[{"data":1,"prerenderedAt":1784},["ShallowReactive",2],{"header-counts":3,"footer-counts":6,"compare-fastgpt-vs-ragflow":8,"compare-a-fastgpt":9,"compare-b-ragflow":1239},{"tools":4,"reviews":5},98,27,{"tools":4,"reviews":5,"playbooks":7,"news":5},22,null,{"id":10,"title":11,"alternatives":12,"api_compatible":17,"body":19,"category":1145,"chinese_friendly":415,"cover":1146,"description":1147,"domestic":1148,"extension":1149,"faq":8,"free":1148,"github":1104,"languages":1150,"lastVerified":8,"meta":1153,"models":1154,"navigation":411,"notSuitable":1161,"opensource":411,"path":1165,"pillar":1166,"platforms":1167,"priceTable":1171,"pricing":1196,"published":1197,"relatedPlaybooks":1198,"relatedReviews":1200,"score":1205,"self_host":411,"seo":1206,"seoTitle":1207,"slug":1208,"sources":1209,"stem":1220,"suitable":1221,"tagline":1227,"tags":1228,"updated":1236,"verdict":1237,"website":1097,"__hash__":1238},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Ffastgpt.md","FastGPT",[13,14,15,16],"agent\u002Fplatform\u002Fdify","agent\u002Fplatform\u002Fcoze","agent\u002Fplatform\u002Flangflow","agent\u002Fplatform\u002Fn8n",[18],"openai",{"type":20,"value":21,"toc":1127},"minimark",[22,27,65,77,80,85,88,104,107,133,137,144,177,184,188,248,255,258,261,264,271,317,325,329,360,364,501,504,507,511,519,647,662,665,835,846,881,887,890,966,970,973,993,996,1014,1017,1087,1090,1115,1123],[23,24,26],"h2",{"id":25},"tldr","TL;DR",[28,29,34,54],"div",{"className":30},[31,32,33],"card","p-5","my-4",[35,36,37,41,42,49,50,53],"p",{},[38,39,40],"strong",{},"一句话："," labring 团队开源的 LLM 知识库 RAG 平台，27k+ GitHub star（截至 2026-03 数据，",[43,44,48],"a",{"href":45,"rel":46},"https:\u002F\u002Fcloud.tencent.com\u002Fdeveloper\u002Farticle\u002F2632669",[47],"nofollow","腾讯云 2026-03 教程"," 引用），Apache 2.0 许可证可商用。",[38,51,52],{},"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。",[35,55,56,57,60,61,64],{},"最大价值在 ",[38,58,59],{},"国内企业知识库 + 私有部署"," 场景。代价是 ",[38,62,63],{},"配置门槛","：docker 基础 + 网络知识 + 一定运维能力。",[66,67,68],"blockquote",{},[35,69,70,71,76],{},"来源说明：本文基于 fastgpt.io 官方页面、github.com\u002Flabring\u002FFastGPT 仓库、",[43,72,75],{"href":73,"rel":74},"https:\u002F\u002Fwww.nanhuantech.com\u002Fzh\u002Fai-reviews\u002Ffastgpt-2025-review",[47],"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。",[23,78,79],{"id":79},"核心特性",[81,82,84],"h3",{"id":83},"知识库管理核心能力","知识库管理（核心能力）",[35,86,87],{},"支持文件类型：",[89,90,91,95,98,101],"ul",{},[92,93,94],"li",{},"文档：PDF \u002F Word \u002F Markdown \u002F TXT \u002F HTML",[92,96,97],{},"表格：Excel \u002F CSV",[92,99,100],{},"网页：URL 抓取 + 定时同步",[92,102,103],{},"API：通过接口推送内容",[35,105,106],{},"处理流程：上传 → 文本切分 → 向量化 → 存储 → 可用于问答。支持：",[89,108,109,115,121,127],{},[92,110,111,114],{},[38,112,113],{},"文件夹分组","：不同主题 \u002F 部门分类",[92,116,117,120],{},[38,118,119],{},"多种分块策略","：默认按段落 \u002F 按 token 数 \u002F 自定义",[92,122,123,126],{},[38,124,125],{},"批量导入","：脚本化大批量同步",[92,128,129,132],{},[38,130,131],{},"定时同步","：网页源自动更新",[81,134,136],{"id":135},"rag-流程编排最强卖点","RAG 流程编排（最强卖点）",[35,138,139,143],{},[43,140,142],{"href":73,"rel":141},[47],"南环 AI 2026 评测"," 总结的 FastGPT RAG 链路：",[145,146,147,153,159,165,171],"ol",{},[92,148,149,152],{},[38,150,151],{},"问题预处理","：改写 \u002F 扩展 \u002F 错词纠正（提升召回率）",[92,154,155,158],{},[38,156,157],{},"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",[92,160,161,164],{},[38,162,163],{},"重排序（Rerank）","：对初步检索结果二次排序，提升相关性",[92,166,167,170],{},[38,168,169],{},"上下文组装","：最优 chunk + 问题 → prompt",[92,172,173,176],{},[38,174,175],{},"答案生成","：调大模型基于检索结果回答 + 引用标注",[35,178,179,180,183],{},"每一步都可视化调参，这是 FastGPT 比 Coze \u002F Dify 在 ",[38,181,182],{},"纯知识库 QA 精度","上更高的原因。",[81,185,187],{"id":186},"多模型支持不绑定厂商","多模型支持（不绑定厂商）",[189,190,191,204],"table",{},[192,193,194],"thead",{},[195,196,197,201],"tr",{},[198,199,200],"th",{},"模型类别",[198,202,203],{},"支持",[205,206,207,216,224,232,240],"tbody",{},[195,208,209,213],{},[210,211,212],"td",{},"国产闭源",[210,214,215],{},"豆包 \u002F 通义千问 \u002F 文心一言 \u002F 智谱 GLM \u002F Moonshot Kimi \u002F MiniMax",[195,217,218,221],{},[210,219,220],{},"开源",[210,222,223],{},"LLaMA \u002F Qwen \u002F ChatGLM \u002F DeepSeek 等可自部署",[195,225,226,229],{},[210,227,228],{},"OpenAI 系",[210,230,231],{},"GPT-5 \u002F GPT-5 mini \u002F o3",[195,233,234,237],{},[210,235,236],{},"Claude 系",[210,238,239],{},"Sonnet 4 \u002F Opus 4 \u002F Haiku",[195,241,242,245],{},[210,243,244],{},"嵌入 \u002F 重排",[210,246,247],{},"BGE \u002F m3e \u002F OpenAI text-embedding-3",[35,249,250,251,254],{},"可以在 ",[38,252,253],{},"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。",[81,256,257],{"id":257},"工作流与高级编排",[35,259,260],{},"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。",[81,262,263],{"id":263},"多向量库选择",[35,265,266,270],{},[43,267,269],{"href":45,"rel":268},[47],"腾讯云教程"," 公开的 4 种向量后端：",[189,272,273,283],{},[192,274,275],{},[195,276,277,280],{},[198,278,279],{},"后端",[198,281,282],{},"适用",[205,284,285,293,301,309],{},[195,286,287,290],{},[210,288,289],{},"PgVector",[210,291,292],{},"5000 万索引以下，新手 \u002F 小规模",[195,294,295,298],{},[210,296,297],{},"Milvus",[210,299,300],{},"亿级以上，高性能",[195,302,303,306],{},[210,304,305],{},"Zilliz Cloud",[210,307,308],{},"Milvus 全托管 SaaS",[195,310,311,314],{},[210,312,313],{},"SeekDB \u002F OceanBase",[210,315,316],{},"企业级国产化",[35,318,319,320,324],{},"部署时选对应 ",[321,322,323],"code",{},"docker-compose.{pgvector|milvus|...}.yml","。",[81,326,328],{"id":327},"api-与-mcp","API 与 MCP",[89,330,331,337,343,354],{},[92,332,333,336],{},[38,334,335],{},"对话 API","：流式 \u002F 非流式 HTTP，OpenAI 兼容",[92,338,339,342],{},[38,340,341],{},"知识库检索 API","：单独调检索（不走生成）做 hybrid pipeline",[92,344,345,348,349,353],{},[38,346,347],{},"MCP Server","：3005 端口暴露 MCP SSE 服务，可被 ",[43,350,352],{"href":351},"\u002Fcoding\u002Fcli\u002Fclaude-code.html","Claude Code"," 等客户端直接接入",[92,355,356,359],{},[38,357,358],{},"Webhook","：回调通知",[23,361,363],{"id":362},"部署-10-分钟docker","部署 10 分钟（Docker）",[365,366,371],"pre",{"className":367,"code":368,"language":369,"meta":370,"style":370},"language-bash shiki shiki-themes github-light github-dark","# 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","bash","",[321,372,373,382,396,406,413,419,433,438,444,452,461,466,472,490,495],{"__ignoreMap":370},[374,375,378],"span",{"class":376,"line":377},"line",1,[374,379,381],{"class":380},"sJ8bj","# 克隆代码\n",[374,383,385,389,393],{"class":376,"line":384},2,[374,386,388],{"class":387},"sScJk","git",[374,390,392],{"class":391},"sZZnC"," clone",[374,394,395],{"class":391}," https:\u002F\u002Fgithub.com\u002Flabring\u002FFastGPT.git\n",[374,397,399,403],{"class":376,"line":398},3,[374,400,402],{"class":401},"sj4cs","cd",[374,404,405],{"class":391}," FastGPT\n",[374,407,409],{"class":376,"line":408},4,[374,410,412],{"emptyLinePlaceholder":411},true,"\n",[374,414,416],{"class":376,"line":415},5,[374,417,418],{"class":380},"# 切到最新稳定版（参考 GitHub releases）\n",[374,420,422,424,427,430],{"class":376,"line":421},6,[374,423,388],{"class":387},[374,425,426],{"class":391}," switch",[374,428,429],{"class":401}," -c",[374,431,432],{"class":401}," 4.14.7.2\n",[374,434,436],{"class":376,"line":435},7,[374,437,412],{"emptyLinePlaceholder":411},[374,439,441],{"class":376,"line":440},8,[374,442,443],{"class":380},"# 选向量库版本（个人 \u002F 小规模选 pg）\n",[374,445,447,449],{"class":376,"line":446},9,[374,448,402],{"class":401},[374,450,451],{"class":391}," deploy\u002Fdocker\u002Fcn\n",[374,453,455,458],{"class":376,"line":454},10,[374,456,457],{"class":387},"wget",[374,459,460],{"class":391}," https:\u002F\u002Fdoc.fastgpt.cn\u002Fdeploy\u002Fconfig\u002Fconfig.json\n",[374,462,464],{"class":376,"line":463},11,[374,465,412],{"emptyLinePlaceholder":411},[374,467,469],{"class":376,"line":468},12,[374,470,471],{"class":380},"# 启动\n",[374,473,475,478,481,484,487],{"class":376,"line":474},13,[374,476,477],{"class":387},"docker-compose",[374,479,480],{"class":401}," -f",[374,482,483],{"class":391}," docker-compose.pg.yml",[374,485,486],{"class":391}," up",[374,488,489],{"class":401}," -d\n",[374,491,493],{"class":376,"line":492},14,[374,494,412],{"emptyLinePlaceholder":411},[374,496,498],{"class":376,"line":497},15,[374,499,500],{"class":380},"# 访问 http:\u002F\u002F\u003Cip>:3000，默认账号 root \u002F 1234\n",[35,502,503],{},"最低配置：2C4G + 20GB 硬盘 + Docker 28+ + Docker Compose 2.20+。",[35,505,506],{},"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。",[23,508,510],{"id":509},"云版-vs-自托管对比","云版 vs 自托管对比",[35,512,513,518],{},[43,514,517],{"href":515,"rel":516},"https:\u002F\u002Ffastgpt.io\u002Fzh\u002Fprice",[47],"fastgpt.io 官方定价"," 公开数据：",[189,520,521,549],{},[192,522,523],{},[195,524,525,528,531,534,537,540,543,546],{},[198,526,527],{},"套餐",[198,529,530],{},"价格",[198,532,533],{},"AI 积分",[198,535,536],{},"知识库索引",[198,538,539],{},"团队",[198,541,542],{},"Agent",[198,544,545],{},"知识库",[198,547,548],{},"QPM",[205,550,551,577,602,626],{},[195,552,553,556,559,562,565,568,571,574],{},[210,554,555],{},"免费",[210,557,558],{},"¥0",[210,560,561],{},"100",[210,563,564],{},"600",[210,566,567],{},"1",[210,569,570],{},"10",[210,572,573],{},"3",[210,575,576],{},"30",[195,578,579,582,585,588,591,594,597,599],{},[210,580,581],{},"基础",[210,583,584],{},"¥99\u002F月",[210,586,587],{},"4000",[210,589,590],{},"6000",[210,592,593],{},"5",[210,595,596],{},"50",[210,598,576],{},[210,600,601],{},"300",[195,603,604,607,610,613,616,618,621,623],{},[210,605,606],{},"高级",[210,608,609],{},"¥599\u002F月",[210,611,612],{},"25000",[210,614,615],{},"36000",[210,617,596],{},[210,619,620],{},"200",[210,622,561],{},[210,624,625],{},"1500",[195,627,628,631,634,637,639,641,643,645],{},[210,629,630],{},"定制",[210,632,633],{},"议价",[210,635,636],{},"弹性",[210,638,636],{},[210,640,636],{},[210,642,636],{},[210,644,636],{},[210,646,636],{},[35,648,649,652,653,656,657,661],{},[38,650,651],{},"云版适合","：不想运维、量小、要快速上线\n",[38,654,655],{},"自托管适合","：量大（10 万+ 日问答）、数据敏感、要深度定制——按 ",[43,658,660],{"href":73,"rel":659},[47],"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\"",[23,663,664],{"id":664},"与同类怎么选",[189,666,667,694],{},[192,668,669],{},[195,670,671,674,676,682,688,691],{},[198,672,673],{},"维度",[198,675,11],{},[198,677,678],{},[43,679,681],{"href":680},"\u002Fagent\u002Fplatform\u002Fdify.html","Dify",[198,683,684],{},[43,685,687],{"href":686},"\u002Fagent\u002Fplatform\u002Fcoze.html","Coze",[198,689,690],{},"RAGFlow",[198,692,693],{},"AnythingLLM",[205,695,696,716,733,752,770,786,803,818],{},[195,697,698,701,704,707,710,713],{},[210,699,700],{},"核心定位",[210,702,703],{},"知识库 QA",[210,705,706],{},"综合 LLMOps",[210,708,709],{},"Bot + 工作流",[210,711,712],{},"文档解析+RAG",[210,714,715],{},"桌面级 KB",[195,717,718,720,723,725,728,730],{},[210,719,220],{},[210,721,722],{},"✅ Apache 2.0",[210,724,722],{},[210,726,727],{},"❌",[210,729,722],{},[210,731,732],{},"✅ MIT",[195,734,735,738,741,744,747,750],{},[210,736,737],{},"私有部署",[210,739,740],{},"★★★★★ docker",[210,742,743],{},"★★★★★",[210,745,746],{},"⚠️ 仅企业版",[210,748,749],{},"★★★★☆",[210,751,743],{},[195,753,754,757,760,762,765,768],{},[210,755,756],{},"RAG 深度",[210,758,759],{},"★★★★★ 最细",[210,761,749],{},[210,763,764],{},"★★★☆☆",[210,766,767],{},"★★★★★ 文档解析最强",[210,769,764],{},[195,771,772,775,777,779,781,783],{},[210,773,774],{},"工作流",[210,776,749],{},[210,778,743],{},[210,780,749],{},[210,782,764],{},[210,784,785],{},"★★☆☆☆",[195,787,788,791,794,796,799,801],{},[210,789,790],{},"上手",[210,792,793],{},"★★★☆☆ 需 docker",[210,795,749],{},[210,797,798],{},"★★★★★ 最简单",[210,800,764],{},[210,802,749],{},[195,804,805,808,810,812,814,816],{},[210,806,807],{},"中文优化",[210,809,743],{},[210,811,749],{},[210,813,743],{},[210,815,749],{},[210,817,764],{},[195,819,820,823,826,828,830,833],{},[210,821,822],{},"多平台发布",[210,824,825],{},"⚠️ API 为主",[210,827,749],{},[210,829,743],{},[210,831,832],{},"⚠️",[210,834,832],{},[35,836,837,840,841,845],{},[38,838,839],{},"怎么选","（综合 ",[43,842,844],{"href":73,"rel":843},[47],"南环 AI 评测","）：",[89,847,848,854,862,869,875],{},[92,849,850,853],{},[38,851,852],{},"核心需求是 RAG 精度"," → FastGPT",[92,855,856,859,860],{},[38,857,858],{},"需要丰富插件 + 复杂工作流 + 多平台发布"," → ",[43,861,681],{"href":680},[92,863,864,859,867],{},[38,865,866],{},"零代码、快速发布到飞书 \u002F 微信",[43,868,687],{"href":686},[92,870,871,874],{},[38,872,873],{},"文档解析（含 OCR \u002F 表格 \u002F 公式）是瓶颈"," → RAGFlow",[92,876,877,880],{},[38,878,879],{},"桌面 \u002F 单机使用"," → AnythingLLM",[35,882,883,886],{},[38,884,885],{},"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。",[23,888,889],{"id":889},"避坑清单",[89,891,892,906,915,921,931,943,949,955],{},[92,893,894,897,898,901,902,905],{},[38,895,896],{},"docker-compose 镜像 tag 不一致","：",[43,899,269],{"href":45,"rel":900},[47]," 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 ",[321,903,904],{},"image:"," 行为正确版本即可",[92,907,908,911,912,914],{},[38,909,910],{},"3000 端口冲突","：默认占用 3000（主服务）\u002F 9000（S3 \u002F MinIO）\u002F 3005（MCP）；改 ",[321,913,477],{}," 的 ports 映射端口",[92,916,917,920],{},[38,918,919],{},"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",[92,922,923,926,927,930],{},[38,924,925],{},"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 ",[43,928,660],{"href":73,"rel":929},[47],"：\"新手 \u002F 小规模 PgVector，中大规模 Milvus，企业 \u002F 国产 OceanBase\"",[92,932,933,897,936,939,940],{},[38,934,935],{},"MinIO 默认密码",[321,937,938],{},"minioadmin\u002Fminioadmin","，",[38,941,942],{},"部署到公网前必须改",[92,944,945,948],{},[38,946,947],{},"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\"",[92,950,951,954],{},[38,952,953],{},"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",[92,956,957,960,961,965],{},[38,958,959],{},"云版 AI 积分会过期","：未用完不能跨月累积（按 ",[43,962,964],{"href":515,"rel":963},[47],"fastgpt.io 定价 FAQ","）",[23,967,969],{"id":968},"适合-不适合","适合 \u002F 不适合",[35,971,972],{},"✅ 适合：",[89,974,975,978,981,984,987,990],{},[92,976,977],{},"企业内部知识库（员工手册 \u002F 制度 \u002F 流程）",[92,979,980],{},"产品 FAQ \u002F 用户手册问答",[92,982,983],{},"医疗 \u002F 法律 \u002F 金融垂直领域知识系统",[92,985,986],{},"数据严格不出网 + Apache 2.0 商用",[92,988,989],{},"有 docker 运维基础的技术团队",[92,991,992],{},"需要把 RAG 当后端服务的开发者（API 接入业务系统）",[35,994,995],{},"❌ 不适合：",[89,997,998,1003,1008,1011],{},[92,999,1000,1001,965],{},"完全非技术用户（去 ",[43,1002,687],{"href":686},[92,1004,1005,1006,965],{},"主要需求是工作流 + 插件集成（去 ",[43,1007,681],{"href":680},[92,1009,1010],{},"文档解析 \u002F OCR 是首要痛点（RAGFlow）",[92,1012,1013],{},"不想自己运维 + 量很小（FastGPT 云免费版起步即可）",[23,1015,1016],{"id":1016},"相关阅读",[89,1018,1019,1032,1053,1076],{},[92,1020,1021,1022,1024,1025,1027,1028],{},"同类对比：",[43,1023,681],{"href":680}," \u002F ",[43,1026,687],{"href":686}," \u002F RAGFlow \u002F AnythingLLM \u002F ",[43,1029,1031],{"href":1030},"\u002Fagent\u002Fplatform\u002Fn8n.html","n8n",[92,1033,1034,1035,1024,1039,1024,1043,1024,1046,1024,1049],{},"概念：",[43,1036,1038],{"href":1037},"\u002Fwiki\u002Frag.html","RAG",[43,1040,1042],{"href":1041},"\u002Fwiki\u002Fembedding.html","Embedding",[43,1044,1045],{"href":1041},"Vector Database",[43,1047,1048],{"href":1037},"Reranker",[43,1050,1052],{"href":1051},"\u002Fwiki\u002Fai-agent.html","AI Agent",[92,1054,1055,1056,1024,1060,1024,1064,1024,1068,1024,1072],{},"模型：",[43,1057,1059],{"href":1058},"\u002Fmodels\u002Fdeepseek-v3.html","DeepSeek-V3",[43,1061,1063],{"href":1062},"\u002Fmodels\u002Fqwen-3.html","Qwen3",[43,1065,1067],{"href":1066},"\u002Fmodels\u002Fglm-5.2.html","GLM-5.2",[43,1069,1071],{"href":1070},"\u002Fmodels\u002Fkimi-k2.html","Kimi K2",[43,1073,1075],{"href":1074},"\u002Fmodels\u002Fdoubao-1-5-pro.html","豆包 Doubao",[92,1077,1078,1079,1024,1083],{},"进阶：",[43,1080,1082],{"href":1081},"\u002Fwiki\u002Fcontext-engineering.html","Context Engineering",[43,1084,1086],{"href":1085},"\u002Fwiki\u002Fprompt-engineering.html","Prompt Engineering",[23,1088,1089],{"id":1089},"来源",[89,1091,1092,1099,1106,1112],{},[92,1093,1094,1095],{},"官网：",[43,1096,1097],{"href":1097,"rel":1098},"https:\u002F\u002Ffastgpt.io",[47],[92,1100,1101,1102],{},"GitHub：",[43,1103,1104],{"href":1104,"rel":1105},"https:\u002F\u002Fgithub.com\u002Flabring\u002FFastGPT",[47],[92,1107,1108,1109],{},"定价：",[43,1110,515],{"href":515,"rel":1111},[47],[92,1113,1114],{},"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",[35,1116,1117,1118,1122],{},"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 ",[43,1119,1121],{"href":1120},"mailto:hello@aiho.net","反馈邮箱"," 反馈。",[1124,1125,1126],"style",{},"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":370,"searchDepth":398,"depth":398,"links":1128},[1129,1130,1138,1139,1140,1141,1142,1143,1144],{"id":25,"depth":384,"text":26},{"id":79,"depth":384,"text":79,"children":1131},[1132,1133,1134,1135,1136,1137],{"id":83,"depth":398,"text":84},{"id":135,"depth":398,"text":136},{"id":186,"depth":398,"text":187},{"id":257,"depth":398,"text":257},{"id":263,"depth":398,"text":263},{"id":327,"depth":398,"text":328},{"id":362,"depth":384,"text":363},{"id":509,"depth":384,"text":510},{"id":664,"depth":384,"text":664},{"id":889,"depth":384,"text":889},{"id":968,"depth":384,"text":969},{"id":1016,"depth":384,"text":1016},{"id":1089,"depth":384,"text":1089},"platform","\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。",false,"md",[1151,1152],"zh","en",{},[1155,1156,1157,1158,1159,1160],"deepseek-v3","qwen-max","doubao-pro","gpt-4o","claude-sonnet-4","kimi",[1162,1163,1164],"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","agent",[1168,1169,1170],"windows","macos","linux",[1172,1177,1182,1187,1192],{"plan":1173,"price":555,"limit":1174,"cn_pay":1175,"note":1176},"Self-host 开源","全功能 + 全数据本地","—","Apache 2.0 可商用",{"plan":1178,"price":1179,"limit":1180,"cn_pay":1175,"note":1181},"云免费版","¥0\u002F月","100 AI 积分 + 600 索引 + 3 知识库","试水",{"plan":1183,"price":584,"limit":1184,"cn_pay":1185,"note":1186},"云基础版","4000 积分 + 6000 索引 + 50 Agent","✅ 微信\u002F支付宝","中小团队 SaaS",{"plan":1188,"price":609,"limit":1189,"cn_pay":1190,"note":1191},"云高级版","25000 积分 + 36000 索引 + 50 成员 + 200 Agent + 1500 QPM","✅","企业级生产",{"plan":1193,"price":633,"limit":1194,"cn_pay":1190,"note":1195},"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","2026-06-18",[1199],"onboarding\u002Ffastgpt-getting-started",[1201,1202,1203,1204],"fastgpt-deep-review","coze-deep-review","coze-vs-dify","dify-deep-review",{"power":408,"ux":408,"price":415,"cn_support":415,"stability":408},{"title":11,"description":1147},"FastGPT 评测 2026：开源知识库问答平台，AI 工作流引擎，对比 Dify","agent\u002Fplatform\u002Ffastgpt",[1210,1212,1214,1216,1218],{"title":1211,"url":1097},"FastGPT 官网",{"title":1213,"url":1104},"FastGPT GitHub",{"title":1215,"url":515},"FastGPT 定价页",{"title":1217,"url":73},"FastGPT 2025 测评（南环 AI）",{"title":1219,"url":45},"FastGPT 部署教程（腾讯云）","tools\u002Fagent\u002Fplatform\u002Ffastgpt",[1222,1223,1224,1225,1226],"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",[1229,1230,1231,1232,1233,1234,1235],"agent-platform","opensource","self-host","rag","china","knowledge-base","labring","2026-06-24","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。","NAay3javdz1FV9ZaZVXCpoCFdZlJu5B4E0Y3oSzmYLk",{"id":1240,"title":690,"alternatives":1241,"api_compatible":8,"body":1243,"category":1145,"chinese_friendly":408,"cover":1761,"description":1762,"domestic":1148,"extension":1149,"faq":8,"free":1148,"github":1745,"languages":1763,"lastVerified":1764,"meta":1765,"models":8,"navigation":411,"notSuitable":8,"opensource":411,"path":1766,"pillar":1166,"platforms":1767,"priceTable":8,"pricing":1769,"published":1770,"relatedPlaybooks":8,"relatedReviews":8,"score":1771,"self_host":1148,"seo":1772,"seoTitle":1773,"slug":1774,"sources":1775,"stem":1778,"suitable":8,"tagline":1779,"tags":1780,"updated":1764,"verdict":1782,"website":1738,"__hash__":1783},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Fragflow.md",[13,1208,1242],"agent\u002Fplatform\u002Fanythingllm",{"type":20,"value":1244,"toc":1748},[1245,1247,1250,1253,1256,1312,1314,1363,1368,1372,1380,1400,1405,1428,1430,1460,1463,1596,1599,1647,1649,1681,1685,1691,1697,1703,1709,1711,1725,1727,1732],[23,1246,26],{"id":25},[35,1248,1249],{},"RAGFlow 是 InfiniFlow（中国团队）出品的开源 RAG 引擎（Apache 2.0），核心卖点是深度文档解析 + 高召回率切片 + 引用溯源。支持 PDF \u002F Word \u002F Excel \u002F PPT \u002F 图片 \u002F 扫描件，内置 OCR + 版面分析 + 表格识别，切片质量远超通用 RAG 方案。Docker 自托管 + Cloud 云端，中文支持好（文档 \u002F UI \u002F 社区）。",[35,1251,1252],{},"适合：需要精准文档问答的企业知识库、复杂文档（表格 \u002F 图文 \u002F 扫描件）场景、中文 RAG 需求、对召回率要求高的业务。不适合：需要复杂 Agent 编排（用 Dify）、资源有限的小服务器、需要精美 UI 的 C 端产品。",[23,1254,1255],{"id":1255},"核心能力",[89,1257,1258,1264,1270,1276,1282,1288,1294,1300,1306],{},[92,1259,1260,1263],{},[38,1261,1262],{},"深度文档解析","：PDF \u002F Word \u002F Excel \u002F PPT \u002F 图片 \u002F 扫描件，版面分析 + 表格识别",[92,1265,1266,1269],{},[38,1267,1268],{},"OCR 引擎","：内置 PaddleOCR \u002F DeepDOC，支持中英文扫描件识别",[92,1271,1272,1275],{},[38,1273,1274],{},"智能切片","：基于版面分析的语义切片，保留段落 \u002F 表格 \u002F 标题结构",[92,1277,1278,1281],{},[38,1279,1280],{},"高召回率","：混合检索（全文 + 向量）+ 重排序（Rerank），召回精度高",[92,1283,1284,1287],{},[38,1285,1286],{},"引用溯源","：回答标注来源文档 + 页码 + 原文片段，可验证",[92,1289,1290,1293],{},[38,1291,1292],{},"多模型接入","：OpenAI \u002F Claude \u002F Ollama \u002F 通义千问 \u002F 智谱 \u002F 月之暗面",[92,1295,1296,1299],{},[38,1297,1298],{},"多向量数据库","：Elasticsearch \u002F Infinity（自研）\u002F Chroma",[92,1301,1302,1305],{},[38,1303,1304],{},"知识库管理","：多知识库 + 文档分类 + 解析状态监控",[92,1307,1308,1311],{},[38,1309,1310],{},"API 接口","：完整 REST API + SDK，可集成到外部系统",[23,1313,530],{"id":530},[189,1315,1316,1328],{},[192,1317,1318],{},[195,1319,1320,1323,1325],{},[198,1321,1322],{},"方案",[198,1324,530],{},[198,1326,1327],{},"核心功能",[205,1329,1330,1341,1352],{},[195,1331,1332,1335,1338],{},[210,1333,1334],{},"开源版",[210,1336,1337],{},"$0",[210,1339,1340],{},"完整功能，Apache 2.0，自托管",[195,1342,1343,1346,1349],{},[210,1344,1345],{},"Cloud",[210,1347,1348],{},"按量付费",[210,1350,1351],{},"托管服务，免运维",[195,1353,1354,1357,1360],{},[210,1355,1356],{},"Enterprise",[210,1358,1359],{},"联系销售",[210,1361,1362],{},"私有部署 + 技术支持 + 定制",[66,1364,1365],{},[35,1366,1367],{},"价格信息基于 2026-07 官网，可能调整。",[23,1369,1371],{"id":1370},"体验与评测资料整理","体验与评测（资料整理）",[66,1373,1374],{},[35,1375,1376,1377],{},"说明：本节基于官方文档与公开评测整理，非本站独立实测环境，具体数据请以官方为准。\n",[38,1378,1379],{},"亮点：",[89,1381,1382,1385,1388,1391,1394,1397],{},[92,1383,1384],{},"文档解析质量在开源 RAG 中最强——复杂表格、多栏排版、图文混排都能正确识别",[92,1386,1387],{},"扫描件 OCR 效果好，中文印刷体识别准确率高",[92,1389,1390],{},"引用溯源到页码 + 原文片段，回答可信度高",[92,1392,1393],{},"混合检索 + Rerank 召回精度明显优于纯向量检索",[92,1395,1396],{},"中国团队出品，中文文档和社区支持好，Issue 响应快",[92,1398,1399],{},"支持通义千问 \u002F 智谱 \u002F 月之暗面等国产模型，国内场景适配好",[35,1401,1402],{},[38,1403,1404],{},"踩坑：",[89,1406,1407,1410,1413,1416,1419,1422,1425],{},[92,1408,1409],{},"资源消耗大——Elasticsearch + Redis + MinIO + RAGFlow 本身，至少 16GB 内存",[92,1411,1412],{},"部署较重，Docker Compose 起来 5+ 容器，配置复杂",[92,1414,1415],{},"大文件解析慢——100 页 PDF 解析 + 切片可能 5-10 分钟",[92,1417,1418],{},"UI 仍有粗糙处，文档管理界面交互不够流畅",[92,1420,1421],{},"Agent 能力弱——RAG 问答是强项，复杂工具调用 \u002F 多步推理不如 Dify",[92,1423,1424],{},"解析失败的重试机制不完善，偶尔卡在 parsing 状态",[92,1426,1427],{},"版本迭代快，升级需注意数据迁移",[23,1429,790],{"id":790},[145,1431,1432,1435,1441,1447,1454,1457],{},[92,1433,1434],{},"系统准备：确保 16GB+ 内存 + Docker + Docker Compose",[92,1436,1437,1438],{},"克隆仓库：",[321,1439,1440],{},"git clone https:\u002F\u002Fgithub.com\u002Finfiniflow\u002Fragflow.git",[92,1442,1443,1444],{},"启动服务：",[321,1445,1446],{},"cd ragflow\u002Fdocker && docker compose up -d",[92,1448,1449,1450,1453],{},"访问 ",[321,1451,1452],{},"http:\u002F\u002Flocalhost:80","，注册管理员账号",[92,1455,1456],{},"配置模型：Settings → Model Providers 添加 LLM + Embedding + Rerank",[92,1458,1459],{},"创建知识库 → 上传文档 → 等待解析完成 → 开始问答",[23,1461,1462],{"id":1462},"对比",[189,1464,1465,1479],{},[192,1466,1467],{},[195,1468,1469,1471,1473,1475,1477],{},[198,1470,673],{},[198,1472,690],{},[198,1474,681],{},[198,1476,11],{},[198,1478,693],{},[205,1480,1481,1498,1511,1526,1540,1553,1566,1580],{},[195,1482,1483,1486,1489,1492,1495],{},[210,1484,1485],{},"文档解析",[210,1487,1488],{},"✅ 最强",[210,1490,1491],{},"中",[210,1493,1494],{},"强",[210,1496,1497],{},"弱",[195,1499,1500,1503,1505,1507,1509],{},[210,1501,1502],{},"表格识别",[210,1504,1190],{},[210,1506,727],{},[210,1508,1190],{},[210,1510,727],{},[195,1512,1513,1516,1519,1521,1524],{},[210,1514,1515],{},"OCR",[210,1517,1518],{},"✅ 内置",[210,1520,727],{},[210,1522,1523],{},"需配置",[210,1525,1523],{},[195,1527,1528,1531,1534,1536,1538],{},[210,1529,1530],{},"召回精度",[210,1532,1533],{},"高",[210,1535,1533],{},[210,1537,1533],{},[210,1539,1491],{},[195,1541,1542,1544,1547,1549,1551],{},[210,1543,1286],{},[210,1545,1546],{},"✅ 页码+片段",[210,1548,1190],{},[210,1550,1190],{},[210,1552,1190],{},[195,1554,1555,1558,1560,1562,1564],{},[210,1556,1557],{},"Agent 能力",[210,1559,1497],{},[210,1561,1494],{},[210,1563,1491],{},[210,1565,581],{},[195,1567,1568,1571,1573,1575,1577],{},[210,1569,1570],{},"资源消耗",[210,1572,1533],{},[210,1574,1491],{},[210,1576,1491],{},[210,1578,1579],{},"低",[195,1581,1582,1585,1588,1591,1593],{},[210,1583,1584],{},"中文支持",[210,1586,1587],{},"✅ 优秀",[210,1589,1590],{},"好",[210,1592,1590],{},[210,1594,1595],{},"一般",[23,1597,1598],{"id":1598},"避坑",[89,1600,1601,1607,1617,1623,1629,1635,1641],{},[92,1602,1603,1606],{},[38,1604,1605],{},"资源一定要够","：低于 16GB 内存别部署，ES + Redis + MinIO 都吃内存",[92,1608,1609,1612,1613,1616],{},[38,1610,1611],{},"Elasticsearch 配置","：默认 JVM 堆偏小，大知识库调 ",[321,1614,1615],{},"ES_JAVA_OPTS"," 到 4-8GB",[92,1618,1619,1622],{},[38,1620,1621],{},"大文件拆分上传","：超过 100 页的 PDF 拆成小文件，解析更稳定",[92,1624,1625,1628],{},[38,1626,1627],{},"解析失败检查格式","：加密 PDF \u002F 损坏文件会卡住，上传前检查",[92,1630,1631,1634],{},[38,1632,1633],{},"Rerank 模型别省","：召回精度提升的关键，用 bge-reranker 或 Cohere Rerank",[92,1636,1637,1640],{},[38,1638,1639],{},"不要当 Agent 平台用","：RAG 问答是核心，复杂工具调用上 Dify",[92,1642,1643,1646],{},[38,1644,1645],{},"定期备份","：ES 数据 + MinIO 文件，升级前完整快照",[23,1648,969],{"id":968},[89,1650,1651,1654,1657,1660,1663,1666,1669,1672,1675,1678],{},[92,1652,1653],{},"✅ 需要精准文档问答的企业知识库",[92,1655,1656],{},"✅ 复杂文档（表格 \u002F 图文 \u002F 扫描件）RAG 场景",[92,1658,1659],{},"✅ 中文 RAG 需求（国产模型 + 中文 OCR）",[92,1661,1662],{},"✅ 对召回率和引用溯源要求高的业务",[92,1664,1665],{},"✅ 有运维能力的团队私有化部署",[92,1667,1668],{},"❌ 需要复杂 Agent 编排（用 Dify）",[92,1670,1671],{},"❌ 资源有限的小服务器（至少 16GB 内存）",[92,1673,1674],{},"❌ 需要精美 C 端 UI 的产品",[92,1676,1677],{},"❌ 无运维能力的团队（用 Cloud 版或 FastGPT）",[92,1679,1680],{},"❌ 纯英文简单文档场景（AnythingLLM 更轻量）",[23,1682,1684],{"id":1683},"faq","FAQ",[35,1686,1687,1690],{},[38,1688,1689],{},"Q: RAGFlow 和 Dify 怎么选？","\nA: RAGFlow 专注 RAG——文档解析 + 检索精度 + 引用溯源是核心强项，适合文档密集型知识库。Dify 是完整 AI 应用平台——工作流 + Agent + RAG + API 管理，功能更全。纯文档问答选 RAGFlow，构建 AI 应用选 Dify，两者也可配合使用。",[35,1692,1693,1696],{},[38,1694,1695],{},"Q: 部署需要什么配置？","\nA: 最低 16GB 内存 + 4 核 CPU + 50GB 磁盘。生产环境建议 32GB 内存 + 8 核 + SSD。Elasticsearch 是内存大户，知识库文档量大时 ES JVM 堆需 8GB+。如果资源有限，考虑用 Infinity（RAGFlow 自研向量库）替代 ES。",[35,1698,1699,1702],{},[38,1700,1701],{},"Q: 支持中文 OCR 吗？","\nA: 支持。内置 PaddleOCR + DeepDOC 引擎，中文印刷体识别准确率高。手写体效果一般，复杂背景的扫描件建议预处理（去噪 \u002F 矫正）后再上传。OCR 默认开启，可在解析模板中配置。",[35,1704,1705,1708],{},[38,1706,1707],{},"Q: 和 FastGPT 比 RAG 精度如何？","\nA: 两者 RAG 精度都属第一梯队。RAGFlow 的优势在文档解析——复杂表格、多栏版面、图文混排的识别更准确，切片质量更高。FastGPT 的优势在工作流编排和知识库管理 UI 更成熟。文档解析要求高选 RAGFlow，流程管理要求高选 FastGPT。",[23,1710,1016],{"id":1016},[35,1712,1713,1716,1717,1716,1721],{},[43,1714,693],{"href":1715},"\u002Fagent\u002Fplatform\u002Fanythingllm.html"," · ",[43,1718,1720],{"href":1719},"\u002Fagent\u002Fplatform\u002Fflowise.html","Flowise",[43,1722,1724],{"href":1723},"\u002Fcoding\u002Fapi\u002Flangfuse.html","Langfuse",[23,1726,1089],{"id":1089},[66,1728,1729],{},[35,1730,1731],{},"本文的价格、版本号与性能数据均参考以下官方渠道整理，可能随时间变动，请以官方实时信息为准。",[89,1733,1734,1741],{},[92,1735,1736],{},[43,1737,1740],{"href":1738,"rel":1739},"https:\u002F\u002Fragflow.io",[47],"官网",[92,1742,1743],{},[43,1744,1747],{"href":1745,"rel":1746},"https:\u002F\u002Fgithub.com\u002Finfiniflow\u002Fragflow",[47],"GitHub",{"title":370,"searchDepth":398,"depth":398,"links":1749},[1750,1751,1752,1753,1754,1755,1756,1757,1758,1759,1760],{"id":25,"depth":384,"text":26},{"id":1255,"depth":384,"text":1255},{"id":530,"depth":384,"text":530},{"id":1370,"depth":384,"text":1371},{"id":790,"depth":384,"text":790},{"id":1462,"depth":384,"text":1462},{"id":1598,"depth":384,"text":1598},{"id":968,"depth":384,"text":969},{"id":1683,"depth":384,"text":1684},{"id":1016,"depth":384,"text":1016},{"id":1089,"depth":384,"text":1089},"\u002Fimg\u002Ftools\u002Fragflow.webp","RAGFlow 真实评测：InfiniFlow 出品的开源 RAG 引擎（Apache 2.0 协议），深度文档解析（PDF\u002FWord\u002FExcel\u002F图片）+ 高召回率切片 + 引用溯源。支持 Docker 自托管，适合需要精准文档问答和知识库检索的企业场景。",[1152,1151],"2026-07-30",{},"\u002Ftools\u002Fagent\u002Fplatform\u002Fragflow",[1170,1768],"docker","Free \u002F 开源（Apache 2.0）\u002F Cloud","2026-07-05",{"power":408,"ux":398,"price":415,"cn_support":408,"stability":398},{"title":690,"description":1762},"RAGFlow - 开源 RAG 引擎评测与部署 | AIHO","agent\u002Fplatform\u002Fragflow",[1776,1777],{"title":1740,"url":1738},{"title":1747,"url":1745},"tools\u002Fagent\u002Fplatform\u002Fragflow","开源 RAG 引擎，深度文档解析 + 高召回率",[1229,1230,1232,1781,1231],"document-parsing","需要精准文档解析和高召回率 RAG 的企业场景首选，深度文档解析（复杂表格\u002F版面\u002FOCR）+ 引用溯源能力在开源 RAG 引擎中最强，中国团队出品中文支持好，但部署资源要求高、Agent 能力弱、UI 仍需打磨。","bLpjRG4MMrBFYyeFSz4XzLMwVql0SyGD3sB3Tr5g38g",1785428441182]