[{"data":1,"prerenderedAt":3315},["ShallowReactive",2],{"header-counts":3,"footer-counts":6,"alt-main-ragflow":8,"alt-list-ragflow":613},{"tools":4,"reviews":5},98,27,{"tools":4,"reviews":5,"playbooks":7,"news":5},23,{"id":9,"title":10,"alternatives":11,"api_compatible":15,"body":22,"category":575,"chinese_friendly":576,"cover":577,"description":578,"domestic":579,"extension":580,"faq":581,"free":579,"github":556,"languages":582,"lastVerified":585,"meta":586,"models":581,"navigation":579,"notSuitable":581,"opensource":579,"path":587,"pillar":588,"platforms":589,"priceTable":581,"pricing":592,"published":593,"relatedPlaybooks":581,"relatedReviews":581,"score":594,"self_host":596,"seo":597,"seoTitle":598,"slug":599,"sources":600,"stem":603,"suitable":581,"tagline":604,"tags":605,"updated":585,"verdict":611,"website":548,"__hash__":612},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Fragflow.md","RAGFlow",[12,13,14],"agent\u002Fplatform\u002Fdify","agent\u002Fplatform\u002Ffastgpt","agent\u002Fplatform\u002Fanythingllm",[16,17,18,19,20,21],"OpenAI","Anthropic","百度文心","阿里通义","Moonshot Kimi","DeepSeek",{"type":23,"value":24,"toc":559},"minimark",[25,30,34,37,40,99,102,157,163,167,175,195,200,223,226,258,261,401,404,452,456,488,492,498,504,510,516,519,534,537,542],[26,27,29],"h2",{"id":28},"tldr","TL;DR",[31,32,33],"p",{},"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）。",[31,35,36],{},"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。",[26,38,39],{"id":39},"核心能力",[41,42,43,51,57,63,69,75,81,87,93],"ul",{},[44,45,46,50],"li",{},[47,48,49],"strong",{},"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",[44,52,53,56],{},[47,54,55],{},"OCR 引擎","：内置 PaddleOCR \u002F DeepDOC，支持中英文扫描件识别",[44,58,59,62],{},[47,60,61],{},"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",[44,64,65,68],{},[47,66,67],{},"高召回率","：混合检索（全文 + 向量）+ 重排序（Rerank），召回精度高",[44,70,71,74],{},[47,72,73],{},"引用溯源","：回答标注来源文档 + 页码 + 原文片段，可验证",[44,76,77,80],{},[47,78,79],{},"多模型接入","：OpenAI \u002F Claude \u002F Ollama \u002F 通义千问 \u002F 智谱 \u002F 月之暗面",[44,82,83,86],{},[47,84,85],{},"多向量数据库","：Elasticsearch \u002F Infinity（自研）\u002F Chroma",[44,88,89,92],{},[47,90,91],{},"知识库管理","：多知识库 + 文档分类 + 解析状态监控",[44,94,95,98],{},[47,96,97],{},"API 接口","：完整 REST API + SDK，可集成到外部系统",[26,100,101],{"id":101},"价格",[103,104,105,120],"table",{},[106,107,108],"thead",{},[109,110,111,115,117],"tr",{},[112,113,114],"th",{},"方案",[112,116,101],{},[112,118,119],{},"核心功能",[121,122,123,135,146],"tbody",{},[109,124,125,129,132],{},[126,127,128],"td",{},"开源版",[126,130,131],{},"$0",[126,133,134],{},"完整功能，Apache 2.0，自托管",[109,136,137,140,143],{},[126,138,139],{},"Cloud",[126,141,142],{},"按量付费",[126,144,145],{},"托管服务，免运维",[109,147,148,151,154],{},[126,149,150],{},"Enterprise",[126,152,153],{},"联系销售",[126,155,156],{},"私有部署 + 技术支持 + 定制",[158,159,160],"blockquote",{},[31,161,162],{},"价格信息基于 2026-07 官网，可能调整。",[26,164,166],{"id":165},"体验与评测资料整理","体验与评测（资料整理）",[158,168,169],{},[31,170,171,172],{},"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",[47,173,174],{},"亮点：",[41,176,177,180,183,186,189,192],{},[44,178,179],{},"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",[44,181,182],{},"扫描件 OCR 效果好，中文印刷体识别准确率高",[44,184,185],{},"引用溯源到页码 + 原文片段，回答可信度高",[44,187,188],{},"混合检索 + Rerank 召回精度明显优于纯向量检索",[44,190,191],{},"中国团队出品，中文文档和社区支持好，Issue 响应快",[44,193,194],{},"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",[31,196,197],{},[47,198,199],{},"踩坑：",[41,201,202,205,208,211,214,217,220],{},[44,203,204],{},"资源消耗大——Elasticsearch + Redis + MinIO + RAGFlow 本身，至少 16GB 内存",[44,206,207],{},"部署较重，Docker Compose 起来 5+ 容器，配置复杂",[44,209,210],{},"大文件解析慢——100 页 PDF 解析 + 切片可能 5-10 分钟",[44,212,213],{},"UI 仍有粗糙处，文档管理界面交互不够流畅",[44,215,216],{},"Agent 能力弱——RAG 问答是强项，复杂工具调用 \u002F 多步推理不如 Dify",[44,218,219],{},"解析失败的重试机制不完善，偶尔卡在 parsing 状态",[44,221,222],{},"版本迭代快，升级需注意数据迁移",[26,224,225],{"id":225},"上手",[227,228,229,232,239,245,252,255],"ol",{},[44,230,231],{},"系统准备：确保 16GB+ 内存 + Docker + Docker Compose",[44,233,234,235],{},"克隆仓库：",[236,237,238],"code",{},"git clone https:\u002F\u002Fgithub.com\u002Finfiniflow\u002Fragflow.git",[44,240,241,242],{},"启动服务：",[236,243,244],{},"cd ragflow\u002Fdocker && docker compose up -d",[44,246,247,248,251],{},"访问 ",[236,249,250],{},"http:\u002F\u002Flocalhost:80","，注册管理员账号",[44,253,254],{},"配置模型：Settings → Model Providers 添加 LLM + Embedding + Rerank",[44,256,257],{},"创建知识库 → 上传文档 → 等待解析完成 → 开始问答",[26,259,260],{"id":260},"对比",[103,262,263,281],{},[106,264,265],{},[109,266,267,270,272,275,278],{},[112,268,269],{},"维度",[112,271,10],{},[112,273,274],{},"Dify",[112,276,277],{},"FastGPT",[112,279,280],{},"AnythingLLM",[121,282,283,300,315,330,344,357,371,385],{},[109,284,285,288,291,294,297],{},[126,286,287],{},"文档解析",[126,289,290],{},"✅ 最强",[126,292,293],{},"中",[126,295,296],{},"强",[126,298,299],{},"弱",[109,301,302,305,308,311,313],{},[126,303,304],{},"表格识别",[126,306,307],{},"✅",[126,309,310],{},"❌",[126,312,307],{},[126,314,310],{},[109,316,317,320,323,325,328],{},[126,318,319],{},"OCR",[126,321,322],{},"✅ 内置",[126,324,310],{},[126,326,327],{},"需配置",[126,329,327],{},[109,331,332,335,338,340,342],{},[126,333,334],{},"召回精度",[126,336,337],{},"高",[126,339,337],{},[126,341,337],{},[126,343,293],{},[109,345,346,348,351,353,355],{},[126,347,73],{},[126,349,350],{},"✅ 页码+片段",[126,352,307],{},[126,354,307],{},[126,356,307],{},[109,358,359,362,364,366,368],{},[126,360,361],{},"Agent 能力",[126,363,299],{},[126,365,296],{},[126,367,293],{},[126,369,370],{},"基础",[109,372,373,376,378,380,382],{},[126,374,375],{},"资源消耗",[126,377,337],{},[126,379,293],{},[126,381,293],{},[126,383,384],{},"低",[109,386,387,390,393,396,398],{},[126,388,389],{},"中文支持",[126,391,392],{},"✅ 优秀",[126,394,395],{},"好",[126,397,395],{},[126,399,400],{},"一般",[26,402,403],{"id":403},"避坑",[41,405,406,412,422,428,434,440,446],{},[44,407,408,411],{},[47,409,410],{},"资源一定要够","：低于 16GB 内存别部署，ES + Redis + MinIO 都吃内存",[44,413,414,417,418,421],{},[47,415,416],{},"Elasticsearch 配置","：默认 JVM 堆偏小，大知识库调 ",[236,419,420],{},"ES_JAVA_OPTS"," 到 4-8GB",[44,423,424,427],{},[47,425,426],{},"大文件拆分上传","：超过 100 页的 PDF 拆成小文件，解析更稳定",[44,429,430,433],{},[47,431,432],{},"解析失败检查格式","：加密 PDF \u002F 损坏文件会卡住，上传前检查",[44,435,436,439],{},[47,437,438],{},"Rerank 模型别省","：召回精度提升的关键，用 bge-reranker 或 Cohere Rerank",[44,441,442,445],{},[47,443,444],{},"不要当 Agent 平台用","：RAG 问答是核心，复杂工具调用上 Dify",[44,447,448,451],{},[47,449,450],{},"定期备份","：ES 数据 + MinIO 文件，升级前完整快照",[26,453,455],{"id":454},"适合-不适合","适合 \u002F 不适合",[41,457,458,461,464,467,470,473,476,479,482,485],{},[44,459,460],{},"✅ 需要精准文档问答的企业知识库",[44,462,463],{},"✅ 复杂文档（表格 \u002F 图文 \u002F 扫描件）RAG 场景",[44,465,466],{},"✅ 中文 RAG 需求（国产模型 + 中文 OCR）",[44,468,469],{},"✅ 对召回率和引用溯源要求高的业务",[44,471,472],{},"✅ 有运维能力的团队私有化部署",[44,474,475],{},"❌ 需要复杂 Agent 编排（用 Dify）",[44,477,478],{},"❌ 资源有限的小服务器（至少 16GB 内存）",[44,480,481],{},"❌ 需要精美 C 端 UI 的产品",[44,483,484],{},"❌ 无运维能力的团队（用 Cloud 版或 FastGPT）",[44,486,487],{},"❌ 纯英文简单文档场景（AnythingLLM 更轻量）",[26,489,491],{"id":490},"faq","FAQ",[31,493,494,497],{},[47,495,496],{},"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。",[31,499,500,503],{},[47,501,502],{},"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。",[31,505,506,509],{},[47,507,508],{},"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。",[31,511,512,515],{},[47,513,514],{},"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。",[26,517,518],{"id":518},"相关阅读",[31,520,521,525,526,525,530],{},[522,523,280],"a",{"href":524},"\u002Fagent\u002Fplatform\u002Fanythingllm.html"," · ",[522,527,529],{"href":528},"\u002Fagent\u002Fplatform\u002Fflowise.html","Flowise",[522,531,533],{"href":532},"\u002Fcoding\u002Fapi\u002Flangfuse.html","Langfuse",[26,535,536],{"id":536},"来源",[158,538,539],{},[31,540,541],{},"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。",[41,543,544,552],{},[44,545,546],{},[522,547,551],{"href":548,"rel":549},"https:\u002F\u002Fragflow.io",[550],"nofollow","官网",[44,553,554],{},[522,555,558],{"href":556,"rel":557},"https:\u002F\u002Fgithub.com\u002Finfiniflow\u002Fragflow",[550],"GitHub",{"title":560,"searchDepth":561,"depth":561,"links":562},"",3,[563,565,566,567,568,569,570,571,572,573,574],{"id":28,"depth":564,"text":29},2,{"id":39,"depth":564,"text":39},{"id":101,"depth":564,"text":101},{"id":165,"depth":564,"text":166},{"id":225,"depth":564,"text":225},{"id":260,"depth":564,"text":260},{"id":403,"depth":564,"text":403},{"id":454,"depth":564,"text":455},{"id":490,"depth":564,"text":491},{"id":518,"depth":564,"text":518},{"id":536,"depth":564,"text":536},"platform",4,"\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。",true,"md",null,[583,584],"en","zh","2026-07-30",{},"\u002Ftools\u002Fagent\u002Fplatform\u002Fragflow","agent",[590,591],"linux","docker","Free \u002F 开源（Apache 2.0）\u002F Cloud","2026-07-05",{"power":576,"ux":561,"price":595,"cn_support":576,"stability":561},5,false,{"title":10,"description":578},"RAGFlow - 开源 RAG 引擎评测与部署 | AIHO","agent\u002Fplatform\u002Fragflow",[601,602],{"title":551,"url":548},{"title":558,"url":556},"tools\u002Fagent\u002Fplatform\u002Fragflow","开源 RAG 引擎，深度文档解析 + 高召回率",[606,607,608,609,610],"agent-platform","opensource","rag","document-parsing","self-host","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。","uQQcwnZ9KyLQk_Zr-LT6Sh0BRwRzRU6ulzOVbEOZTSE",[614,1725,2821],{"id":615,"title":274,"alternatives":616,"api_compatible":620,"body":621,"category":575,"chinese_friendly":576,"cover":1667,"description":1668,"domestic":579,"extension":580,"faq":581,"free":579,"github":913,"languages":1669,"lastVerified":1671,"meta":1672,"models":581,"navigation":579,"notSuitable":581,"opensource":579,"path":1673,"pillar":588,"platforms":1674,"priceTable":1677,"pricing":1694,"published":1695,"relatedPlaybooks":581,"relatedReviews":1696,"score":1701,"self_host":579,"seo":1702,"seoTitle":1703,"slug":12,"sources":1704,"stem":1716,"suitable":581,"tagline":1717,"tags":1718,"updated":1722,"verdict":1723,"website":1609,"__hash__":1724},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Fdify.md",[617,13,618,619],"agent\u002Fplatform\u002Fcoze","agent\u002Fplatform\u002Fn8n","agent\u002Fplatform\u002Flangflow",[16,17,18,19,20,21],{"type":23,"value":622,"toc":1645},[623,625,650,655,658,663,666,735,738,742,745,759,776,780,783,794,798,806,817,821,824,827,831,840,898,905,909,917,980,983,1003,1007,1010,1069,1075,1079,1173,1176,1205,1208,1346,1364,1403,1406,1479,1481,1484,1504,1507,1536,1538,1600,1602,1633,1641],[26,624,29],{"id":28},[626,627,632,638],"div",{"className":628},[629,630,631],"card","p-5","my-4",[31,633,634,637],{},[47,635,636],{},"一句话："," 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。",[31,639,640,641,644,645,649],{},"最大价值是 ",[47,642,643],{},"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 ",[522,646,648],{"href":647},"\u002Fagent\u002Fplatform\u002Fcoze.html","Coze"," 折腾，新手得读 1-2 小时文档。",[158,651,652],{},[31,653,654],{},"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。",[26,656,657],{"id":657},"核心特性",[659,660,662],"h3",{"id":661},"可视化工作流chatflow-workflow","可视化工作流（Chatflow + Workflow）",[31,664,665],{},"Dify 把 LLM 应用拆成两种\"应用类型\"：",[103,667,668,681],{},[106,669,670],{},[109,671,672,675,678],{},[112,673,674],{},"类型",[112,676,677],{},"适合场景",[112,679,680],{},"编排范式",[121,682,683,696,709,722],{},[109,684,685,690,693],{},[126,686,687],{},[47,688,689],{},"Chatbot",[126,691,692],{},"简单对话机器人",[126,694,695],{},"prompt + tools",[109,697,698,703,706],{},[126,699,700],{},[47,701,702],{},"Agent",[126,704,705],{},"自主多步任务",[126,707,708],{},"ReAct \u002F Function Calling",[109,710,711,716,719],{},[126,712,713],{},[47,714,715],{},"Chatflow",[126,717,718],{},"对话型工作流（多轮 + 分支）",[126,720,721],{},"节点 DAG，带聊天上下文",[109,723,724,729,732],{},[126,725,726],{},[47,727,728],{},"Workflow",[126,730,731],{},"单次输入→输出（API 模式）",[126,733,734],{},"节点 DAG，无对话状态",[31,736,737],{},"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\"。",[659,739,741],{"id":740},"rag-知识库","RAG 知识库",[31,743,744],{},"内置完整 RAG 链路：",[227,746,747,750,753,756],{},[44,748,749],{},"上传文档（PDF \u002F Word \u002F Markdown \u002F 网页）",[44,751,752],{},"自动分块 + embedding（可配置分段策略和 embedding 模型）",[44,754,755],{},"混合检索（向量 + 全文 + 重排）",[44,757,758],{},"引用溯源（回答末尾自动附原文片段）",[31,760,761,762,767,768,771,772,775],{},"注意：根据 ",[522,763,766],{"href":764,"rel":765},"https:\u002F\u002Fzhuanlan.zhihu.com\u002Fp\u002F1887141987838309480",[550],"知乎 LLM 实战笔记 2025-03 对比"," 的实测，Dify ",[47,769,770],{},"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 ",[522,773,277],{"href":774},"\u002Fagent\u002Fplatform\u002Ffastgpt.html","（实测准确率高 10+ 个百分点），Dify 胜在工作流而非纯 RAG。",[659,777,779],{"id":778},"模型生态40-提供商","模型生态：40+ 提供商",[31,781,782],{},"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。",[31,784,785,786,788,789,793],{},"国产模型原生支持（不像 ",[522,787,277],{"href":774}," 需要 ",[522,790,792],{"href":791},"\u002Fcoding\u002Fapi\u002Fone-api.html","OneAPI"," 中转），是 Dify 在国内 toB 场景流行的关键。",[659,795,797],{"id":796},"mcp-协议支持","MCP 协议支持",[31,799,800,801,805],{},"Dify 较早接入了 ",[522,802,804],{"href":803},"\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：",[41,807,808,811,814],{},[44,809,810],{},"通过 MCP 调本地 PostgreSQL \u002F SQLite",[44,812,813],{},"通过 MCP 调 GitHub \u002F Slack \u002F Linear",[44,815,816],{},"通过 MCP 调自家内部系统（写一个 MCP Server 即可）",[659,818,820],{"id":819},"api-first","API-first",[31,822,823],{},"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。",[26,825,826],{"id":826},"价格与运行成本",[659,828,830],{"id":829},"云版difyai","云版（dify.ai）",[31,832,833,834,839],{},"根据 ",[522,835,838],{"href":836,"rel":837},"https:\u002F\u002Fwww.tooljunction.io\u002Fai-tools\u002Fdify-ai",[550],"tooljunction.io 2026 评测"," 引用的官方定价：",[103,841,842,854],{},[106,843,844],{},[109,845,846,849,851],{},[112,847,848],{},"套餐",[112,850,101],{},[112,852,853],{},"主要限制",[121,855,856,867,878,889],{},[109,857,858,861,864],{},[126,859,860],{},"Sandbox",[126,862,863],{},"免费",[126,865,866],{},"200 次模型调用，1 app，5MB 知识库",[109,868,869,872,875],{},[126,870,871],{},"Professional",[126,873,874],{},"$59\u002F月起",[126,876,877],{},"5000 调用\u002F月，多 app，50MB 知识库",[109,879,880,883,886],{},[126,881,882],{},"Team",[126,884,885],{},"$159\u002F月起",[126,887,888],{},"团队协作、SSO",[109,890,891,893,895],{},[126,892,150],{},[126,894,153],{},[126,896,897],{},"定制 SLA、私有云",[31,899,900,901,904],{},"注意：云版价格只是 Dify 平台费，",[47,902,903],{},"模型 API 费用另算","（自带 OpenAI \u002F Anthropic key）。",[659,906,908],{"id":907},"自托管推荐","自托管（推荐）",[31,910,911,916],{},[522,912,915],{"href":913,"rel":914},"https:\u002F\u002Fgithub.com\u002Flanggenius\u002Fdify",[550],"官方 GitHub 仓库"," 提供 Docker Compose 部署，社区版完全免费可商用：",[918,919,923],"pre",{"className":920,"code":921,"language":922,"meta":560,"style":560},"language-bash shiki shiki-themes github-light github-dark","git clone https:\u002F\u002Fgithub.com\u002Flanggenius\u002Fdify.git\ncd dify\u002Fdocker\ncp .env.example .env\ndocker compose up -d\n# 默认 http:\u002F\u002Flocalhost \u002F 端口可在 .env 调整\n","bash",[236,924,925,941,950,961,974],{"__ignoreMap":560},[926,927,930,934,938],"span",{"class":928,"line":929},"line",1,[926,931,933],{"class":932},"sScJk","git",[926,935,937],{"class":936},"sZZnC"," clone",[926,939,940],{"class":936}," https:\u002F\u002Fgithub.com\u002Flanggenius\u002Fdify.git\n",[926,942,943,947],{"class":928,"line":564},[926,944,946],{"class":945},"sj4cs","cd",[926,948,949],{"class":936}," dify\u002Fdocker\n",[926,951,952,955,958],{"class":928,"line":561},[926,953,954],{"class":932},"cp",[926,956,957],{"class":936}," .env.example",[926,959,960],{"class":936}," .env\n",[926,962,963,965,968,971],{"class":928,"line":576},[926,964,591],{"class":932},[926,966,967],{"class":936}," compose",[926,969,970],{"class":936}," up",[926,972,973],{"class":945}," -d\n",[926,975,976],{"class":928,"line":595},[926,977,979],{"class":978},"sJ8bj","# 默认 http:\u002F\u002Flocalhost \u002F 端口可在 .env 调整\n",[31,981,982],{},"硬件门槛（社区共识，非官方硬性要求）：",[41,984,985,991,997],{},[44,986,987,990],{},[47,988,989],{},"最低","：2 核 4G，纯外接 API 模式",[44,992,993,996],{},[47,994,995],{},"推荐","：4 核 8G + 至少 30GB 磁盘（向量数据 + 文件存储）",[44,998,999,1002],{},[47,1000,1001],{},"企业","：8 核 16G+，单机日活上千",[659,1004,1006],{"id":1005},"真实-tco","真实 TCO",[31,1008,1009],{},"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）：",[103,1011,1012,1025],{},[106,1013,1014],{},[109,1015,1016,1019,1022],{},[112,1017,1018],{},"成本项",[112,1020,1021],{},"云版 Professional",[112,1023,1024],{},"自托管",[121,1026,1027,1037,1047,1058],{},[109,1028,1029,1032,1035],{},[126,1030,1031],{},"平台费",[126,1033,1034],{},"~$2,100（3 年）",[126,1036,131],{},[109,1038,1039,1042,1044],{},[126,1040,1041],{},"服务器",[126,1043,131],{},[126,1045,1046],{},"~$50\u002F月 × 36 = $1,800",[109,1048,1049,1052,1055],{},[126,1050,1051],{},"模型 API",[126,1053,1054],{},"与下同",[126,1056,1057],{},"与上同",[109,1059,1060,1063,1066],{},[126,1061,1062],{},"运维人力",[126,1064,1065],{},"0",[126,1067,1068],{},"约 0.2 人月",[31,1070,1071,1074],{},[47,1072,1073],{},"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。",[26,1076,1078],{"id":1077},"上手-10-分钟","上手 10 分钟",[918,1080,1082],{"className":920,"code":1081,"language":922,"meta":560,"style":560},"# 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",[236,1083,1084,1089,1097,1103,1111,1121,1127,1133,1139,1144,1150,1155,1161,1167],{"__ignoreMap":560},[926,1085,1086],{"class":928,"line":929},[926,1087,1088],{"class":978},"# 1. 自托管（社区版）\n",[926,1090,1091,1093,1095],{"class":928,"line":564},[926,1092,933],{"class":932},[926,1094,937],{"class":936},[926,1096,940],{"class":936},[926,1098,1099,1101],{"class":928,"line":561},[926,1100,946],{"class":945},[926,1102,949],{"class":936},[926,1104,1105,1107,1109],{"class":928,"line":576},[926,1106,954],{"class":932},[926,1108,957],{"class":936},[926,1110,960],{"class":936},[926,1112,1113,1115,1117,1119],{"class":928,"line":595},[926,1114,591],{"class":932},[926,1116,967],{"class":936},[926,1118,970],{"class":936},[926,1120,973],{"class":945},[926,1122,1124],{"class":928,"line":1123},6,[926,1125,1126],{"emptyLinePlaceholder":579},"\n",[926,1128,1130],{"class":928,"line":1129},7,[926,1131,1132],{"class":978},"# 2. 浏览器打开 http:\u002F\u002Flocalhost\n",[926,1134,1136],{"class":928,"line":1135},8,[926,1137,1138],{"class":978},"#    首次会让你创建 admin 账号\n",[926,1140,1142],{"class":928,"line":1141},9,[926,1143,1126],{"emptyLinePlaceholder":579},[926,1145,1147],{"class":928,"line":1146},10,[926,1148,1149],{"class":978},"# 3. 进入\"设置 → 模型供应商\"，配置 OpenAI \u002F 国产模型 API key\n",[926,1151,1153],{"class":928,"line":1152},11,[926,1154,1126],{"emptyLinePlaceholder":579},[926,1156,1158],{"class":928,"line":1157},12,[926,1159,1160],{"class":978},"# 4. 在主界面\"创建空白应用\"，选 Chatflow 或 Workflow\n",[926,1162,1164],{"class":928,"line":1163},13,[926,1165,1166],{"class":978},"# 5. 拖入\"开始 → LLM → 结束\"节点试一下基础 prompt\n",[926,1168,1170],{"class":928,"line":1169},14,[926,1171,1172],{"class":978},"# 6. 满意了点右上\"发布\"，自动生成 API endpoint\n",[26,1174,1175],{"id":1175},"国内使用注意事项",[227,1177,1178,1184,1190,1196],{},[44,1179,1180,1183],{},[47,1181,1182],{},"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",[44,1185,1186,1189],{},[47,1187,1188],{},"自托管 + 国产模型"," = 完全国内闭环，是 Dify 在国内最大优势",[44,1191,1192,1195],{},[47,1193,1194],{},"Docker 镜像拉取","：国内可能慢，建议配 Docker registry 镜像（阿里云 \u002F 网易）",[44,1197,1198,1201,1202,1204],{},[47,1199,1200],{},"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 ",[522,1203,648],{"href":647}," 迁到 Dify",[26,1206,1207],{"id":1207},"与同类怎么选",[103,1209,1210,1232],{},[106,1211,1212],{},[109,1213,1214,1216,1218,1222,1226],{},[112,1215,269],{},[112,1217,274],{},[112,1219,1220],{},[522,1221,648],{"href":647},[112,1223,1224],{},[522,1225,277],{"href":774},[112,1227,1228],{},[522,1229,1231],{"href":1230},"\u002Fagent\u002Fplatform\u002Fn8n.html","n8n",[121,1233,1234,1248,1261,1277,1291,1305,1319,1332],{},[109,1235,1236,1239,1241,1243,1245],{},[126,1237,1238],{},"开源",[126,1240,307],{},[126,1242,310],{},[126,1244,307],{},[126,1246,1247],{},"✅（fair-code）",[109,1249,1250,1253,1255,1257,1259],{},[126,1251,1252],{},"私有部署",[126,1254,307],{},[126,1256,310],{},[126,1258,307],{},[126,1260,307],{},[109,1262,1263,1266,1269,1272,1274],{},[126,1264,1265],{},"上手难度",[126,1267,1268],{},"★★★☆☆",[126,1270,1271],{},"★★☆☆☆ 最简单",[126,1273,1268],{},[126,1275,1276],{},"★★★★☆",[109,1278,1279,1282,1285,1287,1289],{},[126,1280,1281],{},"工作流编排",[126,1283,1284],{},"★★★★★",[126,1286,1276],{},[126,1288,1268],{},[126,1290,1284],{},[109,1292,1293,1296,1298,1300,1302],{},[126,1294,1295],{},"RAG 精度",[126,1297,1276],{},[126,1299,1268],{},[126,1301,1284],{},[126,1303,1304],{},"★★☆☆☆",[109,1306,1307,1310,1312,1314,1317],{},[126,1308,1309],{},"模型生态",[126,1311,1284],{},[126,1313,1276],{},[126,1315,1316],{},"★★★☆☆（OneAPI 中转）",[126,1318,1276],{},[109,1320,1321,1324,1326,1328,1330],{},[126,1322,1323],{},"中文场景",[126,1325,1276],{},[126,1327,1284],{},[126,1329,1276],{},[126,1331,1268],{},[109,1333,1334,1337,1339,1342,1344],{},[126,1335,1336],{},"字节生态绑定",[126,1338,310],{},[126,1340,1341],{},"✅（飞书\u002F抖音深度集成）",[126,1343,310],{},[126,1345,310],{},[31,1347,1348,1351,1352,1357,1358,1363],{},[47,1349,1350],{},"怎么选","（基于 ",[522,1353,1356],{"href":1354,"rel":1355},"https:\u002F\u002Fwww.besthub.dev\u002Farticles\u002Fcoze-vs-dify-vs-fastgpt-which-ai-agent-platform-fits-your-needs-fa59cf97b798",[550],"BestHub 2025-07"," 和 ",[522,1359,1362],{"href":1360,"rel":1361},"https:\u002F\u002Fwww.cnblogs.com\u002Fuulucias\u002Fp\u002F19449008",[550],"博客园 2026-01"," 两份选型指南综合）：",[41,1365,1366,1372,1380,1387,1394],{},[44,1367,1368,1371],{},[47,1369,1370],{},"数据必须不出内网 + 工作流复杂"," → Dify",[44,1373,1374,1377,1378],{},[47,1375,1376],{},"个人 \u002F 小团队 \u002F 快速原型 + 字节生态"," → ",[522,1379,648],{"href":647},[44,1381,1382,1377,1385],{},[47,1383,1384],{},"核心场景就是企业知识库 QA",[522,1386,277],{"href":774},[44,1388,1389,1377,1392],{},[47,1390,1391],{},"重点是连接外部 SaaS（Slack \u002F Notion \u002F 数据库）",[522,1393,1231],{"href":1230},[44,1395,1396,1377,1399],{},[47,1397,1398],{},"要画图式表达 LangChain pipeline",[522,1400,1402],{"href":1401},"\u002Fagent\u002Fplatform\u002Flangflow.html","Langflow",[26,1404,1405],{"id":1405},"避坑清单",[41,1407,1408,1414,1431,1442,1455,1461,1467,1473],{},[44,1409,1410,1413],{},[47,1411,1412],{},"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。",[44,1415,1416,1422,1423,1426,1427,1430],{},[47,1417,1418,1421],{},[236,1419,1420],{},".env"," 文件改完忘 restart","：",[236,1424,1425],{},"docker compose down && up -d","，不是 ",[236,1428,1429],{},"restart","——后者不重新加载 env。",[44,1432,1433,1422,1436,1441],{},[47,1434,1435],{},"大版本升级会破坏数据库 schema",[522,1437,1440],{"href":1438,"rel":1439},"https:\u002F\u002Fdocs.dify.ai\u002Fzh-hans",[550],"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。",[44,1443,1444,1447,1448,1450,1451,1454],{},[47,1445,1446],{},"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 ",[236,1449,1420],{}," 的 ",[236,1452,1453],{},"UPLOAD_FILE_SIZE_LIMIT"," 并重启容器。",[44,1456,1457,1460],{},[47,1458,1459],{},"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。",[44,1462,1463,1466],{},[47,1464,1465],{},"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。",[44,1468,1469,1472],{},[47,1470,1471],{},"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。",[44,1474,1475,1478],{},[47,1476,1477],{},"国内 Docker 拉取镜像慢","：先配国内 registry，否则首次 pull 可能要 30+ 分钟。",[26,1480,455],{"id":454},[31,1482,1483],{},"✅ 适合：",[41,1485,1486,1489,1492,1495,1498,1501],{},[44,1487,1488],{},"中大型企业 LLM 中台建设",[44,1490,1491],{},"需要私有化部署（金融 \u002F 医疗 \u002F 政府）",[44,1493,1494],{},"想做\"AI 工作流即产品\"的开发团队",[44,1496,1497],{},"同时需要 RAG + Agent + Workflow 三件套",[44,1499,1500],{},"想用国产模型 + 国际模型混合编排",[44,1502,1503],{},"已经接受 Docker + 一定运维投入",[31,1505,1506],{},"❌ 不适合：",[41,1508,1509,1515,1521,1524,1530],{},[44,1510,1511,1512,1514],{},"纯个人玩家做对话机器人（",[522,1513,648],{"href":647}," 更快）",[44,1516,1517,1518,1520],{},"只想做企业知识库 QA（",[522,1519,277],{"href":774}," RAG 更专）",[44,1522,1523],{},"团队完全没运维能力（云版还行，自托管会踩坑）",[44,1525,1526,1527,1529],{},"需要深度对接字节飞书 \u002F 抖音（",[522,1528,648],{"href":647}," 原生）",[44,1531,1532,1533,1535],{},"工作流核心是连接 100+ SaaS（",[522,1534,1231],{"href":1230}," 节点更全）",[26,1537,518],{"id":518},[41,1539,1540,1552,1570,1589],{},[44,1541,1542,1543,1545,1546,1545,1548,1545,1550],{},"同类对比：",[522,1544,648],{"href":647}," \u002F ",[522,1547,277],{"href":774},[522,1549,1231],{"href":1230},[522,1551,1402],{"href":1401},[44,1553,1554,1555,1545,1559,1545,1563,1545,1566],{},"概念基础：",[522,1556,1558],{"href":1557},"\u002Fwiki\u002Fai-agent.html","AI Agent",[522,1560,1562],{"href":1561},"\u002Fwiki\u002Frag.html","RAG",[522,1564,1565],{"href":803},"MCP",[522,1567,1569],{"href":1568},"\u002Fwiki\u002Ffunction-calling.html","Function Calling",[44,1571,1572,1573,1545,1577,1545,1581,1545,1585],{},"模型选型：",[522,1574,1576],{"href":1575},"\u002Fmodels\u002Fgpt-5.html","GPT-5",[522,1578,1580],{"href":1579},"\u002Fmodels\u002Fclaude-sonnet-4.html","Claude Sonnet 4",[522,1582,1584],{"href":1583},"\u002Fmodels\u002Fdeepseek-v3.html","DeepSeek-V3",[522,1586,1588],{"href":1587},"\u002Fmodels\u002Fglm-5.2.html","GLM-5.2",[44,1590,1591,1592,1545,1596],{},"进阶：",[522,1593,1595],{"href":1594},"\u002Fwiki\u002Ffine-tuning-vs-rag.html","Fine-tuning vs RAG",[522,1597,1599],{"href":1598},"\u002Fwiki\u002Fcontext-engineering.html","Context Engineering",[26,1601,536],{"id":536},[41,1603,1604,1611,1617,1623,1630],{},[44,1605,1606,1607],{},"官网：",[522,1608,1609],{"href":1609,"rel":1610},"https:\u002F\u002Fdify.ai",[550],[44,1612,1613,1614],{},"中文文档：",[522,1615,1438],{"href":1438,"rel":1616},[550],[44,1618,1619,1620],{},"GitHub：",[522,1621,913],{"href":913,"rel":1622},[550],[44,1624,1625,1626],{},"官方定价：",[522,1627,1628],{"href":1628,"rel":1629},"https:\u002F\u002Fdify.ai\u002Fpricing",[550],[44,1631,1632],{},"第三方评测：tooljunction.io \u002F chatforest.com \u002F besthub.dev \u002F joshuaopolko.com \u002F 知乎 LLM 实战笔记",[31,1634,1635,1636,1640],{},"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 ",[522,1637,1639],{"href":1638},"mailto:hello@aiho.net","反馈邮箱"," 反馈。",[1642,1643,1644],"style",{},"html pre.shiki code .sScJk, html code.shiki .sScJk{--shiki-default:#6F42C1;--shiki-dark:#B392F0}html pre.shiki code .sZZnC, html code.shiki .sZZnC{--shiki-default:#032F62;--shiki-dark:#9ECBFF}html pre.shiki code .sj4cs, html code.shiki .sj4cs{--shiki-default:#005CC5;--shiki-dark:#79B8FF}html pre.shiki code .sJ8bj, html code.shiki .sJ8bj{--shiki-default:#6A737D;--shiki-dark:#6A737D}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"title":560,"searchDepth":561,"depth":561,"links":1646},[1647,1648,1655,1660,1661,1662,1663,1664,1665,1666],{"id":28,"depth":564,"text":29},{"id":657,"depth":564,"text":657,"children":1649},[1650,1651,1652,1653,1654],{"id":661,"depth":561,"text":662},{"id":740,"depth":561,"text":741},{"id":778,"depth":561,"text":779},{"id":796,"depth":561,"text":797},{"id":819,"depth":561,"text":820},{"id":826,"depth":564,"text":826,"children":1656},[1657,1658,1659],{"id":829,"depth":561,"text":830},{"id":907,"depth":561,"text":908},{"id":1005,"depth":561,"text":1006},{"id":1077,"depth":564,"text":1078},{"id":1175,"depth":564,"text":1175},{"id":1207,"depth":564,"text":1207},{"id":1405,"depth":564,"text":1405},{"id":454,"depth":564,"text":455},{"id":518,"depth":564,"text":518},{"id":536,"depth":564,"text":536},"\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。",[584,583,1670],"ja","2026-08-02",{},"\u002Ftools\u002Fagent\u002Fplatform\u002Fdify",[1675,1676,590,591],"windows","macos",[1678,1682,1686,1690],{"plan":1679,"price":131,"features":1680,"notes":1681},"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":1683,"price":131,"features":1684,"notes":1685},"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":1687,"price":874,"features":1688,"notes":1689},"Cloud Professional","更高调用额度 + 团队协作 + 商用支持","商用云首选",{"plan":1691,"price":1692,"features":1693,"notes":153},"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","2026-06-18",[1697,1698,1699,1700],"coze-deep-review","coze-vs-dify","dify-deep-review","fastgpt-deep-review",{"power":595,"ux":576,"price":595,"cn_support":576,"stability":576},{"title":274,"description":1668},"Dify 评测 2026：开源 LLMOps 与 AI Agent 平台，自托管指南",[1705,1707,1709,1711,1713],{"title":1706,"url":1438},"Dify 官方文档（中文）",{"title":1708,"url":913},"Dify GitHub",{"title":1710,"url":1628},"Dify 官方定价",{"title":1712,"url":1354},"Coze vs Dify vs FastGPT 选型",{"title":1714,"url":1715},"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",[606,607,610,608,1719,1720,1721],"workflow","llmops","mcp","2026-06-24","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。","q61l3oA5zdTKrp-66KGGh7wde1ZGUvjMulHa4GKaOXE",{"id":1726,"title":277,"alternatives":1727,"api_compatible":1728,"body":1730,"category":575,"chinese_friendly":595,"cover":2748,"description":2749,"domestic":579,"extension":580,"faq":581,"free":579,"github":2711,"languages":2750,"lastVerified":1671,"meta":2751,"models":2752,"navigation":579,"notSuitable":2759,"opensource":579,"path":2763,"pillar":588,"platforms":2764,"priceTable":2765,"pricing":2789,"published":1695,"relatedPlaybooks":2790,"relatedReviews":2792,"score":2793,"self_host":579,"seo":2794,"seoTitle":2795,"slug":13,"sources":2796,"stem":2807,"suitable":2808,"tagline":2814,"tags":2815,"updated":1722,"verdict":2819,"website":2705,"__hash__":2820},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Ffastgpt.md",[12,617,619,618],[1729],"openai",{"type":23,"value":1731,"toc":2730},[1732,1734,1763,1774,1776,1780,1783,1797,1800,1826,1830,1837,1869,1876,1880,1933,1940,1943,1946,1949,1956,2002,2009,2013,2044,2048,2154,2157,2160,2164,2172,2295,2310,2312,2469,2479,2513,2519,2521,2596,2598,2600,2620,2622,2640,2642,2697,2699,2722,2727],[26,1733,29],{"id":28},[626,1735,1737,1752],{"className":1736},[629,630,631],[31,1738,1739,1741,1742,1747,1748,1751],{},[47,1740,636],{}," labring 团队开源的 LLM 知识库 RAG 平台，27k+ GitHub star（截至 2026-03 数据，",[522,1743,1746],{"href":1744,"rel":1745},"https:\u002F\u002Fcloud.tencent.com\u002Fdeveloper\u002Farticle\u002F2632669",[550],"腾讯云 2026-03 教程"," 引用），Apache 2.0 许可证可商用。",[47,1749,1750],{},"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。",[31,1753,1754,1755,1758,1759,1762],{},"最大价值在 ",[47,1756,1757],{},"国内企业知识库 + 私有部署"," 场景。代价是 ",[47,1760,1761],{},"配置门槛","：docker 基础 + 网络知识 + 一定运维能力。",[158,1764,1765],{},[31,1766,1767,1768,1773],{},"来源说明：本文基于 fastgpt.io 官方页面、github.com\u002Flabring\u002FFastGPT 仓库、",[522,1769,1772],{"href":1770,"rel":1771},"https:\u002F\u002Fwww.nanhuantech.com\u002Fzh\u002Fai-reviews\u002Ffastgpt-2025-review",[550],"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。",[26,1775,657],{"id":657},[659,1777,1779],{"id":1778},"知识库管理核心能力","知识库管理（核心能力）",[31,1781,1782],{},"支持文件类型：",[41,1784,1785,1788,1791,1794],{},[44,1786,1787],{},"文档：PDF \u002F Word \u002F Markdown \u002F TXT \u002F HTML",[44,1789,1790],{},"表格：Excel \u002F CSV",[44,1792,1793],{},"网页：URL 抓取 + 定时同步",[44,1795,1796],{},"API：通过接口推送内容",[31,1798,1799],{},"处理流程：上传 → 文本切分 → 向量化 → 存储 → 可用于问答。支持：",[41,1801,1802,1808,1814,1820],{},[44,1803,1804,1807],{},[47,1805,1806],{},"文件夹分组","：不同主题 \u002F 部门分类",[44,1809,1810,1813],{},[47,1811,1812],{},"多种分块策略","：默认按段落 \u002F 按 token 数 \u002F 自定义",[44,1815,1816,1819],{},[47,1817,1818],{},"批量导入","：脚本化大批量同步",[44,1821,1822,1825],{},[47,1823,1824],{},"定时同步","：网页源自动更新",[659,1827,1829],{"id":1828},"rag-流程编排最强卖点","RAG 流程编排（最强卖点）",[31,1831,1832,1836],{},[522,1833,1835],{"href":1770,"rel":1834},[550],"南环 AI 2026 评测"," 总结的 FastGPT RAG 链路：",[227,1838,1839,1845,1851,1857,1863],{},[44,1840,1841,1844],{},[47,1842,1843],{},"问题预处理","：改写 \u002F 扩展 \u002F 错词纠正（提升召回率）",[44,1846,1847,1850],{},[47,1848,1849],{},"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",[44,1852,1853,1856],{},[47,1854,1855],{},"重排序（Rerank）","：对初步检索结果二次排序，提升相关性",[44,1858,1859,1862],{},[47,1860,1861],{},"上下文组装","：最优 chunk + 问题 → prompt",[44,1864,1865,1868],{},[47,1866,1867],{},"答案生成","：调大模型基于检索结果回答 + 引用标注",[31,1870,1871,1872,1875],{},"每一步都可视化调参，这是 FastGPT 比 Coze \u002F Dify 在 ",[47,1873,1874],{},"纯知识库 QA 精度","上更高的原因。",[659,1877,1879],{"id":1878},"多模型支持不绑定厂商","多模型支持（不绑定厂商）",[103,1881,1882,1892],{},[106,1883,1884],{},[109,1885,1886,1889],{},[112,1887,1888],{},"模型类别",[112,1890,1891],{},"支持",[121,1893,1894,1902,1909,1917,1925],{},[109,1895,1896,1899],{},[126,1897,1898],{},"国产闭源",[126,1900,1901],{},"豆包 \u002F 通义千问 \u002F 文心一言 \u002F 智谱 GLM \u002F Moonshot Kimi \u002F MiniMax",[109,1903,1904,1906],{},[126,1905,1238],{},[126,1907,1908],{},"LLaMA \u002F Qwen \u002F ChatGLM \u002F DeepSeek 等可自部署",[109,1910,1911,1914],{},[126,1912,1913],{},"OpenAI 系",[126,1915,1916],{},"GPT-5 \u002F GPT-5 mini \u002F o3",[109,1918,1919,1922],{},[126,1920,1921],{},"Claude 系",[126,1923,1924],{},"Sonnet 4 \u002F Opus 4 \u002F Haiku",[109,1926,1927,1930],{},[126,1928,1929],{},"嵌入 \u002F 重排",[126,1931,1932],{},"BGE \u002F m3e \u002F OpenAI text-embedding-3",[31,1934,1935,1936,1939],{},"可以在 ",[47,1937,1938],{},"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。",[659,1941,1942],{"id":1942},"工作流与高级编排",[31,1944,1945],{},"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。",[659,1947,1948],{"id":1948},"多向量库选择",[31,1950,1951,1955],{},[522,1952,1954],{"href":1744,"rel":1953},[550],"腾讯云教程"," 公开的 4 种向量后端：",[103,1957,1958,1968],{},[106,1959,1960],{},[109,1961,1962,1965],{},[112,1963,1964],{},"后端",[112,1966,1967],{},"适用",[121,1969,1970,1978,1986,1994],{},[109,1971,1972,1975],{},[126,1973,1974],{},"PgVector",[126,1976,1977],{},"5000 万索引以下，新手 \u002F 小规模",[109,1979,1980,1983],{},[126,1981,1982],{},"Milvus",[126,1984,1985],{},"亿级以上，高性能",[109,1987,1988,1991],{},[126,1989,1990],{},"Zilliz Cloud",[126,1992,1993],{},"Milvus 全托管 SaaS",[109,1995,1996,1999],{},[126,1997,1998],{},"SeekDB \u002F OceanBase",[126,2000,2001],{},"企业级国产化",[31,2003,2004,2005,2008],{},"部署时选对应 ",[236,2006,2007],{},"docker-compose.{pgvector|milvus|...}.yml","。",[659,2010,2012],{"id":2011},"api-与-mcp","API 与 MCP",[41,2014,2015,2021,2027,2038],{},[44,2016,2017,2020],{},[47,2018,2019],{},"对话 API","：流式 \u002F 非流式 HTTP，OpenAI 兼容",[44,2022,2023,2026],{},[47,2024,2025],{},"知识库检索 API","：单独调检索（不走生成）做 hybrid pipeline",[44,2028,2029,2032,2033,2037],{},[47,2030,2031],{},"MCP Server","：3005 端口暴露 MCP SSE 服务，可被 ",[522,2034,2036],{"href":2035},"\u002Fcoding\u002Fcli\u002Fclaude-code.html","Claude Code"," 等客户端直接接入",[44,2039,2040,2043],{},[47,2041,2042],{},"Webhook","：回调通知",[26,2045,2047],{"id":2046},"部署-10-分钟docker","部署 10 分钟（Docker）",[918,2049,2051],{"className":920,"code":2050,"language":922,"meta":560,"style":560},"# 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",[236,2052,2053,2058,2067,2074,2078,2083,2096,2100,2105,2112,2120,2124,2129,2144,2148],{"__ignoreMap":560},[926,2054,2055],{"class":928,"line":929},[926,2056,2057],{"class":978},"# 克隆代码\n",[926,2059,2060,2062,2064],{"class":928,"line":564},[926,2061,933],{"class":932},[926,2063,937],{"class":936},[926,2065,2066],{"class":936}," https:\u002F\u002Fgithub.com\u002Flabring\u002FFastGPT.git\n",[926,2068,2069,2071],{"class":928,"line":561},[926,2070,946],{"class":945},[926,2072,2073],{"class":936}," FastGPT\n",[926,2075,2076],{"class":928,"line":576},[926,2077,1126],{"emptyLinePlaceholder":579},[926,2079,2080],{"class":928,"line":595},[926,2081,2082],{"class":978},"# 切到最新稳定版（参考 GitHub releases）\n",[926,2084,2085,2087,2090,2093],{"class":928,"line":1123},[926,2086,933],{"class":932},[926,2088,2089],{"class":936}," switch",[926,2091,2092],{"class":945}," -c",[926,2094,2095],{"class":945}," 4.14.7.2\n",[926,2097,2098],{"class":928,"line":1129},[926,2099,1126],{"emptyLinePlaceholder":579},[926,2101,2102],{"class":928,"line":1135},[926,2103,2104],{"class":978},"# 选向量库版本（个人 \u002F 小规模选 pg）\n",[926,2106,2107,2109],{"class":928,"line":1141},[926,2108,946],{"class":945},[926,2110,2111],{"class":936}," deploy\u002Fdocker\u002Fcn\n",[926,2113,2114,2117],{"class":928,"line":1146},[926,2115,2116],{"class":932},"wget",[926,2118,2119],{"class":936}," https:\u002F\u002Fdoc.fastgpt.cn\u002Fdeploy\u002Fconfig\u002Fconfig.json\n",[926,2121,2122],{"class":928,"line":1152},[926,2123,1126],{"emptyLinePlaceholder":579},[926,2125,2126],{"class":928,"line":1157},[926,2127,2128],{"class":978},"# 启动\n",[926,2130,2131,2134,2137,2140,2142],{"class":928,"line":1163},[926,2132,2133],{"class":932},"docker-compose",[926,2135,2136],{"class":945}," -f",[926,2138,2139],{"class":936}," docker-compose.pg.yml",[926,2141,970],{"class":936},[926,2143,973],{"class":945},[926,2145,2146],{"class":928,"line":1169},[926,2147,1126],{"emptyLinePlaceholder":579},[926,2149,2151],{"class":928,"line":2150},15,[926,2152,2153],{"class":978},"# 访问 http:\u002F\u002F\u003Cip>:3000，默认账号 root \u002F 1234\n",[31,2155,2156],{},"最低配置：2C4G + 20GB 硬盘 + Docker 28+ + Docker Compose 2.20+。",[31,2158,2159],{},"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。",[26,2161,2163],{"id":2162},"云版-vs-自托管对比","云版 vs 自托管对比",[31,2165,2166,2171],{},[522,2167,2170],{"href":2168,"rel":2169},"https:\u002F\u002Ffastgpt.io\u002Fzh\u002Fprice",[550],"fastgpt.io 官方定价"," 公开数据：",[103,2173,2174,2199],{},[106,2175,2176],{},[109,2177,2178,2180,2182,2185,2188,2191,2193,2196],{},[112,2179,848],{},[112,2181,101],{},[112,2183,2184],{},"AI 积分",[112,2186,2187],{},"知识库索引",[112,2189,2190],{},"团队",[112,2192,702],{},[112,2194,2195],{},"知识库",[112,2197,2198],{},"QPM",[121,2200,2201,2226,2250,2274],{},[109,2202,2203,2205,2208,2211,2214,2217,2220,2223],{},[126,2204,863],{},[126,2206,2207],{},"¥0",[126,2209,2210],{},"100",[126,2212,2213],{},"600",[126,2215,2216],{},"1",[126,2218,2219],{},"10",[126,2221,2222],{},"3",[126,2224,2225],{},"30",[109,2227,2228,2230,2233,2236,2239,2242,2245,2247],{},[126,2229,370],{},[126,2231,2232],{},"¥99\u002F月",[126,2234,2235],{},"4000",[126,2237,2238],{},"6000",[126,2240,2241],{},"5",[126,2243,2244],{},"50",[126,2246,2225],{},[126,2248,2249],{},"300",[109,2251,2252,2255,2258,2261,2264,2266,2269,2271],{},[126,2253,2254],{},"高级",[126,2256,2257],{},"¥599\u002F月",[126,2259,2260],{},"25000",[126,2262,2263],{},"36000",[126,2265,2244],{},[126,2267,2268],{},"200",[126,2270,2210],{},[126,2272,2273],{},"1500",[109,2275,2276,2279,2282,2285,2287,2289,2291,2293],{},[126,2277,2278],{},"定制",[126,2280,2281],{},"议价",[126,2283,2284],{},"弹性",[126,2286,2284],{},[126,2288,2284],{},[126,2290,2284],{},[126,2292,2284],{},[126,2294,2284],{},[31,2296,2297,2300,2301,2304,2305,2309],{},[47,2298,2299],{},"云版适合","：不想运维、量小、要快速上线\n",[47,2302,2303],{},"自托管适合","：量大（10 万+ 日问答）、数据敏感、要深度定制——按 ",[522,2306,2308],{"href":1770,"rel":2307},[550],"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\"",[26,2311,1207],{"id":1207},[103,2313,2314,2335],{},[106,2315,2316],{},[109,2317,2318,2320,2322,2327,2331,2333],{},[112,2319,269],{},[112,2321,277],{},[112,2323,2324],{},[522,2325,274],{"href":2326},"\u002Fagent\u002Fplatform\u002Fdify.html",[112,2328,2329],{},[522,2330,648],{"href":647},[112,2332,10],{},[112,2334,280],{},[121,2336,2337,2357,2373,2389,2406,2421,2437,2452],{},[109,2338,2339,2342,2345,2348,2351,2354],{},[126,2340,2341],{},"核心定位",[126,2343,2344],{},"知识库 QA",[126,2346,2347],{},"综合 LLMOps",[126,2349,2350],{},"Bot + 工作流",[126,2352,2353],{},"文档解析+RAG",[126,2355,2356],{},"桌面级 KB",[109,2358,2359,2361,2364,2366,2368,2370],{},[126,2360,1238],{},[126,2362,2363],{},"✅ Apache 2.0",[126,2365,2363],{},[126,2367,310],{},[126,2369,2363],{},[126,2371,2372],{},"✅ MIT",[109,2374,2375,2377,2380,2382,2385,2387],{},[126,2376,1252],{},[126,2378,2379],{},"★★★★★ docker",[126,2381,1284],{},[126,2383,2384],{},"⚠️ 仅企业版",[126,2386,1276],{},[126,2388,1284],{},[109,2390,2391,2394,2397,2399,2401,2404],{},[126,2392,2393],{},"RAG 深度",[126,2395,2396],{},"★★★★★ 最细",[126,2398,1276],{},[126,2400,1268],{},[126,2402,2403],{},"★★★★★ 文档解析最强",[126,2405,1268],{},[109,2407,2408,2411,2413,2415,2417,2419],{},[126,2409,2410],{},"工作流",[126,2412,1276],{},[126,2414,1284],{},[126,2416,1276],{},[126,2418,1268],{},[126,2420,1304],{},[109,2422,2423,2425,2428,2430,2433,2435],{},[126,2424,225],{},[126,2426,2427],{},"★★★☆☆ 需 docker",[126,2429,1276],{},[126,2431,2432],{},"★★★★★ 最简单",[126,2434,1268],{},[126,2436,1276],{},[109,2438,2439,2442,2444,2446,2448,2450],{},[126,2440,2441],{},"中文优化",[126,2443,1284],{},[126,2445,1276],{},[126,2447,1284],{},[126,2449,1276],{},[126,2451,1268],{},[109,2453,2454,2457,2460,2462,2464,2467],{},[126,2455,2456],{},"多平台发布",[126,2458,2459],{},"⚠️ API 为主",[126,2461,1276],{},[126,2463,1284],{},[126,2465,2466],{},"⚠️",[126,2468,2466],{},[31,2470,2471,2473,2474,2478],{},[47,2472,1350],{},"（综合 ",[522,2475,2477],{"href":1770,"rel":2476},[550],"南环 AI 评测","）：",[41,2480,2481,2487,2494,2501,2507],{},[44,2482,2483,2486],{},[47,2484,2485],{},"核心需求是 RAG 精度"," → FastGPT",[44,2488,2489,1377,2492],{},[47,2490,2491],{},"需要丰富插件 + 复杂工作流 + 多平台发布",[522,2493,274],{"href":2326},[44,2495,2496,1377,2499],{},[47,2497,2498],{},"零代码、快速发布到飞书 \u002F 微信",[522,2500,648],{"href":647},[44,2502,2503,2506],{},[47,2504,2505],{},"文档解析（含 OCR \u002F 表格 \u002F 公式）是瓶颈"," → RAGFlow",[44,2508,2509,2512],{},[47,2510,2511],{},"桌面 \u002F 单机使用"," → AnythingLLM",[31,2514,2515,2518],{},[47,2516,2517],{},"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。",[26,2520,1405],{"id":1405},[41,2522,2523,2536,2545,2551,2561,2573,2579,2585],{},[44,2524,2525,1422,2528,2531,2532,2535],{},[47,2526,2527],{},"docker-compose 镜像 tag 不一致",[522,2529,1954],{"href":1744,"rel":2530},[550]," 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 ",[236,2533,2534],{},"image:"," 行为正确版本即可",[44,2537,2538,2541,2542,2544],{},[47,2539,2540],{},"3000 端口冲突","：默认占用 3000（主服务）\u002F 9000（S3 \u002F MinIO）\u002F 3005（MCP）；改 ",[236,2543,2133],{}," 的 ports 映射端口",[44,2546,2547,2550],{},[47,2548,2549],{},"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",[44,2552,2553,2556,2557,2560],{},[47,2554,2555],{},"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 ",[522,2558,2308],{"href":1770,"rel":2559},[550],"：\"新手 \u002F 小规模 PgVector，中大规模 Milvus，企业 \u002F 国产 OceanBase\"",[44,2562,2563,1422,2566,2569,2570],{},[47,2564,2565],{},"MinIO 默认密码",[236,2567,2568],{},"minioadmin\u002Fminioadmin","，",[47,2571,2572],{},"部署到公网前必须改",[44,2574,2575,2578],{},[47,2576,2577],{},"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\"",[44,2580,2581,2584],{},[47,2582,2583],{},"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",[44,2586,2587,2590,2591,2595],{},[47,2588,2589],{},"云版 AI 积分会过期","：未用完不能跨月累积（按 ",[522,2592,2594],{"href":2168,"rel":2593},[550],"fastgpt.io 定价 FAQ","）",[26,2597,455],{"id":454},[31,2599,1483],{},[41,2601,2602,2605,2608,2611,2614,2617],{},[44,2603,2604],{},"企业内部知识库（员工手册 \u002F 制度 \u002F 流程）",[44,2606,2607],{},"产品 FAQ \u002F 用户手册问答",[44,2609,2610],{},"医疗 \u002F 法律 \u002F 金融垂直领域知识系统",[44,2612,2613],{},"数据严格不出网 + Apache 2.0 商用",[44,2615,2616],{},"有 docker 运维基础的技术团队",[44,2618,2619],{},"需要把 RAG 当后端服务的开发者（API 接入业务系统）",[31,2621,1506],{},[41,2623,2624,2629,2634,2637],{},[44,2625,2626,2627,2595],{},"完全非技术用户（去 ",[522,2628,648],{"href":647},[44,2630,2631,2632,2595],{},"主要需求是工作流 + 插件集成（去 ",[522,2633,274],{"href":2326},[44,2635,2636],{},"文档解析 \u002F OCR 是首要痛点（RAGFlow）",[44,2638,2639],{},"不想自己运维 + 量很小（FastGPT 云免费版起步即可）",[26,2641,518],{"id":518},[41,2643,2644,2653,2670,2689],{},[44,2645,1542,2646,1545,2648,2650,2651],{},[522,2647,274],{"href":2326},[522,2649,648],{"href":647}," \u002F RAGFlow \u002F AnythingLLM \u002F ",[522,2652,1231],{"href":1230},[44,2654,2655,2656,1545,2658,1545,2662,1545,2665,1545,2668],{},"概念：",[522,2657,1562],{"href":1561},[522,2659,2661],{"href":2660},"\u002Fwiki\u002Fembedding.html","Embedding",[522,2663,2664],{"href":2660},"Vector Database",[522,2666,2667],{"href":1561},"Reranker",[522,2669,1558],{"href":1557},[44,2671,2672,2673,1545,2675,1545,2679,1545,2681,1545,2685],{},"模型：",[522,2674,1584],{"href":1583},[522,2676,2678],{"href":2677},"\u002Fmodels\u002Fqwen-3.html","Qwen3",[522,2680,1588],{"href":1587},[522,2682,2684],{"href":2683},"\u002Fmodels\u002Fkimi-k2.html","Kimi K2",[522,2686,2688],{"href":2687},"\u002Fmodels\u002Fdoubao-1-5-pro.html","豆包 Doubao",[44,2690,1591,2691,1545,2693],{},[522,2692,1599],{"href":1598},[522,2694,2696],{"href":2695},"\u002Fwiki\u002Fprompt-engineering.html","Prompt Engineering",[26,2698,536],{"id":536},[41,2700,2701,2707,2713,2719],{},[44,2702,1606,2703],{},[522,2704,2705],{"href":2705,"rel":2706},"https:\u002F\u002Ffastgpt.io",[550],[44,2708,1619,2709],{},[522,2710,2711],{"href":2711,"rel":2712},"https:\u002F\u002Fgithub.com\u002Flabring\u002FFastGPT",[550],[44,2714,2715,2716],{},"定价：",[522,2717,2168],{"href":2168,"rel":2718},[550],[44,2720,2721],{},"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",[31,2723,2724,2725,1640],{},"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 ",[522,2726,1639],{"href":1638},[1642,2728,2729],{},"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":560,"searchDepth":561,"depth":561,"links":2731},[2732,2733,2741,2742,2743,2744,2745,2746,2747],{"id":28,"depth":564,"text":29},{"id":657,"depth":564,"text":657,"children":2734},[2735,2736,2737,2738,2739,2740],{"id":1778,"depth":561,"text":1779},{"id":1828,"depth":561,"text":1829},{"id":1878,"depth":561,"text":1879},{"id":1942,"depth":561,"text":1942},{"id":1948,"depth":561,"text":1948},{"id":2011,"depth":561,"text":2012},{"id":2046,"depth":564,"text":2047},{"id":2162,"depth":564,"text":2163},{"id":1207,"depth":564,"text":1207},{"id":1405,"depth":564,"text":1405},{"id":454,"depth":564,"text":455},{"id":518,"depth":564,"text":518},{"id":536,"depth":564,"text":536},"\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。",[584,583],{},[2753,2754,2755,2756,2757,2758],"deepseek-v3","qwen-max","doubao-pro","gpt-4o","claude-sonnet-4","kimi",[2760,2761,2762],"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",[1675,1676,590],[2766,2771,2776,2781,2785],{"plan":2767,"price":863,"limit":2768,"cn_pay":2769,"note":2770},"Self-host 开源","全功能 + 全数据本地","—","Apache 2.0 可商用",{"plan":2772,"price":2773,"limit":2774,"cn_pay":2769,"note":2775},"云免费版","¥0\u002F月","100 AI 积分 + 600 索引 + 3 知识库","试水",{"plan":2777,"price":2232,"limit":2778,"cn_pay":2779,"note":2780},"云基础版","4000 积分 + 6000 索引 + 50 Agent","✅ 微信\u002F支付宝","中小团队 SaaS",{"plan":2782,"price":2257,"limit":2783,"cn_pay":307,"note":2784},"云高级版","25000 积分 + 36000 索引 + 50 成员 + 200 Agent + 1500 QPM","企业级生产",{"plan":2786,"price":2281,"limit":2787,"cn_pay":307,"note":2788},"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",[2791],"onboarding\u002Ffastgpt-getting-started",[1700,1697,1698,1699],{"power":576,"ux":576,"price":595,"cn_support":595,"stability":576},{"title":277,"description":2749},"FastGPT 评测 2026：开源知识库问答平台，AI 工作流引擎，对比 Dify",[2797,2799,2801,2803,2805],{"title":2798,"url":2705},"FastGPT 官网",{"title":2800,"url":2711},"FastGPT GitHub",{"title":2802,"url":2168},"FastGPT 定价页",{"title":2804,"url":1770},"FastGPT 2025 测评（南环 AI）",{"title":2806,"url":1744},"FastGPT 部署教程（腾讯云）","tools\u002Fagent\u002Fplatform\u002Ffastgpt",[2809,2810,2811,2812,2813],"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",[606,607,610,608,2816,2817,2818],"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。","3k7sqE3ueQ8CMIEFNqawWzvX4hGkRy7UVY9-HChmMJM",{"id":2822,"title":280,"alternatives":2823,"api_compatible":2824,"body":2834,"category":575,"chinese_friendly":561,"cover":3296,"description":3297,"domestic":596,"extension":580,"faq":581,"free":579,"github":3281,"languages":3298,"lastVerified":585,"meta":3299,"models":581,"navigation":579,"notSuitable":581,"opensource":579,"path":3300,"pillar":588,"platforms":3301,"priceTable":581,"pricing":3302,"published":593,"relatedPlaybooks":581,"relatedReviews":581,"score":3303,"self_host":596,"seo":3304,"seoTitle":3305,"slug":14,"sources":3306,"stem":3309,"suitable":581,"tagline":3310,"tags":3311,"updated":585,"verdict":3313,"website":3275,"__hash__":3314},"tools\u002Ftools\u002Fagent\u002Fplatform\u002Fanythingllm.md",[12,13,619],[16,17,2825,2826,2827,2828,19,18,2829,20,2830,21,2831,2832,2833],"Google","Grok","Mistral","Cohere","腾讯混元","字节豆包","智谱 GLM","Ollama","Hugging Face",{"type":23,"value":2835,"toc":3283},[2836,2838,2841,2844,2846,2899,2901,2943,2947,2949,2955,2979,2983,3003,3005,3033,3035,3150,3152,3193,3195,3224,3226,3232,3238,3244,3250,3252,3263,3265,3269],[26,2837,29],{"id":28},[31,2839,2840],{},"AnythingLLM 是 Mintplex Labs 出品的开源私有部署 LLM 平台，MIT 协议，主打\"一站式 RAG 知识库 + Agent + 多用户权限管理\"。桌面应用 \u002F Docker 双部署模式，接入 OpenAI \u002F Claude \u002F Ollama \u002F Azure 等任意模型，Workspaces 隔离不同知识库，内置向量数据库。适合需要私有化部署 AI 知识库且不写代码的团队。",[31,2842,2843],{},"适合：企业内网知识库、团队共享 AI 助手、需要多用户权限控制的私有化场景。不适合：需要复杂 Agent 编排（用 Dify \u002F Langflow）、需要极致 RAG 召回精度（用 RAGFlow \u002F FastGPT）、需要大规模并发生产服务。",[26,2845,39],{"id":39},[41,2847,2848,2854,2860,2866,2872,2877,2883,2888,2894],{},[44,2849,2850,2853],{},[47,2851,2852],{},"私有化部署","：Docker \u002F 桌面应用（Win\u002FMac\u002FLinux），数据完全在内网",[44,2855,2856,2859],{},[47,2857,2858],{},"Workspaces 知识库隔离","：不同工作区独立向量库 + 文档 + 对话历史",[44,2861,2862,2865],{},[47,2863,2864],{},"多用户权限管理","：管理员 \u002F 用户 \u002F 多工作区角色分配，适合团队使用",[44,2867,2868,2871],{},[47,2869,2870],{},"任意模型接入","：OpenAI \u002F Claude \u002F Azure \u002F Ollama \u002F LM Studio \u002F 本地模型",[44,2873,2874,2876],{},[47,2875,85],{},"：内置 LanceDB，可选 Chroma \u002F Pinecone \u002F Weaviate \u002F Qdrant",[44,2878,2879,2882],{},[47,2880,2881],{},"文档处理","：PDF \u002F Word \u002F Excel \u002F TXT \u002F Markdown \u002F 网页链接，自动切片 + 向量化",[44,2884,2885,2887],{},[47,2886,361],{},"：内置 Web 搜索 \u002F RAG 搜索 \u002F SQL 查询等工具调用",[44,2889,2890,2893],{},[47,2891,2892],{},"嵌入向量","：支持自定义 embedding 模型，兼容 OpenAI \u002F 本地嵌入",[44,2895,2896,2898],{},[47,2897,97],{},"：提供完整 REST API，可集成到外部系统",[26,2900,101],{"id":101},[103,2902,2903,2913],{},[106,2904,2905],{},[109,2906,2907,2909,2911],{},[112,2908,114],{},[112,2910,101],{},[112,2912,119],{},[121,2914,2915,2924,2934],{},[109,2916,2917,2919,2921],{},[126,2918,128],{},[126,2920,131],{},[126,2922,2923],{},"完整功能，MIT 协议，自托管",[109,2925,2926,2928,2931],{},[126,2927,139],{},[126,2929,2930],{},"$30\u002F月起",[126,2932,2933],{},"托管服务，免去运维，含团队协作",[109,2935,2936,2938,2940],{},[126,2937,150],{},[126,2939,153],{},[126,2941,2942],{},"SSO \u002F 审计日志 \u002F 私有部署支持",[158,2944,2945],{},[31,2946,162],{},[26,2948,166],{"id":165},[158,2950,2951],{},[31,2952,171,2953],{},[47,2954,174],{},[41,2956,2957,2964,2967,2970,2973,2976],{},[44,2958,2959,2960,2963],{},"Docker 部署极快，一条 ",[236,2961,2962],{},"docker-compose up"," 起来就能用",[44,2965,2966],{},"Workspaces 隔离设计实用，不同部门知识库互不干扰",[44,2968,2969],{},"接 Ollama 本地模型完全离线运行，数据不出内网",[44,2971,2972],{},"桌面应用适合个人用户，安装即用零配置",[44,2974,2975],{},"文档上传后自动切片 + 向量化，问答效果在通用场景下可接受",[44,2977,2978],{},"多用户权限管理是开源 RAG 平台中少有的完整实现",[31,2980,2981],{},[47,2982,199],{},[41,2984,2985,2988,2991,2994,2997,3000],{},[44,2986,2987],{},"文档切片策略偏简单（固定长度），复杂表格 \u002F 图文混排召回效果一般",[44,2989,2990],{},"大文件（100MB+ PDF）处理偶尔超时，需调超时参数",[44,2992,2993],{},"Agent 能力有限，复杂工具链编排不如 Dify",[44,2995,2996],{},"向量库默认 LanceDB 在数据量大时查询变慢，建议切 Qdrant \u002F Chroma",[44,2998,2999],{},"UI 偶有卡顿，文档列表加载慢",[44,3001,3002],{},"中文文档的 OCR 需要额外配置，默认对扫描件支持有限",[26,3004,225],{"id":225},[227,3006,3007,3014,3021,3024,3027,3030],{},[44,3008,3009,3010,3013],{},"Docker 部署：",[236,3011,3012],{},"docker-compose up -d","（官方提供 docker-compose.yml）",[44,3015,3016,3017,3020],{},"首次访问 ",[236,3018,3019],{},"http:\u002F\u002Flocalhost:3001","，创建管理员账号",[44,3022,3023],{},"Settings → LLM Provider 配置模型（OpenAI API Key 或 Ollama 地址）",[44,3025,3026],{},"创建 Workspace → 上传文档（PDF\u002FWord\u002FTXT）",[44,3028,3029],{},"等待文档向量化完成，在 Chat 中开始问答",[44,3031,3032],{},"Settings → Users 添加团队成员并分配工作区权限",[26,3034,260],{"id":260},[103,3036,3037,3051],{},[106,3038,3039],{},[109,3040,3041,3043,3045,3047,3049],{},[112,3042,269],{},[112,3044,280],{},[112,3046,274],{},[112,3048,277],{},[112,3050,1402],{},[121,3052,3053,3067,3081,3093,3107,3121,3134],{},[109,3054,3055,3058,3061,3063,3065],{},[126,3056,3057],{},"部署门槛",[126,3059,3060],{},"极低",[126,3062,293],{},[126,3064,293],{},[126,3066,293],{},[109,3068,3069,3072,3075,3077,3079],{},[126,3070,3071],{},"多用户权限",[126,3073,3074],{},"✅ 完整",[126,3076,307],{},[126,3078,307],{},[126,3080,310],{},[109,3082,3083,3085,3087,3089,3091],{},[126,3084,1295],{},[126,3086,293],{},[126,3088,337],{},[126,3090,337],{},[126,3092,293],{},[109,3094,3095,3098,3100,3102,3104],{},[126,3096,3097],{},"Agent 编排",[126,3099,370],{},[126,3101,296],{},[126,3103,293],{},[126,3105,3106],{},"强（可视化）",[109,3108,3109,3112,3115,3117,3119],{},[126,3110,3111],{},"模型接入",[126,3113,3114],{},"丰富",[126,3116,3114],{},[126,3118,3114],{},[126,3120,3114],{},[109,3122,3123,3126,3128,3130,3132],{},[126,3124,3125],{},"桌面应用",[126,3127,307],{},[126,3129,310],{},[126,3131,310],{},[126,3133,310],{},[109,3135,3136,3139,3142,3145,3148],{},[126,3137,3138],{},"协议",[126,3140,3141],{},"MIT",[126,3143,3144],{},"Apache 2.0",[126,3146,3147],{},"FastGPT Open",[126,3149,3141],{},[26,3151,403],{"id":403},[41,3153,3154,3160,3166,3172,3181,3187],{},[44,3155,3156,3159],{},[47,3157,3158],{},"切片策略默认偏简单","：对结构化文档（表格\u002F代码）效果差，可调 chunk size",[44,3161,3162,3165],{},[47,3163,3164],{},"LanceDB 大数据量变慢","：文档超过 1 万条建议切 Qdrant 或 Chroma",[44,3167,3168,3171],{},[47,3169,3170],{},"大文件超时","：调整 Docker 超时配置，或拆分文档上传",[44,3173,3174,3177,3178],{},[47,3175,3176],{},"Ollama 连接","：Docker 内访问宿主机 Ollama 需用 ",[236,3179,3180],{},"host.docker.internal",[44,3182,3183,3186],{},[47,3184,3185],{},"embedding 模型选择","：中文场景建议用 bge-large-zh 而非默认 OpenAI embedding",[44,3188,3189,3192],{},[47,3190,3191],{},"不要当生产级 Agent 平台用","：Agent 能力是辅助，复杂编排上 Dify",[26,3194,455],{"id":454},[41,3196,3197,3200,3203,3206,3209,3212,3215,3218,3221],{},[44,3198,3199],{},"✅ 企业内网私有化 AI 知识库",[44,3201,3202],{},"✅ 团队共享 AI 助手 + 多用户权限管理",[44,3204,3205],{},"✅ 接 Ollama 完全离线运行",[44,3207,3208],{},"✅ 个人桌面端快速体验 RAG",[44,3210,3211],{},"✅ 需要快速验证 RAG 概念的原型项目",[44,3213,3214],{},"❌ 需要复杂 Agent 工作流编排（用 Dify \u002F Langflow）",[44,3216,3217],{},"❌ 需要极致 RAG 召回精度（用 RAGFlow \u002F FastGPT）",[44,3219,3220],{},"❌ 大规模并发生产服务（架构未做高可用）",[44,3222,3223],{},"❌ 需要深度文档解析（复杂表格\u002F公式\u002F扫描件）",[26,3225,491],{"id":490},[31,3227,3228,3231],{},[47,3229,3230],{},"Q: AnythingLLM 和 Dify 怎么选？","\nA: AnythingLLM 更轻量，部署快、有桌面应用、多用户权限开箱即用，适合快速搭建团队知识库。Dify 功能更全面，Agent 编排、工作流、API 发布能力更强，适合需要构建复杂 AI 应用的团队。简单知识库选 AnythingLLM，复杂应用选 Dify。",[31,3233,3234,3237],{},[47,3235,3236],{},"Q: 可以完全离线使用吗？","\nA: 可以。接 Ollama 本地模型 + 用本地 embedding 模型（如 bge-large-zh）+ 内置 LanceDB 向量库，整个系统完全离线运行，数据不出内网。适合数据敏感的企业场景。",[31,3239,3240,3243],{},[47,3241,3242],{},"Q: 免费开源版有什么限制？","\nA: MIT 协议开源版功能完整，无用户数 \u002F 文档数 \u002F API 调用限制。Cloud 版和 Enterprise 版主要是托管服务和企业管理功能（SSO \u002F 审计日志），功能层面开源版已够用。",[31,3245,3246,3249],{},[47,3247,3248],{},"Q: 支持中文文档吗？","\nA: 支持，但效果取决于 embedding 模型。默认 OpenAI embedding 对中文尚可，追求精度建议切换 bge-large-zh 或 m3e 模型。OCR 扫描件需额外配置 Tesseract 或接入外部 OCR 服务。",[26,3251,518],{"id":518},[31,3253,3254,525,3257,525,3259],{},[522,3255,10],{"href":3256},"\u002Fagent\u002Fplatform\u002Fragflow.html",[522,3258,529],{"href":528},[522,3260,3262],{"href":3261},"\u002Fagent\u002Fgeneral\u002Fperplexity.html","Perplexity",[26,3264,536],{"id":536},[158,3266,3267],{},[31,3268,541],{},[41,3270,3271,3277],{},[44,3272,3273],{},[522,3274,551],{"href":3275,"rel":3276},"https:\u002F\u002Fuseanything.com",[550],[44,3278,3279],{},[522,3280,558],{"href":3281,"rel":3282},"https:\u002F\u002Fgithub.com\u002FMintplex-Labs\u002Fanything-llm",[550],{"title":560,"searchDepth":561,"depth":561,"links":3284},[3285,3286,3287,3288,3289,3290,3291,3292,3293,3294,3295],{"id":28,"depth":564,"text":29},{"id":39,"depth":564,"text":39},{"id":101,"depth":564,"text":101},{"id":165,"depth":564,"text":166},{"id":225,"depth":564,"text":225},{"id":260,"depth":564,"text":260},{"id":403,"depth":564,"text":403},{"id":454,"depth":564,"text":455},{"id":490,"depth":564,"text":491},{"id":518,"depth":564,"text":518},{"id":536,"depth":564,"text":536},"\u002Fimg\u002Ftools\u002Fanythingllm.webp","AnythingLLM 真实评测：Mintplex Labs 出品的开源私有部署 LLM 平台（MIT 协议），一站式 RAG 知识库 + Agent + 多用户权限管理。支持 Docker\u002F桌面部署，接入 OpenAI\u002FClaude\u002FOllama 等任意模型，适合企业内网私有化 AI 知识库场景。",[583],{},"\u002Ftools\u002Fagent\u002Fplatform\u002Fanythingllm",[1675,1676,590,591],"Free \u002F 开源（MIT）\u002F Cloud",{"power":561,"ux":576,"price":595,"cn_support":561,"stability":561},{"title":280,"description":3297},"AnythingLLM - 开源私有部署 LLM 平台评测 | AIHO",[3307,3308],{"title":551,"url":3275},{"title":558,"url":3281},"tools\u002Fagent\u002Fplatform\u002Fanythingllm","开源私有部署 LLM 平台，一站式 RAG + Agent + 多用户",[606,607,610,608,3312],"multi-user","需要快速搭建私有化 AI 知识库且要求多用户权限管理的企业团队首选，MIT 协议 + 桌面\u002FDocker 双模式 + 任意模型接入降低了部署门槛，但 RAG 精度和 Agent 编排能力不及 Dify\u002FFastGPT 等专业平台。","w0FYQb9Q4nM6-GQPZyzHTVhhjdZKUXmEmR920GlQFQ4",1785660639519]