"测试一下对话效果",[111,1129,1130],{},"右上\"发布\" → 选渠道（飞书 \u002F 抖音 \u002F API \u002F Web）",[111,1132,1133],{},"拿到调用 URL \u002F 飞书机器人 webhook",[20,1135,1136],{},"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。",[16,1138,474],{"id":474},[329,1140,1141,1163],{},[332,1142,1143],{},[335,1144,1145,1147,1149,1153,1157],{},[338,1146,483],{},[338,1148,496],{},[338,1150,1151],{},[100,1152,303],{"href":302},[338,1154,1155],{},[100,1156,486],{"href":683},[338,1158,1159],{},[100,1160,1162],{"href":1161},"\u002Fagent\u002Fplatform\u002Fyuanqi.html","元器 yuanqi",[363,1164,1165,1178,1191,1207,1220,1233,1247,1261],{},[335,1166,1167,1169,1171,1174,1176],{},[368,1168,739],{},[368,1170,531],{},[368,1172,1173],{},"✅",[368,1175,1173],{},[368,1177,531],{},[335,1179,1180,1182,1185,1187,1189],{},[368,1181,538],{},[368,1183,1184],{},"⚠️ 仅企业版",[368,1186,1173],{},[368,1188,1173],{},[368,1190,618],{},[335,1192,1193,1196,1199,1202,1204],{},[368,1194,1195],{},"上手难度",[368,1197,1198],{},"★ 最简单",[368,1200,1201],{},"★★★",[368,1203,1201],{},[368,1205,1206],{},"★★",[335,1208,1209,1212,1214,1216,1218],{},[368,1210,1211],{},"工作流编排",[368,1213,549],{},[368,1215,541],{},[368,1217,561],{},[368,1219,561],{},[335,1221,1222,1225,1227,1229,1231],{},[368,1223,1224],{},"RAG 精度",[368,1226,561],{},[368,1228,549],{},[368,1230,541],{},[368,1232,561],{},[335,1234,1235,1238,1240,1242,1244],{},[368,1236,1237],{},"字节生态",[368,1239,541],{},[368,1241,531],{},[368,1243,531],{},[368,1245,1246],{},"❌（腾讯系）",[335,1248,1249,1252,1255,1257,1259],{},[368,1250,1251],{},"插件市场",[368,1253,1254],{},"★★★★★ 200+",[368,1256,561],{},[368,1258,561],{},[368,1260,561],{},[335,1262,1263,1266,1268,1270,1272],{},[368,1264,1265],{},"中文社区",[368,1267,541],{},[368,1269,549],{},[368,1271,549],{},[368,1273,549],{},[20,1275,1276,1278,1279,1284,1285,832],{},[24,1277,623],{},"（综合 ",[100,1280,1283],{"href":1281,"rel":1282},"https:\u002F\u002Fwww.besthub.dev\u002Farticles\u002Fcoze-vs-dify-vs-fastgpt-which-ai-agent-platform-fits-your-needs-fa59cf97b798",[104],"BestHub 2025-07"," 和 ",[100,1286,1288],{"href":789,"rel":1287},[104],"博客园 2026-01",[246,1290,1291,1297,1304,1311],{},[111,1292,1293,1296],{},[24,1294,1295],{},"快速验证 \u002F 不懂代码 \u002F 1-2 天出原型"," → Coze",[111,1298,1299,639,1302],{},[24,1300,1301],{},"数据安全要求高 \u002F 复杂业务流程 \u002F 有技术团队",[100,1303,303],{"href":302},[111,1305,1306,639,1309],{},[24,1307,1308],{},"核心场景就是企业知识库 QA",[100,1310,486],{"href":683},[111,1312,1313,639,1316],{},[24,1314,1315],{},"QQ \u002F 微信生态 + 腾讯系",[100,1317,1318],{"href":1161},"元器",[16,1320,1321],{"id":1321},"避坑清单",[246,1323,1324,1334,1340,1355,1365,1371,1383,1389],{},[111,1325,1326,1329,1330,1333],{},[24,1327,1328],{},"国内版 vs 国际版的\"双账号陷阱\"","：扣子（coze.cn）和 Coze（coze.com）是",[24,1331,1332],{},"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",[111,1335,1336,1339],{},[24,1337,1338],{},"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",[111,1341,1342,1345,1346,1349,1350,1354],{},[24,1343,1344],{},"飞书多维表格插件数据格式坑","：写入多维表格需要 ",[172,1347,1348],{},"Array\u003CObject>"," 格式，代码节点要做转换（参考 ",[100,1351,1353],{"href":829,"rel":1352},[104],"火山引擎 2025 教程","）",[111,1356,1357,1360,1361,1364],{},[24,1358,1359],{},"工作流读取飞书表格默认 20 条","：要改 ",[172,1362,1363],{},"page_size","，最大 500 条；超过 500 要分页或循环",[111,1366,1367,1370],{},[24,1368,1369],{},"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",[111,1372,1373,1376,1377,1379,1380,1382],{},[24,1374,1375],{},"企业版\"私有化\"是有限的","：完全数据不出网仍建议 ",[100,1378,303],{"href":302}," \u002F ",[100,1381,486],{"href":683}," 自托管",[111,1384,1385,1388],{},[24,1386,1387],{},"国际版接 Claude \u002F GPT 需要 BYOK","：自己绑海外信用卡，平台不代付",[111,1390,1391,1394],{},[24,1392,1393],{},"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",[16,1396,1398],{"id":1397},"适合-不适合","适合 \u002F 不适合",[20,1400,1401],{},"✅ 适合：",[246,1403,1404,1407,1410,1413,1416],{},[111,1405,1406],{},"产品 \u002F 运营 \u002F 非技术人员快速做 Bot",[111,1408,1409],{},"在飞书 \u002F 抖音 \u002F 头条生态内做集成",[111,1411,1412],{},"个人副业（小红书账号批量内容生成等）",[111,1414,1415],{},"中小企业客服 Bot（3 天上线）",[111,1417,1418],{},"想用豆包 \u002F DeepSeek 国产模型的人",[20,1420,1421],{},"❌ 不适合：",[246,1423,1424,1427,1430,1435,1442],{},[111,1425,1426],{},"金融 \u002F 政府 \u002F 医疗（数据敏感，需自托管）",[111,1428,1429],{},"复杂业务系统深度集成（自由度不够）",[111,1431,1432,1433,1354],{},"反感字节生态（去 ",[100,1434,303],{"href":302},[111,1436,1437,1438,1379,1440,1354],{},"希望开源 \u002F 完全自主可控（去 ",[100,1439,303],{"href":302},[100,1441,486],{"href":683},[111,1443,1444,1445,1379,1447,1451],{},"海外 toB SaaS 产品后端（",[100,1446,303],{"href":302},[100,1448,1450],{"href":1449},"\u002Fagent\u002Fplatform\u002Flangflow.html","Langflow"," 更合适）",[16,1453,676],{"id":676},[246,1455,1456,1471,1488,1507],{},[111,1457,1458,1459,1379,1461,1379,1463,1379,1465,1379,1467],{},"同类对比：",[100,1460,303],{"href":302},[100,1462,486],{"href":683},[100,1464,1318],{"href":1161},[100,1466,1450],{"href":1449},[100,1468,1470],{"href":1469},"\u002Fagent\u002Fplatform\u002Fn8n.html","n8n",[111,1472,1473,1474,1379,1478,1379,1481,1379,1485],{},"概念：",[100,1475,1477],{"href":1476},"\u002Fwiki\u002Fai-agent.html","AI Agent",[100,1479,738],{"href":1480},"\u002Fwiki\u002Frag.html",[100,1482,1484],{"href":1483},"\u002Fwiki\u002Ffunction-calling.html","Function Calling",[100,1486,1487],{"href":1476},"Multi-Agent",[111,1489,1490,1491,1379,1495,1379,1499,1379,1503],{},"模型：",[100,1492,1494],{"href":1493},"\u002Fmodels\u002Fdoubao-1-5-pro.html","豆包 Doubao",[100,1496,1498],{"href":1497},"\u002Fmodels\u002Fdeepseek-v3.html","DeepSeek-V3",[100,1500,1502],{"href":1501},"\u002Fmodels\u002Fqwen-3.html","Qwen3",[100,1504,1506],{"href":1505},"\u002Fmodels\u002Fglm-5.2.html","GLM-5.2",[111,1508,1509,1510,1379,1514],{},"进阶：",[100,1511,1513],{"href":1512},"\u002Fwiki\u002Fprompt-engineering.html","Prompt Engineering",[100,1515,1517],{"href":1516},"\u002Fwiki\u002Fcontext-engineering.html","Context Engineering",[16,1519,1520],{"id":1520},"来源",[246,1522,1523,1529,1535,1542,1545],{},[111,1524,1525,1526],{},"国内版：",[100,1527,770],{"href":770,"rel":1528},[104],[111,1530,1531,1532],{},"国际版：",[100,1533,776],{"href":776,"rel":1534},[104],[111,1536,1537,1538],{},"官方文档：",[100,1539,1540],{"href":1540,"rel":1541},"https:\u002F\u002Fdocs.coze.com",[104],[111,1543,1544],{},"第三方选型评测：cnblogs.com \u002F besthub.dev \u002F aibotgo.net",[111,1546,1547],{},"实战教程：火山引擎社区、今日头条 涛哥讲AI",[20,1549,1550,1551,1554],{},"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 ",[100,1552,1553],{"href":1553},"\u002Fsubmit"," 反馈。",{"title":170,"searchDepth":202,"depth":202,"links":1556},[1557,1558,1564,1565,1566,1567,1568,1569,1570],{"id":754,"depth":192,"text":755},{"id":801,"depth":192,"text":801,"children":1559},[1560,1561,1562,1563],{"id":805,"depth":202,"text":806},{"id":894,"depth":202,"text":895},{"id":941,"depth":202,"text":942},{"id":968,"depth":202,"text":969},{"id":1061,"depth":192,"text":1061},{"id":1096,"depth":192,"text":1097},{"id":474,"depth":192,"text":474},{"id":1321,"depth":192,"text":1321},{"id":1397,"depth":192,"text":1398},{"id":676,"depth":192,"text":676},{"id":1520,"depth":192,"text":1520},"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,[1578,1579],"zh","en",{},[1582,1583,1584,1585,1586],"doubao-pro","deepseek-v3","qwen-max","gpt-4o (国际版)","claude (国际版)",[1588,1589,1590],"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",[1594],"web",[1596,1601,1606,1609],{"plan":1597,"price":373,"limit":1598,"cn_pay":1599,"note":1600},"免费版（扣子）","免费模型限额 + 基础功能","—","个人试水 \u002F 1-2 天 MVP",{"plan":1077,"price":1602,"limit":1603,"cn_pay":1604,"note":1605},"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":1083,"price":448,"limit":1607,"cn_pay":1173,"note":1608},"VPC、合规、SLA、私有化（部分）","B 端落地",{"plan":1610,"price":1611,"cn_pay":1612,"note":1613,"limit":1599},"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":211,"ux":1572,"price":211,"cn_support":1572,"stability":211},{"title":496,"description":1574},"Coze 扣子评测 2026：字节 AI Agent 平台，对比 Dify",[1620,1622,1624,1626,1628],{"title":1621,"url":770},"Coze 国内版（扣子）",{"title":1623,"url":776},"Coze 国际版",{"title":1625,"url":1540},"Coze 官方文档",{"title":1627,"url":789},"Coze vs Dify vs FastGPT 选型 2026",{"title":1629,"url":1281},"BestHub 三平台对比","tools\u002Fagent\u002Fplatform\u002Fcoze",[1632,1633,1634,1635,1636],"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",[1639,1640,1641,1642,1643,1644],"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":1649,"title":303,"alternatives":1650,"api_compatible":1576,"body":1653,"category":1571,"chinese_friendly":211,"cover":2626,"description":2627,"domestic":1575,"extension":726,"faq":1576,"free":1575,"github":1932,"languages":2628,"meta":2630,"models":1576,"navigation":728,"notSuitable":1576,"opensource":728,"path":2631,"pillar":1592,"platforms":2632,"priceTable":2636,"pricing":2653,"published":1615,"relatedPlaybooks":1576,"relatedReviews":1576,"score":2654,"self_host":728,"seo":2655,"seoTitle":2656,"slug":733,"sources":2657,"stem":2669,"suitable":1576,"tagline":2670,"tags":2671,"updated":1645,"verdict":2677,"website":2572,"__hash__":2678},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Fdify.md",[734,732,1651,1652],"agent\u002Fplatform\u002Fn8n","agent\u002Fplatform\u002Flangflow",{"type":13,"value":1654,"toc":2604},[1655,1657,1674,1679,1681,1685,1688,1756,1759,1763,1766,1780,1796,1800,1803,1814,1818,1826,1837,1841,1844,1846,1850,1859,1917,1924,1928,1936,1986,1989,2009,2013,2016,2076,2082,2084,2176,2179,2208,2210,2338,2350,2385,2387,2459,2461,2463,2483,2485,2514,2516,2563,2565,2596,2601],[16,1656,755],{"id":754},[36,1658,1660,1665],{"className":1659},[39,40,41],[20,1661,1662,1664],{},[24,1663,763],{}," 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。",[20,1666,782,1667,1670,1671,1673],{},[24,1668,1669],{},"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 ",[100,1672,496],{"href":495}," 折腾，新手得读 1-2 小时文档。",[794,1675,1676],{},[20,1677,1678],{},"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。",[16,1680,801],{"id":801},[803,1682,1684],{"id":1683},"可视化工作流chatflow-workflow","可视化工作流（Chatflow + Workflow）",[20,1686,1687],{},"Dify 把 LLM 应用拆成两种\"应用类型\"：",[329,1689,1690,1703],{},[332,1691,1692],{},[335,1693,1694,1697,1700],{},[338,1695,1696],{},"类型",[338,1698,1699],{},"适合场景",[338,1701,1702],{},"编排范式",[363,1704,1705,1718,1730,1743],{},[335,1706,1707,1712,1715],{},[368,1708,1709],{},[24,1710,1711],{},"Chatbot",[368,1713,1714],{},"简单对话机器人",[368,1716,1717],{},"prompt + tools",[335,1719,1720,1724,1727],{},[368,1721,1722],{},[24,1723,355],{},[368,1725,1726],{},"自主多步任务",[368,1728,1729],{},"ReAct \u002F Function Calling",[335,1731,1732,1737,1740],{},[368,1733,1734],{},[24,1735,1736],{},"Chatflow",[368,1738,1739],{},"对话型工作流（多轮 + 分支）",[368,1741,1742],{},"节点 DAG，带聊天上下文",[335,1744,1745,1750,1753],{},[368,1746,1747],{},[24,1748,1749],{},"Workflow",[368,1751,1752],{},"单次输入→输出（API 模式）",[368,1754,1755],{},"节点 DAG，无对话状态",[20,1757,1758],{},"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\"。",[803,1760,1762],{"id":1761},"rag-知识库","RAG 知识库",[20,1764,1765],{},"内置完整 RAG 链路：",[108,1767,1768,1771,1774,1777],{},[111,1769,1770],{},"上传文档（PDF \u002F Word \u002F Markdown \u002F 网页）",[111,1772,1773],{},"自动分块 + embedding（可配置分段策略和 embedding 模型）",[111,1775,1776],{},"混合检索（向量 + 全文 + 重排）",[111,1778,1779],{},"引用溯源（回答末尾自动附原文片段）",[20,1781,1782,1783,1788,1789,1792,1793,1795],{},"注意：根据 ",[100,1784,1787],{"href":1785,"rel":1786},"https:\u002F\u002Fzhuanlan.zhihu.com\u002Fp\u002F1887141987838309480",[104],"知乎 LLM 实战笔记 2025-03 对比"," 的实测，Dify ",[24,1790,1791],{},"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 ",[100,1794,486],{"href":683},"（实测准确率高 10+ 个百分点），Dify 胜在工作流而非纯 RAG。",[803,1797,1799],{"id":1798},"模型生态40-提供商","模型生态：40+ 提供商",[20,1801,1802],{},"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。",[20,1804,1805,1806,1808,1809,1813],{},"国产模型原生支持（不像 ",[100,1807,486],{"href":683}," 需要 ",[100,1810,1812],{"href":1811},"\u002Fcoding\u002Fapi\u002Fone-api.html","OneAPI"," 中转），是 Dify 在国内 toB 场景流行的关键。",[803,1815,1817],{"id":1816},"mcp-协议支持","MCP 协议支持",[20,1819,1820,1821,1825],{},"Dify 较早接入了 ",[100,1822,1824],{"href":1823},"\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：",[246,1827,1828,1831,1834],{},[111,1829,1830],{},"通过 MCP 调本地 PostgreSQL \u002F SQLite",[111,1832,1833],{},"通过 MCP 调 GitHub \u002F Slack \u002F Linear",[111,1835,1836],{},"通过 MCP 调自家内部系统（写一个 MCP Server 即可）",[803,1838,1840],{"id":1839},"api-first","API-first",[20,1842,1843],{},"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。",[16,1845,1061],{"id":1061},[803,1847,1849],{"id":1848},"云版difyai","云版（dify.ai）",[20,1851,1852,1853,1858],{},"根据 ",[100,1854,1857],{"href":1855,"rel":1856},"https:\u002F\u002Fwww.tooljunction.io\u002Fai-tools\u002Fdify-ai",[104],"tooljunction.io 2026 评测"," 引用的官方定价：",[329,1860,1861,1872],{},[332,1862,1863],{},[335,1864,1865,1867,1869],{},[338,1866,340],{},[338,1868,343],{},[338,1870,1871],{},"主要限制",[363,1873,1874,1884,1895,1906],{},[335,1875,1876,1879,1881],{},[368,1877,1878],{},"Sandbox",[368,1880,370],{},[368,1882,1883],{},"200 次模型调用，1 app，5MB 知识库",[335,1885,1886,1889,1892],{},[368,1887,1888],{},"Professional",[368,1890,1891],{},"$59\u002F月起",[368,1893,1894],{},"5000 调用\u002F月，多 app，50MB 知识库",[335,1896,1897,1900,1903],{},[368,1898,1899],{},"Team",[368,1901,1902],{},"$159\u002F月起",[368,1904,1905],{},"团队协作、SSO",[335,1907,1908,1911,1914],{},[368,1909,1910],{},"Enterprise",[368,1912,1913],{},"联系销售",[368,1915,1916],{},"定制 SLA、私有云",[20,1918,1919,1920,1923],{},"注意：云版价格只是 Dify 平台费，",[24,1921,1922],{},"模型 API 费用另算","（自带 OpenAI \u002F Anthropic key）。",[803,1925,1927],{"id":1926},"自托管推荐","自托管（推荐）",[20,1929,1930,1935],{},[100,1931,1934],{"href":1932,"rel":1933},"https:\u002F\u002Fgithub.com\u002Flanggenius\u002Fdify",[104],"官方 GitHub 仓库"," 提供 Docker Compose 部署，社区版完全免费可商用：",[165,1937,1939],{"className":167,"code":1938,"language":169,"meta":170,"style":170},"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",[172,1940,1941,1950,1957,1968,1980],{"__ignoreMap":170},[175,1942,1943,1945,1947],{"class":177,"line":178},[175,1944,182],{"class":181},[175,1946,186],{"class":185},[175,1948,1949],{"class":185}," https:\u002F\u002Fgithub.com\u002Flanggenius\u002Fdify.git\n",[175,1951,1952,1954],{"class":177,"line":192},[175,1953,196],{"class":195},[175,1955,1956],{"class":185}," dify\u002Fdocker\n",[175,1958,1959,1962,1965],{"class":177,"line":202},[175,1960,1961],{"class":181},"cp",[175,1963,1964],{"class":185}," .env.example",[175,1966,1967],{"class":185}," .env\n",[175,1969,1970,1973,1976,1978],{"class":177,"line":211},[175,1971,1972],{"class":181},"docker",[175,1974,1975],{"class":185}," compose",[175,1977,223],{"class":185},[175,1979,226],{"class":195},[175,1981,1982],{"class":177,"line":1572},[175,1983,1985],{"class":1984},"sJ8bj","# 默认 http:\u002F\u002Flocalhost \u002F 端口可在 .env 调整\n",[20,1987,1988],{},"硬件门槛（社区共识，非官方硬性要求）：",[246,1990,1991,1997,2003],{},[111,1992,1993,1996],{},[24,1994,1995],{},"最低","：2 核 4G，纯外接 API 模式",[111,1998,1999,2002],{},[24,2000,2001],{},"推荐","：4 核 8G + 至少 30GB 磁盘（向量数据 + 文件存储）",[111,2004,2005,2008],{},[24,2006,2007],{},"企业","：8 核 16G+，单机日活上千",[803,2010,2012],{"id":2011},"真实-tco","真实 TCO",[20,2014,2015],{},"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）：",[329,2017,2018,2031],{},[332,2019,2020],{},[335,2021,2022,2025,2028],{},[338,2023,2024],{},"成本项",[338,2026,2027],{},"云版 Professional",[338,2029,2030],{},"自托管",[363,2032,2033,2044,2054,2065],{},[335,2034,2035,2038,2041],{},[368,2036,2037],{},"平台费",[368,2039,2040],{},"~$2,100（3 年）",[368,2042,2043],{},"$0",[335,2045,2046,2049,2051],{},[368,2047,2048],{},"服务器",[368,2050,2043],{},[368,2052,2053],{},"~$50\u002F月 × 36 = $1,800",[335,2055,2056,2059,2062],{},[368,2057,2058],{},"模型 API",[368,2060,2061],{},"与下同",[368,2063,2064],{},"与上同",[335,2066,2067,2070,2073],{},[368,2068,2069],{},"运维人力",[368,2071,2072],{},"0",[368,2074,2075],{},"约 0.2 人月",[20,2077,2078,2081],{},[24,2079,2080],{},"结论","：日活 \u003C 100 用云版省心；> 500 或数据敏感场景自托管 ROI 更好。",[16,2083,1097],{"id":1096},[165,2085,2087],{"className":167,"code":2086,"language":169,"meta":170,"style":170},"# 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",[172,2088,2089,2094,2102,2108,2116,2126,2132,2138,2144,2149,2155,2160,2166,2171],{"__ignoreMap":170},[175,2090,2091],{"class":177,"line":178},[175,2092,2093],{"class":1984},"# 1. 自托管（社区版）\n",[175,2095,2096,2098,2100],{"class":177,"line":192},[175,2097,182],{"class":181},[175,2099,186],{"class":185},[175,2101,1949],{"class":185},[175,2103,2104,2106],{"class":177,"line":202},[175,2105,196],{"class":195},[175,2107,1956],{"class":185},[175,2109,2110,2112,2114],{"class":177,"line":211},[175,2111,1961],{"class":181},[175,2113,1964],{"class":185},[175,2115,1967],{"class":185},[175,2117,2118,2120,2122,2124],{"class":177,"line":1572},[175,2119,1972],{"class":181},[175,2121,1975],{"class":185},[175,2123,223],{"class":185},[175,2125,226],{"class":195},[175,2127,2129],{"class":177,"line":2128},6,[175,2130,2131],{"emptyLinePlaceholder":728},"\n",[175,2133,2135],{"class":177,"line":2134},7,[175,2136,2137],{"class":1984},"# 2. 浏览器打开 http:\u002F\u002Flocalhost\n",[175,2139,2141],{"class":177,"line":2140},8,[175,2142,2143],{"class":1984},"#    首次会让你创建 admin 账号\n",[175,2145,2147],{"class":177,"line":2146},9,[175,2148,2131],{"emptyLinePlaceholder":728},[175,2150,2152],{"class":177,"line":2151},10,[175,2153,2154],{"class":1984},"# 3. 进入\"设置 → 模型供应商\"，配置 OpenAI \u002F 国产模型 API key\n",[175,2156,2158],{"class":177,"line":2157},11,[175,2159,2131],{"emptyLinePlaceholder":728},[175,2161,2163],{"class":177,"line":2162},12,[175,2164,2165],{"class":1984},"# 4. 在主界面\"创建空白应用\"，选 Chatflow 或 Workflow\n",[175,2167,2168],{"class":177,"line":8},[175,2169,2170],{"class":1984},"# 5. 拖入\"开始 → LLM → 结束\"节点试一下基础 prompt\n",[175,2172,2173],{"class":177,"line":5},[175,2174,2175],{"class":1984},"# 6. 满意了点右上\"发布\"，自动生成 API endpoint\n",[16,2177,2178],{"id":2178},"国内使用注意事项",[108,2180,2181,2187,2193,2199],{},[111,2182,2183,2186],{},[24,2184,2185],{},"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",[111,2188,2189,2192],{},[24,2190,2191],{},"自托管 + 国产模型"," = 完全国内闭环，是 Dify 在国内最大优势",[111,2194,2195,2198],{},[24,2196,2197],{},"Docker 镜像拉取","：国内可能慢，建议配 Docker registry 镜像（阿里云 \u002F 网易）",[111,2200,2201,2204,2205,2207],{},[24,2202,2203],{},"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 ",[100,2206,496],{"href":495}," 迁到 Dify",[16,2209,474],{"id":474},[329,2211,2212,2232],{},[332,2213,2214],{},[335,2215,2216,2218,2220,2224,2228],{},[338,2217,483],{},[338,2219,303],{},[338,2221,2222],{},[100,2223,496],{"href":495},[338,2225,2226],{},[100,2227,486],{"href":683},[338,2229,2230],{},[100,2231,1470],{"href":1469},[363,2233,2234,2247,2259,2272,2284,2297,2311,2324],{},[335,2235,2236,2238,2240,2242,2244],{},[368,2237,739],{},[368,2239,1173],{},[368,2241,531],{},[368,2243,1173],{},[368,2245,2246],{},"✅（fair-code）",[335,2248,2249,2251,2253,2255,2257],{},[368,2250,538],{},[368,2252,1173],{},[368,2254,531],{},[368,2256,1173],{},[368,2258,1173],{},[335,2260,2261,2263,2265,2268,2270],{},[368,2262,1195],{},[368,2264,561],{},[368,2266,2267],{},"★★☆☆☆ 最简单",[368,2269,561],{},[368,2271,549],{},[335,2273,2274,2276,2278,2280,2282],{},[368,2275,1211],{},[368,2277,541],{},[368,2279,549],{},[368,2281,561],{},[368,2283,541],{},[335,2285,2286,2288,2290,2292,2294],{},[368,2287,1224],{},[368,2289,549],{},[368,2291,561],{},[368,2293,541],{},[368,2295,2296],{},"★★☆☆☆",[335,2298,2299,2302,2304,2306,2309],{},[368,2300,2301],{},"模型生态",[368,2303,541],{},[368,2305,549],{},[368,2307,2308],{},"★★★☆☆（OneAPI 中转）",[368,2310,549],{},[335,2312,2313,2316,2318,2320,2322],{},[368,2314,2315],{},"中文场景",[368,2317,549],{},[368,2319,541],{},[368,2321,549],{},[368,2323,561],{},[335,2325,2326,2329,2331,2334,2336],{},[368,2327,2328],{},"字节生态绑定",[368,2330,531],{},[368,2332,2333],{},"✅（飞书\u002F抖音深度集成）",[368,2335,531],{},[368,2337,531],{},[20,2339,2340,2342,2343,1284,2346,2349],{},[24,2341,623],{},"（基于 ",[100,2344,1283],{"href":1281,"rel":2345},[104],[100,2347,1288],{"href":789,"rel":2348},[104]," 两份选型指南综合）：",[246,2351,2352,2358,2365,2371,2378],{},[111,2353,2354,2357],{},[24,2355,2356],{},"数据必须不出内网 + 工作流复杂"," → Dify",[111,2359,2360,639,2363],{},[24,2361,2362],{},"个人 \u002F 小团队 \u002F 快速原型 + 字节生态",[100,2364,496],{"href":495},[111,2366,2367,639,2369],{},[24,2368,1308],{},[100,2370,486],{"href":683},[111,2372,2373,639,2376],{},[24,2374,2375],{},"重点是连接外部 SaaS（Slack \u002F Notion \u002F 数据库）",[100,2377,1470],{"href":1469},[111,2379,2380,639,2383],{},[24,2381,2382],{},"要画图式表达 LangChain pipeline",[100,2384,1450],{"href":1449},[16,2386,1321],{"id":1321},[246,2388,2389,2395,2411,2422,2435,2441,2447,2453],{},[111,2390,2391,2394],{},[24,2392,2393],{},"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。",[111,2396,2397,59,2403,2406,2407,2410],{},[24,2398,2399,2402],{},[172,2400,2401],{},".env"," 文件改完忘 restart",[172,2404,2405],{},"docker compose down && up -d","，不是 ",[172,2408,2409],{},"restart","——后者不重新加载 env。",[111,2412,2413,59,2416,2421],{},[24,2414,2415],{},"大版本升级会破坏数据库 schema",[100,2417,2420],{"href":2418,"rel":2419},"https:\u002F\u002Fdocs.dify.ai\u002Fzh-hans",[104],"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。",[111,2423,2424,2427,2428,2430,2431,2434],{},[24,2425,2426],{},"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 ",[172,2429,2401],{}," 的 ",[172,2432,2433],{},"UPLOAD_FILE_SIZE_LIMIT"," 并重启容器。",[111,2436,2437,2440],{},[24,2438,2439],{},"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。",[111,2442,2443,2446],{},[24,2444,2445],{},"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。",[111,2448,2449,2452],{},[24,2450,2451],{},"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。",[111,2454,2455,2458],{},[24,2456,2457],{},"国内 Docker 拉取镜像慢","：先配国内 registry，否则首次 pull 可能要 30+ 分钟。",[16,2460,1398],{"id":1397},[20,2462,1401],{},[246,2464,2465,2468,2471,2474,2477,2480],{},[111,2466,2467],{},"中大型企业 LLM 中台建设",[111,2469,2470],{},"需要私有化部署（金融 \u002F 医疗 \u002F 政府）",[111,2472,2473],{},"想做\"AI 工作流即产品\"的开发团队",[111,2475,2476],{},"同时需要 RAG + Agent + Workflow 三件套",[111,2478,2479],{},"想用国产模型 + 国际模型混合编排",[111,2481,2482],{},"已经接受 Docker + 一定运维投入",[20,2484,1421],{},[246,2486,2487,2493,2499,2502,2508],{},[111,2488,2489,2490,2492],{},"纯个人玩家做对话机器人（",[100,2491,496],{"href":495}," 更快）",[111,2494,2495,2496,2498],{},"只想做企业知识库 QA（",[100,2497,486],{"href":683}," RAG 更专）",[111,2500,2501],{},"团队完全没运维能力（云版还行，自托管会踩坑）",[111,2503,2504,2505,2507],{},"需要深度对接字节飞书 \u002F 抖音（",[100,2506,496],{"href":495}," 原生）",[111,2509,2510,2511,2513],{},"工作流核心是连接 100+ SaaS（",[100,2512,1470],{"href":1469}," 节点更全）",[16,2515,676],{"id":676},[246,2517,2518,2528,2540,2555],{},[111,2519,1458,2520,1379,2522,1379,2524,1379,2526],{},[100,2521,496],{"href":495},[100,2523,486],{"href":683},[100,2525,1470],{"href":1469},[100,2527,1450],{"href":1449},[111,2529,2530,2531,1379,2533,1379,2535,1379,2538],{},"概念基础：",[100,2532,1477],{"href":1476},[100,2534,738],{"href":1480},[100,2536,2537],{"href":1823},"MCP",[100,2539,1484],{"href":1483},[111,2541,2542,2543,1379,2547,1379,2551,1379,2553],{},"模型选型：",[100,2544,2546],{"href":2545},"\u002Fmodels\u002Fgpt-5.html","GPT-5",[100,2548,2550],{"href":2549},"\u002Fmodels\u002Fclaude-sonnet-4.html","Claude Sonnet 4",[100,2552,1498],{"href":1497},[100,2554,1506],{"href":1505},[111,2556,1509,2557,1379,2561],{},[100,2558,2560],{"href":2559},"\u002Fwiki\u002Ffine-tuning-vs-rag.html","Fine-tuning vs RAG",[100,2562,1517],{"href":1516},[16,2564,1520],{"id":1520},[246,2566,2567,2574,2580,2586,2593],{},[111,2568,2569,2570],{},"官网：",[100,2571,2572],{"href":2572,"rel":2573},"https:\u002F\u002Fdify.ai",[104],[111,2575,2576,2577],{},"中文文档：",[100,2578,2418],{"href":2418,"rel":2579},[104],[111,2581,2582,2583],{},"GitHub：",[100,2584,1932],{"href":1932,"rel":2585},[104],[111,2587,2588,2589],{},"官方定价：",[100,2590,2591],{"href":2591,"rel":2592},"https:\u002F\u002Fdify.ai\u002Fpricing",[104],[111,2594,2595],{},"第三方评测：tooljunction.io \u002F chatforest.com \u002F besthub.dev \u002F joshuaopolko.com \u002F 知乎 LLM 实战笔记",[20,2597,2598,2599,1554],{},"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 ",[100,2600,1553],{"href":1553},[710,2602,2603],{},"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":170,"searchDepth":202,"depth":202,"links":2605},[2606,2607,2614,2619,2620,2621,2622,2623,2624,2625],{"id":754,"depth":192,"text":755},{"id":801,"depth":192,"text":801,"children":2608},[2609,2610,2611,2612,2613],{"id":1683,"depth":202,"text":1684},{"id":1761,"depth":202,"text":1762},{"id":1798,"depth":202,"text":1799},{"id":1816,"depth":202,"text":1817},{"id":1839,"depth":202,"text":1840},{"id":1061,"depth":192,"text":1061,"children":2615},[2616,2617,2618],{"id":1848,"depth":202,"text":1849},{"id":1926,"depth":202,"text":1927},{"id":2011,"depth":202,"text":2012},{"id":1096,"depth":192,"text":1097},{"id":2178,"depth":192,"text":2178},{"id":474,"depth":192,"text":474},{"id":1321,"depth":192,"text":1321},{"id":1397,"depth":192,"text":1398},{"id":676,"depth":192,"text":676},{"id":1520,"depth":192,"text":1520},"\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。",[1578,1579,2629],"ja",{},"\u002Ftools\u002Fagent\u002Fplatform\u002Fdify",[2633,2634,2635,1972],"windows","macos","linux",[2637,2641,2645,2649],{"plan":2638,"price":2043,"features":2639,"notes":2640},"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":2642,"price":2043,"features":2643,"notes":2644},"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":2646,"price":1891,"features":2647,"notes":2648},"Cloud Professional","更高调用额度 + 团队协作 + 商用支持","商用云首选",{"plan":2650,"price":2651,"features":2652,"notes":1913},"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":1572,"ux":211,"price":1572,"cn_support":211,"stability":211},{"title":303,"description":2627},"Dify 评测 2026：开源 LLMOps 与 AI Agent 平台，自托管指南",[2658,2660,2662,2664,2666],{"title":2659,"url":2418},"Dify 官方文档（中文）",{"title":2661,"url":1932},"Dify GitHub",{"title":2663,"url":2591},"Dify 官方定价",{"title":2665,"url":1281},"Coze vs Dify vs FastGPT 选型",{"title":2667,"url":2668},"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",[1639,2672,2673,2674,1641,2675,2676],"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":2680,"title":486,"alternatives":2681,"api_compatible":2684,"body":2686,"category":1571,"chinese_friendly":1572,"cover":3638,"description":3639,"domestic":1575,"extension":726,"faq":1576,"free":1575,"github":3601,"languages":3640,"meta":3641,"models":3642,"navigation":728,"notSuitable":3646,"opensource":728,"path":3650,"pillar":1592,"platforms":3651,"priceTable":3652,"pricing":3674,"published":1615,"relatedPlaybooks":3675,"relatedReviews":3677,"score":3679,"self_host":728,"seo":3680,"seoTitle":1576,"slug":732,"sources":3681,"stem":3692,"suitable":3693,"tagline":3699,"tags":3700,"updated":1645,"verdict":3704,"website":3595,"__hash__":3705},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Ffastgpt.md",[733,734,2682,2683],"agent\u002Fplatform\u002Fragflow","agent\u002Fplatform\u002Fanythingllm",[2685],"openai",{"type":13,"value":2687,"toc":3620},[2688,2690,2717,2726,2728,2732,2735,2749,2752,2778,2782,2789,2817,2824,2828,2881,2888,2891,2894,2897,2904,2950,2956,2960,2991,2995,3095,3098,3101,3105,3111,3208,3223,3225,3373,3381,3413,3419,3421,3494,3496,3498,3518,3520,3538,3540,3587,3589,3612,3617],[16,2689,755],{"id":754},[36,2691,2693,2706],{"className":2692},[39,40,41],[20,2694,2695,2697,2698,2701,2702,2705],{},[24,2696,763],{}," labring 团队开源的 LLM 知识库 RAG 平台，27k+ GitHub star（截至 2026-03 数据，",[100,2699,264],{"href":262,"rel":2700},[104]," 引用），Apache 2.0 许可证可商用。",[24,2703,2704],{},"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。",[20,2707,2708,2709,2712,2713,2716],{},"最大价值在 ",[24,2710,2711],{},"国内企业知识库 + 私有部署"," 场景。代价是 ",[24,2714,2715],{},"配置门槛","：docker 基础 + 网络知识 + 一定运维能力。",[794,2718,2719],{},[20,2720,2721,2722,2725],{},"来源说明：本文基于 fastgpt.io 官方页面、github.com\u002Flabring\u002FFastGPT 仓库、",[100,2723,105],{"href":102,"rel":2724},[104],"、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。",[16,2727,801],{"id":801},[803,2729,2731],{"id":2730},"知识库管理核心能力","知识库管理（核心能力）",[20,2733,2734],{},"支持文件类型：",[246,2736,2737,2740,2743,2746],{},[111,2738,2739],{},"文档：PDF \u002F Word \u002F Markdown \u002F TXT \u002F HTML",[111,2741,2742],{},"表格：Excel \u002F CSV",[111,2744,2745],{},"网页：URL 抓取 + 定时同步",[111,2747,2748],{},"API：通过接口推送内容",[20,2750,2751],{},"处理流程：上传 → 文本切分 → 向量化 → 存储 → 可用于问答。支持：",[246,2753,2754,2760,2766,2772],{},[111,2755,2756,2759],{},[24,2757,2758],{},"文件夹分组","：不同主题 \u002F 部门分类",[111,2761,2762,2765],{},[24,2763,2764],{},"多种分块策略","：默认按段落 \u002F 按 token 数 \u002F 自定义",[111,2767,2768,2771],{},[24,2769,2770],{},"批量导入","：脚本化大批量同步",[111,2773,2774,2777],{},[24,2775,2776],{},"定时同步","：网页源自动更新",[803,2779,2781],{"id":2780},"rag-流程编排最强卖点","RAG 流程编排（最强卖点）",[20,2783,2784,2788],{},[100,2785,2787],{"href":102,"rel":2786},[104],"南环 AI 2026 评测"," 总结的 FastGPT RAG 链路：",[108,2790,2791,2796,2801,2807,2812],{},[111,2792,2793,2795],{},[24,2794,115],{},"：改写 \u002F 扩展 \u002F 错词纠正（提升召回率）",[111,2797,2798,2800],{},[24,2799,121],{},"：语义检索 \u002F 关键词 BM25 \u002F 混合检索，可调相似度阈值",[111,2802,2803,2806],{},[24,2804,2805],{},"重排序（Rerank）","：对初步检索结果二次排序，提升相关性",[111,2808,2809,2811],{},[24,2810,133],{},"：最优 chunk + 问题 → prompt",[111,2813,2814,2816],{},[24,2815,139],{},"：调大模型基于检索结果回答 + 引用标注",[20,2818,2819,2820,2823],{},"每一步都可视化调参，这是 FastGPT 比 Coze \u002F Dify 在 ",[24,2821,2822],{},"纯知识库 QA 精度","上更高的原因。",[803,2825,2827],{"id":2826},"多模型支持不绑定厂商","多模型支持（不绑定厂商）",[329,2829,2830,2840],{},[332,2831,2832],{},[335,2833,2834,2837],{},[338,2835,2836],{},"模型类别",[338,2838,2839],{},"支持",[363,2841,2842,2850,2857,2865,2873],{},[335,2843,2844,2847],{},[368,2845,2846],{},"国产闭源",[368,2848,2849],{},"豆包 \u002F 通义千问 \u002F 文心一言 \u002F 智谱 GLM \u002F Moonshot Kimi \u002F MiniMax",[335,2851,2852,2854],{},[368,2853,739],{},[368,2855,2856],{},"LLaMA \u002F Qwen \u002F ChatGLM \u002F DeepSeek 等可自部署",[335,2858,2859,2862],{},[368,2860,2861],{},"OpenAI 系",[368,2863,2864],{},"GPT-5 \u002F GPT-5 mini \u002F o3",[335,2866,2867,2870],{},[368,2868,2869],{},"Claude 系",[368,2871,2872],{},"Sonnet 4 \u002F Opus 4 \u002F Haiku",[335,2874,2875,2878],{},[368,2876,2877],{},"嵌入 \u002F 重排",[368,2879,2880],{},"BGE \u002F m3e \u002F OpenAI text-embedding-3",[20,2882,2883,2884,2887],{},"可以在 ",[24,2885,2886],{},"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。",[803,2889,2890],{"id":2890},"工作流与高级编排",[20,2892,2893],{},"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。",[803,2895,2896],{"id":2896},"多向量库选择",[20,2898,2899,2903],{},[100,2900,2902],{"href":262,"rel":2901},[104],"腾讯云教程"," 公开的 4 种向量后端：",[329,2905,2906,2916],{},[332,2907,2908],{},[335,2909,2910,2913],{},[338,2911,2912],{},"后端",[338,2914,2915],{},"适用",[363,2917,2918,2926,2934,2942],{},[335,2919,2920,2923],{},[368,2921,2922],{},"PgVector",[368,2924,2925],{},"5000 万索引以下，新手 \u002F 小规模",[335,2927,2928,2931],{},[368,2929,2930],{},"Milvus",[368,2932,2933],{},"亿级以上，高性能",[335,2935,2936,2939],{},[368,2937,2938],{},"Zilliz Cloud",[368,2940,2941],{},"Milvus 全托管 SaaS",[335,2943,2944,2947],{},[368,2945,2946],{},"SeekDB \u002F OceanBase",[368,2948,2949],{},"企业级国产化",[20,2951,2952,2953,69],{},"部署时选对应 ",[172,2954,2955],{},"docker-compose.{pgvector|milvus|...}.yml",[803,2957,2959],{"id":2958},"api-与-mcp","API 与 MCP",[246,2961,2962,2968,2974,2985],{},[111,2963,2964,2967],{},[24,2965,2966],{},"对话 API","：流式 \u002F 非流式 HTTP，OpenAI 兼容",[111,2969,2970,2973],{},[24,2971,2972],{},"知识库检索 API","：单独调检索（不走生成）做 hybrid pipeline",[111,2975,2976,2979,2980,2984],{},[24,2977,2978],{},"MCP Server","：3005 端口暴露 MCP SSE 服务，可被 ",[100,2981,2983],{"href":2982},"\u002Fcoding\u002Fcli\u002Fclaude-code.html","Claude Code"," 等客户端直接接入",[111,2986,2987,2990],{},[24,2988,2989],{},"Webhook","：回调通知",[16,2992,2994],{"id":2993},"部署-10-分钟docker","部署 10 分钟（Docker）",[165,2996,2998],{"className":167,"code":2997,"language":169,"meta":170,"style":170},"# 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",[172,2999,3000,3005,3013,3020,3024,3029,3042,3046,3051,3058,3064,3068,3073,3085,3089],{"__ignoreMap":170},[175,3001,3002],{"class":177,"line":178},[175,3003,3004],{"class":1984},"# 克隆代码\n",[175,3006,3007,3009,3011],{"class":177,"line":192},[175,3008,182],{"class":181},[175,3010,186],{"class":185},[175,3012,189],{"class":185},[175,3014,3015,3017],{"class":177,"line":202},[175,3016,196],{"class":195},[175,3018,3019],{"class":185}," FastGPT\n",[175,3021,3022],{"class":177,"line":211},[175,3023,2131],{"emptyLinePlaceholder":728},[175,3025,3026],{"class":177,"line":1572},[175,3027,3028],{"class":1984},"# 切到最新稳定版（参考 GitHub releases）\n",[175,3030,3031,3033,3036,3039],{"class":177,"line":2128},[175,3032,182],{"class":181},[175,3034,3035],{"class":185}," switch",[175,3037,3038],{"class":195}," -c",[175,3040,3041],{"class":195}," 4.14.7.2\n",[175,3043,3044],{"class":177,"line":2134},[175,3045,2131],{"emptyLinePlaceholder":728},[175,3047,3048],{"class":177,"line":2140},[175,3049,3050],{"class":1984},"# 选向量库版本（个人 \u002F 小规模选 pg）\n",[175,3052,3053,3055],{"class":177,"line":2146},[175,3054,196],{"class":195},[175,3056,3057],{"class":185}," deploy\u002Fdocker\u002Fcn\n",[175,3059,3060,3062],{"class":177,"line":2151},[175,3061,205],{"class":181},[175,3063,208],{"class":185},[175,3065,3066],{"class":177,"line":2157},[175,3067,2131],{"emptyLinePlaceholder":728},[175,3069,3070],{"class":177,"line":2162},[175,3071,3072],{"class":1984},"# 启动\n",[175,3074,3075,3077,3079,3081,3083],{"class":177,"line":8},[175,3076,214],{"class":181},[175,3078,217],{"class":195},[175,3080,220],{"class":185},[175,3082,223],{"class":185},[175,3084,226],{"class":195},[175,3086,3087],{"class":177,"line":5},[175,3088,2131],{"emptyLinePlaceholder":728},[175,3090,3092],{"class":177,"line":3091},15,[175,3093,3094],{"class":1984},"# 访问 http:\u002F\u002F\u003Cip>:3000，默认账号 root \u002F 1234\n",[20,3096,3097],{},"最低配置：2C4G + 20GB 硬盘 + Docker 28+ + Docker Compose 2.20+。",[20,3099,3100],{},"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。",[16,3102,3104],{"id":3103},"云版-vs-自托管对比","云版 vs 自托管对比",[20,3106,3107,3110],{},[100,3108,327],{"href":325,"rel":3109},[104]," 公开数据：",[329,3112,3113,3134],{},[332,3114,3115],{},[335,3116,3117,3119,3121,3123,3126,3128,3130,3132],{},[338,3118,340],{},[338,3120,343],{},[338,3122,346],{},[338,3124,3125],{},"知识库索引",[338,3127,352],{},[338,3129,355],{},[338,3131,358],{},[338,3133,361],{},[363,3135,3136,3154,3172,3190],{},[335,3137,3138,3140,3142,3144,3146,3148,3150,3152],{},[368,3139,370],{},[368,3141,373],{},[368,3143,376],{},[368,3145,379],{},[368,3147,382],{},[368,3149,385],{},[368,3151,388],{},[368,3153,391],{},[335,3155,3156,3158,3160,3162,3164,3166,3168,3170],{},[368,3157,396],{},[368,3159,399],{},[368,3161,402],{},[368,3163,405],{},[368,3165,408],{},[368,3167,411],{},[368,3169,391],{},[368,3171,416],{},[335,3173,3174,3176,3178,3180,3182,3184,3186,3188],{},[368,3175,421],{},[368,3177,424],{},[368,3179,427],{},[368,3181,430],{},[368,3183,411],{},[368,3185,435],{},[368,3187,376],{},[368,3189,440],{},[335,3191,3192,3194,3196,3198,3200,3202,3204,3206],{},[368,3193,445],{},[368,3195,448],{},[368,3197,451],{},[368,3199,451],{},[368,3201,451],{},[368,3203,451],{},[368,3205,451],{},[368,3207,451],{},[20,3209,3210,3213,3214,3217,3218,3222],{},[24,3211,3212],{},"云版适合","：不想运维、量小、要快速上线\n",[24,3215,3216],{},"自托管适合","：量大（10 万+ 日问答）、数据敏感、要深度定制——按 ",[100,3219,3221],{"href":102,"rel":3220},[104],"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\"",[16,3224,474],{"id":474},[329,3226,3227,3248],{},[332,3228,3229],{},[335,3230,3231,3233,3235,3239,3243,3245],{},[338,3232,483],{},[338,3234,486],{},[338,3236,3237],{},[100,3238,303],{"href":302},[338,3240,3241],{},[100,3242,496],{"href":495},[338,3244,499],{},[338,3246,3247],{},"AnythingLLM",[363,3249,3250,3266,3282,3297,3313,3327,3344,3358],{},[335,3251,3252,3254,3256,3258,3260,3263],{},[368,3253,506],{},[368,3255,509],{},[368,3257,512],{},[368,3259,515],{},[368,3261,3262],{},"文档解析+RAG",[368,3264,3265],{},"桌面级 KB",[335,3267,3268,3270,3273,3275,3277,3279],{},[368,3269,739],{},[368,3271,3272],{},"✅ Apache 2.0",[368,3274,3272],{},[368,3276,531],{},[368,3278,3272],{},[368,3280,3281],{},"✅ MIT",[335,3283,3284,3286,3289,3291,3293,3295],{},[368,3285,538],{},[368,3287,3288],{},"★★★★★ docker",[368,3290,541],{},[368,3292,1184],{},[368,3294,549],{},[368,3296,541],{},[335,3298,3299,3301,3304,3306,3308,3311],{},[368,3300,554],{},[368,3302,3303],{},"★★★★★ 最细",[368,3305,549],{},[368,3307,561],{},[368,3309,3310],{},"★★★★★ 文档解析最强",[368,3312,561],{},[335,3314,3315,3317,3319,3321,3323,3325],{},[368,3316,568],{},[368,3318,549],{},[368,3320,541],{},[368,3322,549],{},[368,3324,561],{},[368,3326,2296],{},[335,3328,3329,3332,3335,3337,3340,3342],{},[368,3330,3331],{},"上手",[368,3333,3334],{},"★★★☆☆ 需 docker",[368,3336,549],{},[368,3338,3339],{},"★★★★★ 最简单",[368,3341,561],{},[368,3343,549],{},[335,3345,3346,3348,3350,3352,3354,3356],{},[368,3347,595],{},[368,3349,541],{},[368,3351,549],{},[368,3353,541],{},[368,3355,549],{},[368,3357,561],{},[335,3359,3360,3363,3365,3367,3369,3371],{},[368,3361,3362],{},"多平台发布",[368,3364,611],{},[368,3366,549],{},[368,3368,541],{},[368,3370,618],{},[368,3372,618],{},[20,3374,3375,1278,3377,832],{},[24,3376,623],{},[100,3378,3380],{"href":102,"rel":3379},[104],"南环 AI 评测",[246,3382,3383,3388,3395,3402,3407],{},[111,3384,3385,632],{},[24,3386,3387],{},"核心需求是 RAG 精度",[111,3389,3390,639,3393],{},[24,3391,3392],{},"需要丰富插件 + 复杂工作流 + 多平台发布",[100,3394,303],{"href":302},[111,3396,3397,639,3400],{},[24,3398,3399],{},"零代码、快速发布到飞书 \u002F 微信",[100,3401,496],{"href":495},[111,3403,3404,654],{},[24,3405,3406],{},"文档解析（含 OCR \u002F 表格 \u002F 公式）是瓶颈",[111,3408,3409,3412],{},[24,3410,3411],{},"桌面 \u002F 单机使用"," → AnythingLLM",[20,3414,3415,3418],{},[24,3416,3417],{},"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。",[16,3420,1321],{"id":1321},[246,3422,3423,3436,3445,3451,3461,3472,3478,3484],{},[111,3424,3425,59,3428,3431,3432,3435],{},[24,3426,3427],{},"docker-compose 镜像 tag 不一致",[100,3429,2902],{"href":262,"rel":3430},[104]," 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 ",[172,3433,3434],{},"image:"," 行为正确版本即可",[111,3437,3438,3441,3442,3444],{},[24,3439,3440],{},"3000 端口冲突","：默认占用 3000（主服务）\u002F 9000（S3 \u002F MinIO）\u002F 3005（MCP）；改 ",[172,3443,214],{}," 的 ports 映射端口",[111,3446,3447,3450],{},[24,3448,3449],{},"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",[111,3452,3453,3456,3457,3460],{},[24,3454,3455],{},"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 ",[100,3458,3221],{"href":102,"rel":3459},[104],"：\"新手 \u002F 小规模 PgVector，中大规模 Milvus，企业 \u002F 国产 OceanBase\"",[111,3462,3463,59,3466,3468,3469],{},[24,3464,3465],{},"MinIO 默认密码",[172,3467,279],{},"，",[24,3470,3471],{},"部署到公网前必须改",[111,3473,3474,3477],{},[24,3475,3476],{},"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\"",[111,3479,3480,3483],{},[24,3481,3482],{},"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",[111,3485,3486,3489,3490,1354],{},[24,3487,3488],{},"云版 AI 积分会过期","：未用完不能跨月累积（按 ",[100,3491,3493],{"href":325,"rel":3492},[104],"fastgpt.io 定价 FAQ",[16,3495,1398],{"id":1397},[20,3497,1401],{},[246,3499,3500,3503,3506,3509,3512,3515],{},[111,3501,3502],{},"企业内部知识库（员工手册 \u002F 制度 \u002F 流程）",[111,3504,3505],{},"产品 FAQ \u002F 用户手册问答",[111,3507,3508],{},"医疗 \u002F 法律 \u002F 金融垂直领域知识系统",[111,3510,3511],{},"数据严格不出网 + Apache 2.0 商用",[111,3513,3514],{},"有 docker 运维基础的技术团队",[111,3516,3517],{},"需要把 RAG 当后端服务的开发者（API 接入业务系统）",[20,3519,1421],{},[246,3521,3522,3527,3532,3535],{},[111,3523,3524,3525,1354],{},"完全非技术用户（去 ",[100,3526,496],{"href":495},[111,3528,3529,3530,1354],{},"主要需求是工作流 + 插件集成（去 ",[100,3531,303],{"href":302},[111,3533,3534],{},"文档解析 \u002F OCR 是首要痛点（RAGFlow）",[111,3536,3537],{},"不想自己运维 + 量很小（FastGPT 云免费版起步即可）",[16,3539,676],{"id":676},[246,3541,3542,3551,3567,3581],{},[111,3543,1458,3544,1379,3546,3548,3549],{},[100,3545,303],{"href":302},[100,3547,496],{"href":495}," \u002F RAGFlow \u002F AnythingLLM \u002F ",[100,3550,1470],{"href":1469},[111,3552,1473,3553,1379,3555,1379,3559,1379,3562,1379,3565],{},[100,3554,738],{"href":1480},[100,3556,3558],{"href":3557},"\u002Fwiki\u002Fembedding.html","Embedding",[100,3560,3561],{"href":3557},"Vector Database",[100,3563,3564],{"href":1480},"Reranker",[100,3566,1477],{"href":1476},[111,3568,1490,3569,1379,3571,1379,3573,1379,3575,1379,3579],{},[100,3570,1498],{"href":1497},[100,3572,1502],{"href":1501},[100,3574,1506],{"href":1505},[100,3576,3578],{"href":3577},"\u002Fmodels\u002Fkimi-k2.html","Kimi K2",[100,3580,1494],{"href":1493},[111,3582,1509,3583,1379,3585],{},[100,3584,1517],{"href":1516},[100,3586,1513],{"href":1512},[16,3588,1520],{"id":1520},[246,3590,3591,3597,3603,3609],{},[111,3592,2569,3593],{},[100,3594,3595],{"href":3595,"rel":3596},"https:\u002F\u002Ffastgpt.io",[104],[111,3598,2582,3599],{},[100,3600,3601],{"href":3601,"rel":3602},"https:\u002F\u002Fgithub.com\u002Flabring\u002FFastGPT",[104],[111,3604,3605,3606],{},"定价：",[100,3607,325],{"href":325,"rel":3608},[104],[111,3610,3611],{},"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",[20,3613,3614,3615,1554],{},"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 ",[100,3616,1553],{"href":1553},[710,3618,3619],{},"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":170,"searchDepth":202,"depth":202,"links":3621},[3622,3623,3631,3632,3633,3634,3635,3636,3637],{"id":754,"depth":192,"text":755},{"id":801,"depth":192,"text":801,"children":3624},[3625,3626,3627,3628,3629,3630],{"id":2730,"depth":202,"text":2731},{"id":2780,"depth":202,"text":2781},{"id":2826,"depth":202,"text":2827},{"id":2890,"depth":202,"text":2890},{"id":2896,"depth":202,"text":2896},{"id":2958,"depth":202,"text":2959},{"id":2993,"depth":192,"text":2994},{"id":3103,"depth":192,"text":3104},{"id":474,"depth":192,"text":474},{"id":1321,"depth":192,"text":1321},{"id":1397,"depth":192,"text":1398},{"id":676,"depth":192,"text":676},{"id":1520,"depth":192,"text":1520},"\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。",[1578,1579],{},[1583,1584,1582,3643,3644,3645],"gpt-4o","claude-sonnet-4","kimi",[3647,3648,3649],"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",[2633,2634,2635],[3653,3657,3662,3666,3670],{"plan":3654,"price":370,"limit":3655,"cn_pay":1599,"note":3656},"Self-host 开源","全功能 + 全数据本地","Apache 2.0 可商用",{"plan":3658,"price":3659,"limit":3660,"cn_pay":1599,"note":3661},"云免费版","¥0\u002F月","100 AI 积分 + 600 索引 + 3 知识库","试水",{"plan":3663,"price":399,"limit":3664,"cn_pay":1604,"note":3665},"云基础版","4000 积分 + 6000 索引 + 50 Agent","中小团队 SaaS",{"plan":3667,"price":424,"limit":3668,"cn_pay":1173,"note":3669},"云高级版","25000 积分 + 36000 索引 + 50 成员 + 200 Agent + 1500 QPM","企业级生产",{"plan":3671,"price":448,"limit":3672,"cn_pay":1173,"note":3673},"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",[3676],"onboarding\u002Ffastgpt-getting-started",[3678],"fastgpt-deep-review",{"power":211,"ux":211,"price":1572,"cn_support":1572,"stability":211},{"title":486,"description":3639},[3682,3684,3686,3688,3690],{"title":3683,"url":3595},"FastGPT 官网",{"title":3685,"url":3601},"FastGPT GitHub",{"title":3687,"url":325},"FastGPT 定价页",{"title":3689,"url":102},"FastGPT 2025 测评（南环 AI）",{"title":3691,"url":262},"FastGPT 部署教程（腾讯云）","tools\u002Fagent\u002Fplatform\u002Ffastgpt",[3694,3695,3696,3697,3698],"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",[1639,2672,2673,2674,3701,3702,3703],"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",1783006601069]