[{"data":1,"prerenderedAt":5132},["ShallowReactive",2],{"header-counts":3,"footer-counts":6,"tool-\u002Ftools\u002Fcoding\u002Flocal\u002Fgpt4all":8,"cat-rank-coding-local":495,"tool-related-coding\u002Flocal\u002Fgpt4all":4100,"tool-reviews-coding\u002Flocal\u002Fgpt4all":4101,"tool-alts-coding\u002Flocal\u002Fgpt4all":4102},{"tools":4,"reviews":5},98,27,{"tools":4,"reviews":5,"playbooks":7,"news":5},22,{"id":9,"title":10,"alternatives":11,"api_compatible":15,"body":16,"category":460,"chinese_friendly":449,"cover":461,"description":462,"domestic":463,"extension":464,"faq":15,"free":463,"github":441,"languages":465,"lastVerified":467,"meta":468,"models":15,"navigation":469,"notSuitable":15,"opensource":469,"path":470,"pillar":471,"platforms":472,"priceTable":15,"pricing":476,"published":477,"relatedPlaybooks":15,"relatedReviews":15,"score":478,"self_host":463,"seo":481,"seoTitle":482,"slug":483,"sources":484,"stem":487,"suitable":15,"tagline":488,"tags":489,"updated":467,"verdict":493,"website":433,"__hash__":494},"tools\u002Ftools\u002Fcoding\u002Flocal\u002Fgpt4all.md","GPT4All",[12,13,14],"coding\u002Flocal\u002Follama","coding\u002Flocal\u002Flm-studio","coding\u002Flocal\u002Fjan",null,{"type":17,"value":18,"toc":444},"minimark",[19,24,33,36,39,91,94,100,133,136,140,148,165,170,187,190,215,218,316,319,351,355,378,382,388,394,400,403,419,422,427],[20,21,23],"h2",{"id":22},"tldr","TL;DR",[25,26,27,28,32],"p",{},"GPT4All 是 Nomic AI 出品的本地 LLM 推理工具，最大卖点是",[29,30,31],"strong",{},"CPU 也能跑","——没有独立显卡的普通笔记本 \u002F 办公本照样运行本地大模型。MIT 开源，桌面客户端一键下载 GGUF 模型，开箱即用。适合无 GPU 设备、隐私优先、轻量本地推理场景。",[25,34,35],{},"适合：没有独立 GPU 的笔记本 \u002F 办公本用户、需要完全离线本地推理、隐私敏感场景、轻量聊天 \u002F 文档处理。不适合：需要高速推理（CPU 太慢）、需要 OpenAI 兼容 API 给应用接入（用 Ollama）、需要大模型（32B+，CPU 跑不动）。",[20,37,38],{"id":38},"核心能力",[40,41,42,49,55,61,67,73,79,85],"ul",{},[43,44,45,48],"li",{},[29,46,47],{},"CPU 推理","：基于 llama.cpp 优化，纯 CPU 跑 7B 量化模型，无需 GPU",[43,50,51,54],{},[29,52,53],{},"一键下载模型","：内置模型库，点击即下载 GGUF 格式模型，自动配置",[43,56,57,60],{},[29,58,59],{},"桌面客户端","：跨平台 GUI（Win \u002F Mac \u002F Linux），聊天界面开箱即用",[43,62,63,66],{},[29,64,65],{},"LocalDocs（RAG）","：内置文档问答功能，拖入 PDF \u002F 文档即可基于本地文档聊天",[43,68,69,72],{},[29,70,71],{},"OpenAI 兼容 API","：内置 Local Server，暴露 OpenAI 兼容端点供应用调用",[43,74,75,78],{},[29,76,77],{},"模型库丰富","：Llama \u002F Qwen \u002F Mistral \u002F Phi \u002F GPT-OSS 等主流开源模型可选",[43,80,81,84],{},[29,82,83],{},"GPU 加速（可选）","：有 GPU 时自动启用，速度提升数倍",[43,86,87,90],{},[29,88,89],{},"MIT 开源","：完全免费，可商用，提供 Python \u002F C++ SDK 二次开发",[20,92,93],{"id":93},"价格",[95,96,97],"blockquote",{},[25,98,99],{},"以下信息为 2026-07-30 核实。",[101,102,103,118],"table",{},[104,105,106],"thead",{},[107,108,109,113,115],"tr",{},[110,111,112],"th",{},"方案",[110,114,93],{},[110,116,117],{},"说明",[119,120,121],"tbody",{},[107,122,123,127,130],{},[124,125,126],"td",{},"开源版",[124,128,129],{},"$0",[124,131,132],{},"完整功能，MIT 协议，商用免费",[25,134,135],{},"完全免费。成本在于硬件（CPU \u002F 内存）和你选用的模型。",[20,137,139],{"id":138},"体验与评测资料整理","体验与评测（资料整理）",[95,141,142],{},[25,143,144,145],{},"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",[29,146,147],{},"亮点：",[40,149,150,153,156,159,162],{},[43,151,152],{},"纯 CPU 跑 Qwen2.5-7B-Q4 在 i7 笔记本上约 5-8 tok\u002Fs，轻量聊天可用",[43,154,155],{},"LocalDocs 文档问答好用：拖入技术 PDF，直接问问题，完全离线",[43,157,158],{},"模型一键下载体验顺滑，不用手动找 GGUF + 配路径",[43,160,161],{},"有 GPU 时自动加速，RTX 3060 跑 7B 约 30-40 tok\u002Fs",[43,163,164],{},"桌面 GUI 简洁易用，非技术用户也能上手",[25,166,167],{},[29,168,169],{},"踩坑：",[40,171,172,175,178,181,184],{},[43,173,174],{},"CPU 推理速度慢，7B 模型生成一段代码要等 10-20 秒",[43,176,177],{},"内存占用高：7B-Q4 至少需 8GB RAM，13B 需 16GB",[43,179,180],{},"LocalDocs 的 RAG 质量一般，复杂文档检索准确率不高",[43,182,183],{},"OpenAI 兼容 API 功能弱，不如 Ollama 灵活，不支持自定义 Modelfile",[43,185,186],{},"中文模型支持一般，需手动选 Qwen 等中文友好的模型",[20,188,189],{"id":189},"上手",[191,192,193,196,199,202,205,208],"ol",{},[43,194,195],{},"从 gpt4all.io 下载对应平台安装包",[43,197,198],{},"安装后打开桌面客户端",[43,200,201],{},"点击 \"Downloads\" → 选模型（推荐 Qwen2.5-7B-Q4 或 Llama3.1-8B-Q4）",[43,203,204],{},"下载完成后回到 Chat → 选模型 → 开始聊天",[43,206,207],{},"文档问答：LocalDocs → 添加文档文件夹 → 在聊天中勾选引用",[43,209,210,211],{},"API 接入：Settings → 启用 API Server → 端点 ",[212,213,214],"code",{},"http:\u002F\u002Flocalhost:4891\u002Fv1",[20,216,217],{"id":217},"对比",[101,219,220,235],{},[104,221,222],{},[107,223,224,227,229,232],{},[110,225,226],{},"维度",[110,228,10],{},[110,230,231],{},"Ollama",[110,233,234],{},"LM Studio",[119,236,237,249,262,276,289,303],{},[107,238,239,241,244,247],{},[124,240,47],{},[124,242,243],{},"✅ 优化好",[124,245,246],{},"✅",[124,248,246],{},[107,250,251,254,257,260],{},[124,252,253],{},"GUI",[124,255,256],{},"✅ 桌面",[124,258,259],{},"❌ CLI",[124,261,246],{},[107,263,264,267,270,273],{},[124,265,266],{},"模型下载",[124,268,269],{},"✅ 内置库",[124,271,272],{},"✅ CLI",[124,274,275],{},"✅ GUI 库",[107,277,278,281,284,287],{},[124,279,280],{},"OpenAI API",[124,282,283],{},"基础",[124,285,286],{},"强（Modelfile）",[124,288,246],{},[107,290,291,294,297,300],{},[124,292,293],{},"文档 RAG",[124,295,296],{},"✅ LocalDocs",[124,298,299],{},"需配 WebUI",[124,301,302],{},"需插件",[107,304,305,308,311,313],{},[124,306,307],{},"开源",[124,309,310],{},"MIT",[124,312,310],{},[124,314,315],{},"闭源",[20,317,318],{"id":318},"避坑",[40,320,321,327,333,339,345],{},[43,322,323,326],{},[29,324,325],{},"CPU 用户别跑大模型","：13B 以上 CPU 跑基本不可用，坚持 7B 量化",[43,328,329,332],{},[29,330,331],{},"LocalDocs 别放太多文件","：文件多了索引慢且检索质量下降，分批放",[43,334,335,338],{},[29,336,337],{},"API Server 别对公网开","：默认无鉴权，仅本地用",[43,340,341,344],{},[29,342,343],{},"内存不够会崩","：8GB RAM 只能跑 7B-Q4，别勉强 13B",[43,346,347,350],{},[29,348,349],{},"中文场景选 Qwen","：Llama 系列中文能力弱，Qwen \u002F GLM 更合适",[20,352,354],{"id":353},"适合-不适合","适合 \u002F 不适合",[40,356,357,360,363,366,369,372,375],{},[43,358,359],{},"✅ 没有独立 GPU 的笔记本 \u002F 办公本",[43,361,362],{},"✅ 需要完全离线 + 隐私优先的本地推理",[43,364,365],{},"✅ 轻量聊天 \u002F 文档问答场景",[43,367,368],{},"✅ 非技术用户（GUI 友好）",[43,370,371],{},"❌ 需要高速推理（有 GPU 直接 Ollama）",[43,373,374],{},"❌ 需要给应用 \u002F IDE 接入 API（用 Ollama）",[43,376,377],{},"❌ 需要跑大模型（32B+，CPU 跑不动）",[20,379,381],{"id":380},"faq","FAQ",[25,383,384,387],{},[29,385,386],{},"Q: CPU 跑 7B 模型速度能接受吗？","\nA: 看用途。轻量聊天 \u002F 短问答可以接受（5-8 tok\u002Fs）。但代码生成 \u002F 长文本输出等待时间较长（一段代码 10-20 秒）。有 GPU 强烈建议用 GPU。",[25,389,390,393],{},[29,391,392],{},"Q: GPT4All 和 Ollama 怎么选？","\nA: 没 GPU + 要 GUI → GPT4All。有 GPU + 要 API 接入应用 → Ollama。两者底层都基于 llama.cpp，模型格式通用（GGUF）。",[25,395,396,399],{},[29,397,398],{},"Q: LocalDocs 文档问答好用吗？","\nA: 基础可用但 RAG 质量一般。简单 PDF 问答没问题，复杂文档 \u002F 多文件检索准确率不高。专业 RAG 需求建议配 Open WebUI + 向量数据库。",[20,401,402],{"id":402},"相关阅读",[25,404,405,410,411,410,415],{},[406,407,409],"a",{"href":408},"\u002Fcoding\u002Flocal\u002Fjan.html","Jan"," · ",[406,412,414],{"href":413},"\u002Fcoding\u002Flocal\u002Fvllm.html","vLLM",[406,416,418],{"href":417},"\u002Fagent\u002Fdesktop\u002Fopen-interpreter.html","Open Interpreter",[20,420,421],{"id":421},"来源",[95,423,424],{},[25,425,426],{},"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。",[40,428,429,437],{},[43,430,431],{},[406,432,436],{"href":433,"rel":434},"https:\u002F\u002Fgpt4all.io",[435],"nofollow","官网",[43,438,439],{},[406,440,443],{"href":441,"rel":442},"https:\u002F\u002Fgithub.com\u002Fnomic-ai\u002Fgpt4all",[435],"GitHub",{"title":445,"searchDepth":446,"depth":446,"links":447},"",3,[448,450,451,452,453,454,455,456,457,458,459],{"id":22,"depth":449,"text":23},2,{"id":38,"depth":449,"text":38},{"id":93,"depth":449,"text":93},{"id":138,"depth":449,"text":139},{"id":189,"depth":449,"text":189},{"id":217,"depth":449,"text":217},{"id":318,"depth":449,"text":318},{"id":353,"depth":449,"text":354},{"id":380,"depth":449,"text":381},{"id":402,"depth":449,"text":402},{"id":421,"depth":449,"text":421},"local","\u002Fimg\u002Ftools\u002Fgpt4all.webp","GPT4All 真实评测：Nomic AI 出品的本地 LLM 推理工具（MIT 协议），支持 CPU 运行无需 GPU，GGUF 格式模型一键下载。跨平台桌面客户端，适合没有独立显卡、需要在普通笔记本上跑本地大模型的用户。",false,"md",[466],"en","2026-07-30",{},true,"\u002Ftools\u002Fcoding\u002Flocal\u002Fgpt4all","coding",[473,474,475],"windows","macos","linux","Free \u002F 开源（MIT）","2026-07-05",{"power":446,"ux":479,"price":480,"cn_support":449,"stability":479},4,5,{"title":10,"description":462},"GPT4All - 本地 LLM 推理引擎评测与使用 | AIHO","coding\u002Flocal\u002Fgpt4all",[485,486],{"title":436,"url":433},{"title":443,"url":441},"tools\u002Fcoding\u002Flocal\u002Fgpt4all","本地 LLM 推理，Nomic AI 出品，支持 CPU 运行",[460,490,491,492],"cpu-inference","opensource","gguf","没有 GPU 也能跑本地大模型的首选，CPU 推理 + 一键下载模型体验顺滑；但推理速度慢、API 能力弱，有 GPU 用户建议直接 Ollama。","oTtyXhNkQBIi7aJI5YJrsiu_c1A81ifErtlA5QOs9UI",[496,976,1292,1706,2183,2670,3133,3598],{"id":497,"title":498,"alternatives":499,"api_compatible":15,"body":502,"category":460,"chinese_friendly":480,"cover":917,"description":918,"domestic":463,"extension":464,"faq":919,"free":463,"github":15,"languages":932,"lastVerified":15,"meta":934,"models":15,"navigation":469,"notSuitable":15,"opensource":469,"path":935,"pillar":471,"platforms":936,"priceTable":938,"pricing":947,"published":948,"relatedPlaybooks":949,"relatedReviews":15,"score":952,"self_host":469,"seo":953,"seoTitle":954,"slug":955,"sources":956,"stem":964,"suitable":15,"tagline":965,"tags":966,"updated":959,"verdict":973,"website":974,"__hash__":975},"tools\u002Ftools\u002Fcoding\u002Flocal\u002Fcherry-studio.md","Cherry Studio",[500,13,12,501],"coding\u002Flocal\u002Flobe-chat","coding\u002Flocal\u002Fopen-webui",{"type":17,"value":503,"toc":905},[504,506,509,512,514,558,560,573,578,582,586,603,607,624,626,650,652,793,795,827,829,852,854,880,882],[20,505,23],{"id":22},[25,507,508],{},"Cherry Studio 是一款开源、跨平台（Windows \u002F macOS \u002F Linux \u002F Android）的桌面 AI 客户端，定位『全能 AI 工作台』：把 OpenAI \u002F Anthropic \u002F Google \u002F DeepSeek 等云端模型，以及 Ollama \u002F LM Studio 本地模型，全部聚合到同一个桌面应用里管理。内置 300+ 助手模板、本地 RAG 知识库、Markdown + Mermaid 渲染、MCP 协议支持，所有对话数据本地存储 + WebDAV 备份。AGPL-3.0 开源、GitHub 60k+ stars，企业版可联系商务做私有化部署。",[25,510,511],{},"适合：中文 AI 重度用户、想统一管理多家模型、需要本地知识库 RAG、关注数据本地存储的开发者 \u002F 研究者。不适合：要 Web 端访问 \u002F Docker 自托管 \u002F 团队多人共享 \u002F iOS 端使用。",[20,513,38],{"id":38},[40,515,516,522,528,534,540,546,552],{},[43,517,518,521],{},[29,519,520],{},"多模型聚合","：OpenAI \u002F Claude \u002F Gemini \u002F DeepSeek \u002F Qwen \u002F Kimi \u002F Moonshot 等云端 + Ollama \u002F LM Studio 本地",[43,523,524,527],{},[29,525,526],{},"本地 RAG 知识库","：拖拽 PDF \u002F Word \u002F Excel \u002F PPT \u002F 网址 \u002F sitemap → 自动向量化 → 检索增强问答 + 来源追溯",[43,529,530,533],{},[29,531,532],{},"300+ 助手模板","：编程 \u002F 写作 \u002F 翻译 \u002F 学习 \u002F 角色扮演开箱即用，可自定义 System Prompt",[43,535,536,539],{},[29,537,538],{},"MCP 协议","：扩展工具调用 \u002F 联网搜索 \u002F 文件操作",[43,541,542,545],{},[29,543,544],{},"数据本地优先","：对话历史本地存储，WebDAV 同步，不上传第三方",[43,547,548,551],{},[29,549,550],{},"多模态","：图片识别 \u002F PDF 阅读 \u002F Markdown + Mermaid + 代码高亮",[43,553,554,557],{},[29,555,556],{},"AI 绘画 + 翻译","：内置主流 SD \u002F DALL·E \u002F 翻译 API 集成",[20,559,93],{"id":93},[40,561,562,567],{},[43,563,564,566],{},[29,565,126],{},"：完全免费，AGPL-3.0",[43,568,569,572],{},[29,570,571],{},"Enterprise","：私有化部署 + 团队协作 + 资源管控，联系销售",[95,574,575],{},[25,576,577],{},"模型 API 费用按你自己绑定的供应商计费；本地 Ollama \u002F LM Studio 零成本。",[20,579,581],{"id":580},"实测mac-m2-中型知识库","实测（Mac M2 + 中型知识库）",[25,583,584],{},[29,585,147],{},[40,587,588,591,594,597,600],{},[43,589,590],{},"中文 UI \u002F 文档 \u002F 社区都顶级，零门槛上手",[43,592,593],{},"本地 RAG 拖入 30+ PDF 后向量化 \u003C 2 分钟（用 bge-m3）",[43,595,596],{},"多模型并排回答：让 Claude \u002F GPT \u002F DeepSeek 同回一个问题做比较",[43,598,599],{},"MCP 接 Brave Search + 自定义工具流畅",[43,601,602],{},"WebDAV 同步坚果云 \u002F 阿里云盘，桌面 + 移动设备数据互通",[25,604,605],{},[29,606,169],{},[40,608,609,612,615,618,621],{},[43,610,611],{},"没有 Web 端 \u002F Docker 自托管（要这个用 LobeChat）",[43,613,614],{},"iOS 版尚未发布（roadmap 中）",[43,616,617],{},"大型 PDF（>100 MB）向量化偶有失败，要切小",[43,619,620],{},"助手市场质量参差，要自筛",[43,622,623],{},"模型 API 调用全靠你自己付费，新手要先理解 API Key 概念",[20,625,189],{"id":189},[191,627,628,631,634,641,644,647],{},[43,629,630],{},"cherry-ai.com 下载客户端（或 GitHub releases）",[43,632,633],{},"设置 → 模型服务 → 填 OpenAI \u002F Claude \u002F DeepSeek API Key",[43,635,636,637],{},"（可选）本地：装 Ollama → Cherry Studio 自动识别 endpoint ",[406,638,639],{"href":639,"rel":640},"http:\u002F\u002Flocalhost:11434",[435],[43,642,643],{},"新建知识库 → 拖文件 \u002F 加网址 → 等向量化",[43,645,646],{},"新对话 → 选模型 → 勾知识库 → 提问",[43,648,649],{},"进阶：自定义助手（System Prompt）+ MCP 扩展工具",[20,651,217],{"id":217},[101,653,654,670],{},[104,655,656],{},[107,657,658,660,662,665,667],{},[110,659,226],{},[110,661,498],{},[110,663,664],{},"LobeChat",[110,666,234],{},[110,668,669],{},"Open WebUI",[119,671,672,688,702,717,730,746,761,776],{},[107,673,674,677,680,683,685],{},[124,675,676],{},"形态",[124,678,679],{},"桌面",[124,681,682],{},"Web + 桌面",[124,684,679],{},[124,686,687],{},"Docker \u002F 桌面",[107,689,690,692,695,697,700],{},[124,691,520],{},[124,693,694],{},"✅ 云 + 本地",[124,696,694],{},[124,698,699],{},"本地为主",[124,701,694],{},[107,703,704,707,710,712,715],{},[124,705,706],{},"知识库 RAG",[124,708,709],{},"✅ 强",[124,711,709],{},[124,713,714],{},"弱",[124,716,246],{},[107,718,719,722,724,726,728],{},[124,720,721],{},"MCP",[124,723,246],{},[124,725,246],{},[124,727,714],{},[124,729,246],{},[107,731,732,735,738,741,744],{},[124,733,734],{},"自托管 \u002F Web",[124,736,737],{},"无 Web",[124,739,740],{},"✅ Docker",[124,742,743],{},"无",[124,745,740],{},[107,747,748,751,754,756,759],{},[124,749,750],{},"中文",[124,752,753],{},"5\u002F5",[124,755,753],{},[124,757,758],{},"4\u002F5",[124,760,758],{},[107,762,763,766,769,771,774],{},[124,764,765],{},"开源协议",[124,767,768],{},"AGPL-3.0",[124,770,310],{},[124,772,773],{},"闭源（免费）",[124,775,310],{},[107,777,778,781,784,787,790],{},[124,779,780],{},"GitHub Stars",[124,782,783],{},"60k+",[124,785,786],{},"72k+",[124,788,789],{},"–",[124,791,792],{},"126k+",[20,794,318],{"id":318},[40,796,797,803,809,815,821],{},[43,798,799,802],{},[29,800,801],{},"API Key 别明文外泄","：客户端配置文件以明文存 Key，机器借出前先清；团队共享用企业版 \u002F 自建中转",[43,804,805,808],{},[29,806,807],{},"知识库别一次塞太多","：单库 1000+ 文档检索质量明显下降，按主题切分多个知识库",[43,810,811,814],{},[29,812,813],{},"嵌入模型选择","：免费 bge-m3 够用；专业用付费 Pro\u002FBAAI\u002Fbge-m3 或 OpenAI text-embedding-3",[43,816,817,820],{},[29,818,819],{},"WebDAV 同步先小范围测","：知识库向量数据较大，先备份对话再开同步",[43,822,823,826],{},[29,824,825],{},"MCP 工具来源要可控","：MCP 是给 AI 真实工具能力，第三方插件审一遍代码",[20,828,354],{"id":353},[40,830,831,834,837,840,843,846,849],{},[43,832,833],{},"✅ 中文用户、AI 重度使用 \u002F 多模型管理",[43,835,836],{},"✅ 需要本地 RAG 知识库",[43,838,839],{},"✅ 关注数据隐私 \u002F 本地存储",[43,841,842],{},"✅ 想用 Ollama \u002F LM Studio 本地模型",[43,844,845],{},"❌ 需要 Web 端 \u002F Docker 自托管",[43,847,848],{},"❌ 团队多人共享 \u002F SSO",[43,850,851],{},"❌ iOS 主力用户",[20,853,402],{"id":402},[40,855,856,862,868,874],{},[43,857,858],{},[406,859,861],{"href":860},"\u002Ftools\u002Fcoding\u002Flocal\u002Flobe-chat","LobeChat 评测",[43,863,864],{},[406,865,867],{"href":866},"\u002Ftools\u002Fcoding\u002Flocal\u002Flm-studio","LM Studio 评测",[43,869,870],{},[406,871,873],{"href":872},"\u002Ftools\u002Fcoding\u002Flocal\u002Follama","Ollama 评测",[43,875,876],{},[406,877,879],{"href":878},"\u002Fplaybook\u002Fonboarding\u002Frag-pipeline-build","RAG Pipeline 搭建 Playbook",[20,881,421],{"id":421},[191,883,884,891,898],{},[43,885,886,887],{},"Cherry Studio 官网（功能 + 下载）",[406,888,889],{"href":889,"rel":890},"https:\u002F\u002Fwww.cherry-ai.com\u002F",[435],[43,892,893,894],{},"MBLUO Studio — Cherry Studio 评测 2026 ",[406,895,896],{"href":896,"rel":897},"https:\u002F\u002Fmbluostudio.com\u002Ftools\u002Fcherry-studio",[435],[43,899,900,901],{},"Cursor IDE 博客 — Cherry Studio 完全指南（2025-03）",[406,902,903],{"href":903,"rel":904},"https:\u002F\u002Fwww.cursor-ide.com\u002Fblog\u002Fcherry-studio-guide",[435],{"title":445,"searchDepth":446,"depth":446,"links":906},[907,908,909,910,911,912,913,914,915,916],{"id":22,"depth":449,"text":23},{"id":38,"depth":449,"text":38},{"id":93,"depth":449,"text":93},{"id":580,"depth":449,"text":581},{"id":189,"depth":449,"text":189},{"id":217,"depth":449,"text":217},{"id":318,"depth":449,"text":318},{"id":353,"depth":449,"text":354},{"id":402,"depth":449,"text":402},{"id":421,"depth":449,"text":421},"\u002Fimg\u002Ftools\u002Fcherry-studio.webp","Cherry Studio 真实评测：开源跨平台桌面 AI 客户端，集成 OpenAI \u002F Claude \u002F Gemini \u002F DeepSeek + Ollama \u002F LM Studio 本地模型，内置 300+ 助手模板 + 本地 RAG 知识库。AGPL-3.0 开源、GitHub 60k+ stars，企业版另询。",[920,923,926,929],{"q":921,"a":922},"Cherry Studio 真的免费吗？","是。客户端完全免费、AGPL-3.0 开源，模型调用走你自己的 API Key（OpenAI \u002F Claude \u002F Gemini \u002F DeepSeek 等付费）或本地 Ollama \u002F LM Studio（零成本）。",{"q":924,"a":925},"本地知识库怎么用？","在『知识库』面板新建，拖文件 \u002F 加网址 \u002F 填 sitemap，系统自动向量化（默认 BAAI\u002Fbge-m3 或硅基流动的 Pro 版）；提问时勾选要检索的知识库，AI 会基于检索片段答题并标出来源。",{"q":927,"a":928},"和 LobeChat 怎么选？","都开源、多模型、有 RAG。LobeChat 是 Web + 桌面双形态，可自托管 Docker，72k stars；Cherry Studio 是纯桌面（Win\u002FMac\u002FLinux\u002FAndroid），不支持 Web 部署但桌面体验更精细，60k+ stars。要 Web 访问 \u002F 公司多人共享选 LobeChat；个人重度选 Cherry Studio。",{"q":930,"a":931},"支持 MCP \u002F 插件吗？","支持 MCP（Model Context Protocol）扩展，配合自定义助手（System Prompt）可扩展工具调用、联网搜索等能力。",[933,466],"zh",{},"\u002Ftools\u002Fcoding\u002Flocal\u002Fcherry-studio",[473,474,475,937],"android",[939,943],{"plan":126,"price":940,"features":941,"notes":942},"免费","300+ 助手模板 \u002F 云端 + 本地模型 \u002F 知识库 \u002F MCP \u002F WebDAV 备份","AGPL-3.0 开源",{"plan":571,"price":944,"features":945,"notes":946},"联系销售","私有化部署 \u002F 团队协作 \u002F AI 资源管控 \u002F 知识库管理","面向企业团队","开源免费 \u002F 企业版联系销售","2026-06-19",[950,951],"onboarding\u002Frag-pipeline-build","onboarding\u002Fcursor-mcp-deep-integration",{"power":479,"ux":480,"price":480,"cn_support":480,"stability":479},{"title":498,"description":918},"Cherry Studio 评测 2026：AI 客户端工具，多模型桌面助手，开源免费","coding\u002Flocal\u002Fcherry-studio",[957,960,962],{"name":958,"url":889,"accessed":959},"Cherry Studio 官网","2026-06-24",{"name":961,"url":896,"accessed":959},"MBLUO Studio — Cherry Studio 评测",{"name":963,"url":903,"accessed":959},"Cursor IDE 博客 — Cherry Studio 指南","tools\u002Fcoding\u002Flocal\u002Fcherry-studio","全能 AI 客户端：多模型聚合 + 本地知识库 + 300+ 助手模板，跨平台桌面应用",[460,967,968,969,970,971,972],"desktop","multi-model","knowledge-base","rag","open-source","china","国产 AI 桌面客户端第一梯队，多模型聚合 + 本地 RAG + 中文体验顶级。需要 Web 部署 \u002F 自托管选 LobeChat；只要桌面体验完整选 Cherry Studio。","https:\u002F\u002Fcherry-ai.com","CVmnny2iirvdFz56djfCs7oCOn1Mjv6hZCZ5SNO0U0Q",{"id":9,"title":10,"alternatives":977,"api_compatible":15,"body":978,"category":460,"chinese_friendly":449,"cover":461,"description":462,"domestic":463,"extension":464,"faq":15,"free":463,"github":441,"languages":1283,"lastVerified":467,"meta":1284,"models":15,"navigation":469,"notSuitable":15,"opensource":469,"path":470,"pillar":471,"platforms":1285,"priceTable":15,"pricing":476,"published":477,"relatedPlaybooks":15,"relatedReviews":15,"score":1286,"self_host":463,"seo":1287,"seoTitle":482,"slug":483,"sources":1288,"stem":487,"suitable":15,"tagline":488,"tags":1291,"updated":467,"verdict":493,"website":433,"__hash__":494},[12,13,14],{"type":17,"value":979,"toc":1270},[980,982,986,988,990,1024,1026,1030,1052,1054,1056,1062,1074,1078,1090,1092,1108,1110,1186,1188,1210,1212,1228,1230,1234,1238,1242,1244,1252,1254,1258],[20,981,23],{"id":22},[25,983,27,984,32],{},[29,985,31],{},[25,987,35],{},[20,989,38],{"id":38},[40,991,992,996,1000,1004,1008,1012,1016,1020],{},[43,993,994,48],{},[29,995,47],{},[43,997,998,54],{},[29,999,53],{},[43,1001,1002,60],{},[29,1003,59],{},[43,1005,1006,66],{},[29,1007,65],{},[43,1009,1010,72],{},[29,1011,71],{},[43,1013,1014,78],{},[29,1015,77],{},[43,1017,1018,84],{},[29,1019,83],{},[43,1021,1022,90],{},[29,1023,89],{},[20,1025,93],{"id":93},[95,1027,1028],{},[25,1029,99],{},[101,1031,1032,1042],{},[104,1033,1034],{},[107,1035,1036,1038,1040],{},[110,1037,112],{},[110,1039,93],{},[110,1041,117],{},[119,1043,1044],{},[107,1045,1046,1048,1050],{},[124,1047,126],{},[124,1049,129],{},[124,1051,132],{},[25,1053,135],{},[20,1055,139],{"id":138},[95,1057,1058],{},[25,1059,144,1060],{},[29,1061,147],{},[40,1063,1064,1066,1068,1070,1072],{},[43,1065,152],{},[43,1067,155],{},[43,1069,158],{},[43,1071,161],{},[43,1073,164],{},[25,1075,1076],{},[29,1077,169],{},[40,1079,1080,1082,1084,1086,1088],{},[43,1081,174],{},[43,1083,177],{},[43,1085,180],{},[43,1087,183],{},[43,1089,186],{},[20,1091,189],{"id":189},[191,1093,1094,1096,1098,1100,1102,1104],{},[43,1095,195],{},[43,1097,198],{},[43,1099,201],{},[43,1101,204],{},[43,1103,207],{},[43,1105,210,1106],{},[212,1107,214],{},[20,1109,217],{"id":217},[101,1111,1112,1124],{},[104,1113,1114],{},[107,1115,1116,1118,1120,1122],{},[110,1117,226],{},[110,1119,10],{},[110,1121,231],{},[110,1123,234],{},[119,1125,1126,1136,1146,1156,1166,1176],{},[107,1127,1128,1130,1132,1134],{},[124,1129,47],{},[124,1131,243],{},[124,1133,246],{},[124,1135,246],{},[107,1137,1138,1140,1142,1144],{},[124,1139,253],{},[124,1141,256],{},[124,1143,259],{},[124,1145,246],{},[107,1147,1148,1150,1152,1154],{},[124,1149,266],{},[124,1151,269],{},[124,1153,272],{},[124,1155,275],{},[107,1157,1158,1160,1162,1164],{},[124,1159,280],{},[124,1161,283],{},[124,1163,286],{},[124,1165,246],{},[107,1167,1168,1170,1172,1174],{},[124,1169,293],{},[124,1171,296],{},[124,1173,299],{},[124,1175,302],{},[107,1177,1178,1180,1182,1184],{},[124,1179,307],{},[124,1181,310],{},[124,1183,310],{},[124,1185,315],{},[20,1187,318],{"id":318},[40,1189,1190,1194,1198,1202,1206],{},[43,1191,1192,326],{},[29,1193,325],{},[43,1195,1196,332],{},[29,1197,331],{},[43,1199,1200,338],{},[29,1201,337],{},[43,1203,1204,344],{},[29,1205,343],{},[43,1207,1208,350],{},[29,1209,349],{},[20,1211,354],{"id":353},[40,1213,1214,1216,1218,1220,1222,1224,1226],{},[43,1215,359],{},[43,1217,362],{},[43,1219,365],{},[43,1221,368],{},[43,1223,371],{},[43,1225,374],{},[43,1227,377],{},[20,1229,381],{"id":380},[25,1231,1232,387],{},[29,1233,386],{},[25,1235,1236,393],{},[29,1237,392],{},[25,1239,1240,399],{},[29,1241,398],{},[20,1243,402],{"id":402},[25,1245,1246,410,1248,410,1250],{},[406,1247,409],{"href":408},[406,1249,414],{"href":413},[406,1251,418],{"href":417},[20,1253,421],{"id":421},[95,1255,1256],{},[25,1257,426],{},[40,1259,1260,1265],{},[43,1261,1262],{},[406,1263,436],{"href":433,"rel":1264},[435],[43,1266,1267],{},[406,1268,443],{"href":441,"rel":1269},[435],{"title":445,"searchDepth":446,"depth":446,"links":1271},[1272,1273,1274,1275,1276,1277,1278,1279,1280,1281,1282],{"id":22,"depth":449,"text":23},{"id":38,"depth":449,"text":38},{"id":93,"depth":449,"text":93},{"id":138,"depth":449,"text":139},{"id":189,"depth":449,"text":189},{"id":217,"depth":449,"text":217},{"id":318,"depth":449,"text":318},{"id":353,"depth":449,"text":354},{"id":380,"depth":449,"text":381},{"id":402,"depth":449,"text":402},{"id":421,"depth":449,"text":421},[466],{},[473,474,475],{"power":446,"ux":479,"price":480,"cn_support":449,"stability":479},{"title":10,"description":462},[1289,1290],{"title":436,"url":433},{"title":443,"url":441},[460,490,491,492],{"id":1293,"title":409,"alternatives":1294,"api_compatible":15,"body":1295,"category":460,"chinese_friendly":446,"cover":1687,"description":1688,"domestic":463,"extension":464,"faq":15,"free":463,"github":1672,"languages":1689,"lastVerified":467,"meta":1690,"models":15,"navigation":469,"notSuitable":15,"opensource":469,"path":1691,"pillar":471,"platforms":1692,"priceTable":15,"pricing":1693,"published":477,"relatedPlaybooks":15,"relatedReviews":15,"score":1694,"self_host":463,"seo":1695,"seoTitle":1696,"slug":14,"sources":1697,"stem":1700,"suitable":15,"tagline":1701,"tags":1702,"updated":467,"verdict":1704,"website":1666,"__hash__":1705},"tools\u002Ftools\u002Fcoding\u002Flocal\u002Fjan.md",[13,955,12],{"type":17,"value":1296,"toc":1674},[1297,1299,1306,1309,1311,1360,1362,1366,1389,1392,1394,1400,1417,1421,1438,1440,1460,1462,1565,1567,1599,1601,1623,1625,1631,1637,1643,1645,1654,1656,1660],[20,1298,23],{"id":22},[25,1300,1301,1302,1305],{},"Jan 是开源的本地 LLM 桌面客户端，定位是",[29,1303,1304],{},"ChatGPT 的离线替代品","。界面设计精美，操作体验接近 ChatGPT，支持 GGUF 模型一键下载、本地推理、多模型切换、插件扩展。AGPL 开源，完全免费。适合想要一个好看好用的本地 AI 聊天工具、隐私优先的用户。",[25,1307,1308],{},"适合：想要 ChatGPT 颜值和体验的本地替代、个人离线聊天、隐私敏感场景、非技术用户（GUI 友好）。不适合：需要 OpenAI 兼容 API 给应用接入（用 Ollama）、需要模型调参 \u002F 量化选择（用 LM Studio）、企业商用（AGPL 限制）。",[20,1310,38],{"id":38},[40,1312,1313,1319,1324,1330,1336,1342,1348,1354],{},[43,1314,1315,1318],{},[29,1316,1317],{},"ChatGPT 式界面","：聊天 UI 设计精美，多会话管理、Markdown 渲染、代码高亮",[43,1320,1321,1323],{},[29,1322,53],{},"：内置模型市场，搜索 GGUF 模型点击下载，自动配置",[43,1325,1326,1329],{},[29,1327,1328],{},"本地推理","：基于 llama.cpp，支持 CPU \u002F GPU 加速，完全离线运行",[43,1331,1332,1335],{},[29,1333,1334],{},"多模型切换","：一个会话可切换不同模型对比输出，方便评估",[43,1337,1338,1341],{},[29,1339,1340],{},"插件系统","：支持扩展功能，如网页搜索、文档分析、API 代理等",[43,1343,1344,1347],{},[29,1345,1346],{},"远程 API 接入","：除了本地模型，也支持接 OpenAI \u002F Anthropic 等云端 API",[43,1349,1350,1353],{},[29,1351,1352],{},"跨平台桌面 App","：Win \u002F Mac \u002F Linux 原生安装包，Electron 构建",[43,1355,1356,1359],{},[29,1357,1358],{},"隐私优先","：所有数据本地存储，无遥测，无云端调用（本地模型模式）",[20,1361,93],{"id":93},[95,1363,1364],{},[25,1365,99],{},[101,1367,1368,1378],{},[104,1369,1370],{},[107,1371,1372,1374,1376],{},[110,1373,112],{},[110,1375,93],{},[110,1377,117],{},[119,1379,1380],{},[107,1381,1382,1384,1386],{},[124,1383,126],{},[124,1385,129],{},[124,1387,1388],{},"完整功能，AGPL 协议",[25,1390,1391],{},"完全免费。注意 AGPL 协议：个人使用无限制，但二次开发 \u002F 商用需遵守开源传染条款。",[20,1393,139],{"id":138},[95,1395,1396],{},[25,1397,144,1398],{},[29,1399,147],{},[40,1401,1402,1405,1408,1411,1414],{},[43,1403,1404],{},"界面设计是同类最佳：比 LM Studio \u002F GPT4All 好看很多，接近 ChatGPT 体验",[43,1406,1407],{},"模型下载体验顺滑：搜索 → 下载 → 使用，全程 GUI，零命令行",[43,1409,1410],{},"多模型对比实用：同一问题切换模型看不同回答，选模型很方便",[43,1412,1413],{},"插件系统有潜力：网页搜索插件让本地模型也能联网",[43,1415,1416],{},"支持云端 API 混用：本地模型 + GPT-4o 切换，一个客户端搞定",[25,1418,1419],{},[29,1420,169],{},[40,1422,1423,1426,1429,1432,1435],{},[43,1424,1425],{},"Electron 应用内存占用偏高，老设备偶有卡顿",[43,1427,1428],{},"模型管理不如 LM Studio：量化版本选择少，调参选项有限",[43,1430,1431],{},"API Server 功能弱：有 OpenAI 兼容端点但不如 Ollama 灵活",[43,1433,1434],{},"插件生态尚不成熟，可用插件不多",[43,1436,1437],{},"AGPL 协议对企业不友好，商用需注意合规",[20,1439,189],{"id":189},[191,1441,1442,1445,1448,1451,1454,1457],{},[43,1443,1444],{},"从 jan.ai 下载对应平台安装包",[43,1446,1447],{},"安装后打开 Jan，界面类似 ChatGPT",[43,1449,1450],{},"点击模型市场（Hub）→ 搜索推荐模型（Qwen2.5-7B \u002F Llama3.1-8B）",[43,1452,1453],{},"下载模型后，新建会话 → 选择模型 → 开始聊天",[43,1455,1456],{},"多模型对比：同一会话切换模型或开多个会话",[43,1458,1459],{},"接云端 API：Settings → API Keys → 填入 OpenAI Key 即可混用",[20,1461,217],{"id":217},[101,1463,1464,1478],{},[104,1465,1466],{},[107,1467,1468,1470,1472,1474,1476],{},[110,1469,226],{},[110,1471,409],{},[110,1473,234],{},[110,1475,231],{},[110,1477,498],{},[119,1479,1480,1496,1510,1523,1537,1551],{},[107,1481,1482,1485,1488,1491,1494],{},[124,1483,1484],{},"界面颜值",[124,1486,1487],{},"高",[124,1489,1490],{},"中",[124,1492,1493],{},"无 GUI",[124,1495,1487],{},[107,1497,1498,1501,1503,1506,1508],{},[124,1499,1500],{},"模型管理",[124,1502,1490],{},[124,1504,1505],{},"强",[124,1507,1505],{},[124,1509,1490],{},[107,1511,1512,1515,1517,1519,1521],{},[124,1513,1514],{},"API 接入",[124,1516,283],{},[124,1518,246],{},[124,1520,1505],{},[124,1522,246],{},[107,1524,1525,1528,1530,1533,1535],{},[124,1526,1527],{},"插件扩展",[124,1529,246],{},[124,1531,1532],{},"❌",[124,1534,1532],{},[124,1536,246],{},[107,1538,1539,1542,1544,1546,1549],{},[124,1540,1541],{},"云端 API 混用",[124,1543,246],{},[124,1545,246],{},[124,1547,1548],{},"需配",[124,1550,246],{},[107,1552,1553,1555,1558,1560,1562],{},[124,1554,765],{},[124,1556,1557],{},"AGPL",[124,1559,315],{},[124,1561,310],{},[124,1563,1564],{},"Apache",[20,1566,318],{"id":318},[40,1568,1569,1575,1581,1587,1593],{},[43,1570,1571,1574],{},[29,1572,1573],{},"别指望它做 API 服务器","：Jan 的 API Server 功能基础，给应用接入用 Ollama",[43,1576,1577,1580],{},[29,1578,1579],{},"模型选对量化","：默认下载的可能不是最优量化，手动选 Q4_K_M 平衡速度质量",[43,1582,1583,1586],{},[29,1584,1585],{},"Electron 吃内存","：8GB RAM 设备跑大模型 + Jan 本身会卡，关其他应用",[43,1588,1589,1592],{},[29,1590,1591],{},"AGPL 商用注意","：企业内部署需法务确认 AGPL 合规",[43,1594,1595,1598],{},[29,1596,1597],{},"插件别装太多","：部分插件质量参差，可能影响稳定性",[20,1600,354],{"id":353},[40,1602,1603,1606,1609,1612,1615,1617,1620],{},[43,1604,1605],{},"✅ 想要 ChatGPT 颜值和体验的本地替代",[43,1607,1608],{},"✅ 个人离线聊天 \u002F 隐私优先场景",[43,1610,1611],{},"✅ 非技术用户（GUI 友好，零命令行）",[43,1613,1614],{},"✅ 本地 + 云端 API 混用需求",[43,1616,374],{},[43,1618,1619],{},"❌ 需要精细模型调参 \u002F 量化管理（用 LM Studio）",[43,1621,1622],{},"❌ 企业商用（AGPL 限制）",[20,1624,381],{"id":380},[25,1626,1627,1630],{},[29,1628,1629],{},"Q: Jan 和 LM Studio 怎么选？","\nA: 颜值和聊天体验选 Jan，模型管理和调参选 LM Studio。Jan 更像 ChatGPT，LM Studio 更像模型工具箱。两者都免费，可以都装。",[25,1632,1633,1636],{},[29,1634,1635],{},"Q: 能给 Cursor \u002F Cline 接入吗？","\nA: Jan 有 OpenAI 兼容 API Server（默认端口 1337），理论上可以。但不如 Ollama 稳定灵活，推荐用 Ollama 做 API 服务器。",[25,1638,1639,1642],{},[29,1640,1641],{},"Q: AGPL 协议影响个人使用吗？","\nA: 不影响。AGPL 主要约束网络服务分发场景。个人本地使用完全无限制。只有你把 Jan 改造后对外提供 SaaS 服务才需开源你的修改。",[20,1644,402],{"id":402},[25,1646,1647,410,1650,410,1652],{},[406,1648,10],{"href":1649},"\u002Fcoding\u002Flocal\u002Fgpt4all.html",[406,1651,414],{"href":413},[406,1653,418],{"href":417},[20,1655,421],{"id":421},[95,1657,1658],{},[25,1659,426],{},[40,1661,1662,1668],{},[43,1663,1664],{},[406,1665,436],{"href":1666,"rel":1667},"https:\u002F\u002Fjan.ai",[435],[43,1669,1670],{},[406,1671,443],{"href":1672,"rel":1673},"https:\u002F\u002Fgithub.com\u002Fjanhq\u002Fjan",[435],{"title":445,"searchDepth":446,"depth":446,"links":1675},[1676,1677,1678,1679,1680,1681,1682,1683,1684,1685,1686],{"id":22,"depth":449,"text":23},{"id":38,"depth":449,"text":38},{"id":93,"depth":449,"text":93},{"id":138,"depth":449,"text":139},{"id":189,"depth":449,"text":189},{"id":217,"depth":449,"text":217},{"id":318,"depth":449,"text":318},{"id":353,"depth":449,"text":354},{"id":380,"depth":449,"text":381},{"id":402,"depth":449,"text":402},{"id":421,"depth":449,"text":421},"\u002Fimg\u002Ftools\u002Fjan.webp","Jan 真实评测：开源本地 LLM 桌面客户端（AGPL 协议），定位 ChatGPT 的离线替代，支持 GGUF 模型一键下载 + 本地推理 + 插件扩展。跨平台桌面 app，适合需要完全离线、隐私优先的本地 AI 聊天场景。",[466],{},"\u002Ftools\u002Fcoding\u002Flocal\u002Fjan",[473,474,475],"Free \u002F 开源（AGPL）",{"power":446,"ux":479,"price":480,"cn_support":446,"stability":446},{"title":409,"description":1688},"Jan - 开源本地 LLM 桌面客户端评测 | AIHO",[1698,1699],{"title":436,"url":1666},{"title":443,"url":1672},"tools\u002Fcoding\u002Flocal\u002Fjan","开源本地 LLM 桌面客户端，定位 ChatGPT 的离线替代",[460,967,491,492,1703],"offline","颜值最高、最像 ChatGPT 的开源本地 LLM 客户端，离线聊天体验好；但 API 能力和模型管理不如 Ollama\u002FLM Studio，定位偏轻量个人使用。","SYftPw58WAWIAEA4qHpFo2Ki3Z1H9q97Rp4jQR560KE",{"id":1707,"title":234,"alternatives":1708,"api_compatible":15,"body":1709,"category":460,"chinese_friendly":446,"cover":2133,"description":2134,"domestic":463,"extension":464,"faq":2135,"free":463,"github":15,"languages":2148,"lastVerified":15,"meta":2149,"models":15,"navigation":469,"notSuitable":15,"opensource":463,"path":866,"pillar":471,"platforms":2150,"priceTable":2151,"pricing":2159,"published":948,"relatedPlaybooks":2160,"relatedReviews":15,"score":2162,"self_host":469,"seo":2163,"seoTitle":2164,"slug":13,"sources":2165,"stem":2172,"suitable":15,"tagline":2173,"tags":2174,"updated":959,"verdict":2180,"website":2181,"__hash__":2182},"tools\u002Ftools\u002Fcoding\u002Flocal\u002Flm-studio.md",[12,501,955,500],{"type":17,"value":1710,"toc":2121},[1711,1713,1720,1723,1725,1782,1784,1798,1803,1807,1811,1828,1832,1849,1851,1877,1879,2012,2014,2046,2048,2071,2073,2096,2098],[20,1712,23],{"id":22},[25,1714,1715,1716,1719],{},"LM Studio 是 Windows \u002F macOS \u002F Linux 桌面应用，让你像浏览 App Store 一样发现、下载、运行本地大模型（GGUF \u002F MLX 格式）。底层基于 llama.cpp + MLX，Mac M 系列原生优化。0.3+ 起新增 Headless 模式 + ",[212,1717,1718],{},"lms"," CLI，可在服务器跑 OpenAI 兼容 API（默认 :1234）。个人 \u002F 评估完全免费，商用咨询。",[25,1721,1722],{},"适合：本地 LLM 入门 \u002F 评估、Mac 用户、需要 GUI 调参 \u002F 模型比较、想给 IDE \u002F 应用接本地 OpenAI 兼容 endpoint 的开发者。不适合：多用户并发生产服务（用 vLLM）、嵌入式 \u002F 边缘部署（用 llama.cpp）、纯 CLI 工作流（用 Ollama）。",[20,1724,38],{"id":38},[40,1726,1727,1733,1739,1745,1755,1764,1770,1776],{},[43,1728,1729,1732],{},[29,1730,1731],{},"模型浏览器","：内置 Hugging Face 检索，按 GGUF \u002F MLX \u002F 大小筛选、一键下载",[43,1734,1735,1738],{},[29,1736,1737],{},"聊天界面","：System Prompt \u002F temperature \u002F top-p \u002F context size 可视化调参",[43,1740,1741,1744],{},[29,1742,1743],{},"多模型并存 \u002F 切换","：同时加载多模型在不同会话中比较",[43,1746,1747,1750,1751,1754],{},[29,1748,1749],{},"OpenAI 兼容 Local Server","：",[212,1752,1753],{},"http:\u002F\u002Flocalhost:1234\u002Fv1","，任何 SDK 即接即用",[43,1756,1757,1750,1760,1763],{},[29,1758,1759],{},"Headless \u002F CLI",[212,1761,1762],{},"lms server start --port 1234","，无 GUI 可跑",[43,1765,1766,1769],{},[29,1767,1768],{},"PDF \u002F 文档对话","：内置基础 RAG，丢文件就能聊",[43,1771,1772,1775],{},[29,1773,1774],{},"MLX 原生支持（Mac）","：M1+ 上比 GGUF + Metal 快 30–50%",[43,1777,1778,1781],{},[29,1779,1780],{},"持续批处理","：Codersera 2026 测得 50–90 tok\u002Fs（消费级 GPU + 中等模型）",[20,1783,93],{"id":93},[40,1785,1786,1792],{},[43,1787,1788,1791],{},[29,1789,1790],{},"个人 \u002F 评估","：免费，全功能可用",[43,1793,1794,1797],{},[29,1795,1796],{},"商用","：邮件 \u002F 官网联系 LM Studio 团队",[95,1799,1800],{},[25,1801,1802],{},"模型本身免费（开源权重），LM Studio 不抽水任何 token 费用。",[20,1804,1806],{"id":1805},"实测mac-m2-pro-qwen3-coder-7b-gguf-q4_k_m","实测（Mac M2 Pro + Qwen3-Coder-7B GGUF Q4_K_M）",[25,1808,1809],{},[29,1810,147],{},[40,1812,1813,1816,1819,1822,1825],{},[43,1814,1815],{},"模型浏览器极舒服：搜「qwen3-coder」直接列出 GGUF + MLX 各 quant，标硬件兼容度",[43,1817,1818],{},"加载 7B Q4 模型 \u003C 3 秒，生成 ~75 tok\u002Fs",[43,1820,1821],{},"Local Server 开了 Cursor 直接接 baseURL → 本地代码补全零成本",[43,1823,1824],{},"MLX 版同模型 ~110 tok\u002Fs，差距显著",[43,1826,1827],{},"多窗口加载 2 个模型并排测，调 prompt 直观",[25,1829,1830],{},[29,1831,169],{},[40,1833,1834,1837,1840,1843,1846],{},[43,1835,1836],{},"模型库依赖 Hugging Face，国内访问要镜像 \u002F 代理",[43,1838,1839],{},"GPU 显存吃满后会自动 offload 到 CPU，无提示就慢下来",[43,1841,1842],{},"Headless 模式相对 Ollama 偏新，文档稍少",[43,1844,1845],{},"闭源应用（虽免费），不适合企业合规挂钩",[43,1847,1848],{},"中文 UI 可用但部分菜单仍英文",[20,1850,189],{"id":189},[191,1852,1853,1856,1859,1862,1865,1872],{},[43,1854,1855],{},"lmstudio.ai 下载（Mac \u002F Windows \u002F Linux）",[43,1857,1858],{},"打开 → Discover 标签 → 搜模型（如 qwen3-coder、deepseek-v3 GGUF\u002FMLX）→ Download",[43,1860,1861],{},"Chat 标签 → 选模型 → 调参聊天",[43,1863,1864],{},"Local Server 标签 → Start Server → 默认端口 1234",[43,1866,1867,1868,1871],{},"在你的应用里：",[212,1869,1870],{},"baseURL = \"http:\u002F\u002Flocalhost:1234\u002Fv1\"","，API Key 任意",[43,1873,1874,1875],{},"Headless：",[212,1876,1762],{},[20,1878,217],{"id":217},[101,1880,1881,1896],{},[104,1882,1883],{},[107,1884,1885,1887,1889,1891,1893],{},[110,1886,226],{},[110,1888,234],{},[110,1890,231],{},[110,1892,669],{},[110,1894,1895],{},"llama.cpp",[119,1897,1898,1914,1930,1944,1958,1972,1985,1997],{},[107,1899,1900,1902,1905,1908,1911],{},[124,1901,676],{},[124,1903,1904],{},"GUI + CLI",[124,1906,1907],{},"CLI Daemon",[124,1909,1910],{},"Docker UI",[124,1912,1913],{},"二进制",[107,1915,1916,1919,1922,1925,1927],{},[124,1917,1918],{},"模型浏览",[124,1920,1921],{},"✅ 内置",[124,1923,1924],{},"CLI pull",[124,1926,743],{},[124,1928,1929],{},"手动",[107,1931,1932,1935,1937,1939,1942],{},[124,1933,1934],{},"参数调优 GUI",[124,1936,246],{},[124,1938,1532],{},[124,1940,1941],{},"部分",[124,1943,1532],{},[107,1945,1946,1948,1951,1954,1956],{},[124,1947,71],{},[124,1949,1950],{},"✅ :1234",[124,1952,1953],{},"✅ :11434",[124,1955,246],{},[124,1957,246],{},[107,1959,1960,1963,1965,1968,1970],{},[124,1961,1962],{},"MLX (Mac)",[124,1964,246],{},[124,1966,1967],{},"✅ 0.19+",[124,1969,789],{},[124,1971,789],{},[107,1973,1974,1977,1979,1981,1983],{},[124,1975,1976],{},"多用户并发",[124,1978,714],{},[124,1980,714],{},[124,1982,246],{},[124,1984,1490],{},[107,1986,1987,1989,1991,1993,1995],{},[124,1988,307],{},[124,1990,773],{},[124,1992,310],{},[124,1994,310],{},[124,1996,310],{},[107,1998,1999,2002,2005,2008,2010],{},[124,2000,2001],{},"上手难度",[124,2003,2004],{},"极低",[124,2006,2007],{},"低",[124,2009,1490],{},[124,2011,1487],{},[20,2013,318],{"id":318},[40,2015,2016,2022,2028,2034,2040],{},[43,2017,2018,2021],{},[29,2019,2020],{},"国内下模型走镜像","：HF 直连慢 \u002F 卡，配 HF_ENDPOINT=hf-mirror.com",[43,2023,2024,2027],{},[29,2025,2026],{},"显存爆 ≠ 报错","：GPU 装不下会无声 offload 到 CPU，关注生成速度，必要时降 quant 或换小模型",[43,2029,2030,2033],{},[29,2031,2032],{},"MLX 优先（Mac M 系列）","：能下 MLX 版就别下 GGUF，速度差距明显",[43,2035,2036,2039],{},[29,2037,2038],{},"Local Server 暴露要谨慎","：默认 0.0.0.0 + 无鉴权，对外开放前加反代 + Bearer",[43,2041,2042,2045],{},[29,2043,2044],{},"闭源合规要核","：企业内部使用前查 license；商用必须联系官方",[20,2047,354],{"id":353},[40,2049,2050,2053,2056,2059,2062,2065,2068],{},[43,2051,2052],{},"✅ 本地 LLM 入门 \u002F 评估",[43,2054,2055],{},"✅ Mac M 系列用户",[43,2057,2058],{},"✅ 想给 Cursor \u002F Cline 接本地 OpenAI 兼容 endpoint",[43,2060,2061],{},"✅ 需要 GUI 调参 \u002F 模型比较",[43,2063,2064],{},"❌ 多用户并发生产服务",[43,2066,2067],{},"❌ 嵌入式 \u002F 边缘设备",[43,2069,2070],{},"❌ 强合规 \u002F 必须开源审计",[20,2072,402],{"id":402},[40,2074,2075,2079,2085,2090],{},[43,2076,2077],{},[406,2078,873],{"href":872},[43,2080,2081],{},[406,2082,2084],{"href":2083},"\u002Ftools\u002Fcoding\u002Flocal\u002Fopen-webui","Open WebUI 评测",[43,2086,2087],{},[406,2088,2089],{"href":935},"Cherry Studio 评测",[43,2091,2092],{},[406,2093,2095],{"href":2094},"\u002Fplaybook\u002Fonboarding\u002Fclaude-code-getting-started","Claude Code 上手 Playbook",[20,2097,421],{"id":421},[191,2099,2100,2107,2114],{},[43,2101,2102,2103],{},"LM Studio 官网 ",[406,2104,2105],{"href":2105,"rel":2106},"https:\u002F\u002Flmstudio.ai\u002F",[435],[43,2108,2109,2110],{},"Codersera — LM Studio Complete Guide 2026 ",[406,2111,2112],{"href":2112,"rel":2113},"https:\u002F\u002Fcodersera.com\u002Fblog\u002Flm-studio-complete-guide-2026\u002F",[435],[43,2115,2116,2117],{},"Codersera — Ollama vs LM Studio vs vLLM vs llama.cpp vs MLX 2026 ",[406,2118,2119],{"href":2119,"rel":2120},"https:\u002F\u002Fcodersera.com\u002Fblog\u002Follama-vs-lm-studio-vs-vllm-vs-llama-cpp-vs-mlx-2026\u002F",[435],{"title":445,"searchDepth":446,"depth":446,"links":2122},[2123,2124,2125,2126,2127,2128,2129,2130,2131,2132],{"id":22,"depth":449,"text":23},{"id":38,"depth":449,"text":38},{"id":93,"depth":449,"text":93},{"id":1805,"depth":449,"text":1806},{"id":189,"depth":449,"text":189},{"id":217,"depth":449,"text":217},{"id":318,"depth":449,"text":318},{"id":353,"depth":449,"text":354},{"id":402,"depth":449,"text":402},{"id":421,"depth":449,"text":421},"\u002Fimg\u002Ftools\u002Flm-studio.webp","LM Studio 真实评测：跨平台桌面应用，运行本地 GGUF \u002F MLX 大模型。50–90 tok\u002Fs 持续批处理、OpenAI 兼容本地 API（默认端口 1234）、Headless 模式、Mac \u002F Win 双端。对个人开发者免费，企业咨询。",[2136,2139,2142,2145],{"q":2137,"a":2138},"和 Ollama 怎么选？","LM Studio 是 GUI 优先（模型浏览器 + 参数面板 + 聊天界面），适合个人 \u002F 评估 \u002F 上手。Ollama 是 CLI \u002F Daemon 优先（后台跑 + REST API），适合应用嵌入 \u002F 脚本调用。两者都基于 llama.cpp，在 Mac M 系列上都已用 MLX。",{"q":2140,"a":2141},"支持 MLX 吗？","支持。Mac M1+ 上可加载 MLX 格式模型，速度比 GGUF + Metal 快 30–50%。模型搜索时筛选 MLX 即可。",{"q":2143,"a":2144},"OpenAI 兼容 API 怎么用？","开 Local Server → 默认端口 1234 → `http:\u002F\u002Flocalhost:1234\u002Fv1`。任何 OpenAI SDK 把 baseURL 改这个就能跑本地模型，零代码改动。",{"q":2146,"a":2147},"Headless 模式？","0.3+ 起支持 `lms server start` CLI 启动后台服务，无 GUI 即可跑 OpenAI 兼容 API，适合服务器 \u002F SSH 场景。",[466,933],{},[473,474,475],[2152,2155],{"plan":1790,"price":940,"features":2153,"notes":2154},"全功能 GUI + Headless API + GGUF\u002FMLX","供个人 \u002F 评估使用",{"plan":1796,"price":2156,"features":2157,"notes":2158},"联系咨询","团队部署 \u002F 商用 license","邮件 \u002F 官网联系","免费（个人 \u002F 评估） \u002F 企业 \u002F 商用咨询",[950,2161],"onboarding\u002Fclaude-code-getting-started",{"power":479,"ux":480,"price":480,"cn_support":446,"stability":479},{"title":234,"description":2134},"LM Studio 评测 2026：本地运行开源大模型，图形化界面，AI 模型管理",[2166,2168,2170],{"name":2167,"url":2105,"accessed":959},"LM Studio 官网",{"name":2169,"url":2112,"accessed":959},"Codersera — LM Studio Complete Guide 2026",{"name":2171,"url":2119,"accessed":959},"Codersera — Ollama vs LM Studio vs vLLM 2026","tools\u002Fcoding\u002Flocal\u002Flm-studio","本地 LLM 的 GUI 首选——模型浏览器 + GGUF\u002FMLX 推理 + OpenAI 兼容 API + Mac 原生优化",[460,2175,492,2176,2177,2178,2179],"gui","mlx","llama-cpp","mac","openai-compatible","Mac \u002F Windows 桌面本地 LLM 的 GUI 首选——上手最快、模型浏览最舒服、自带 OpenAI 兼容 API。批量服务 \u002F 多用户场景用 vLLM；纯 CLI \u002F 嵌入应用走 Ollama。","https:\u002F\u002Flmstudio.ai","LobnLABcHoL2A6Bu-tfBMAlwTLq_jKyEJkNmD662RWU",{"id":2184,"title":664,"alternatives":2185,"api_compatible":15,"body":2186,"category":460,"chinese_friendly":480,"cover":2622,"description":2623,"domestic":463,"extension":464,"faq":2624,"free":463,"github":15,"languages":2637,"lastVerified":15,"meta":2638,"models":15,"navigation":469,"notSuitable":15,"opensource":469,"path":860,"pillar":471,"platforms":2639,"priceTable":2642,"pricing":2650,"published":948,"relatedPlaybooks":2651,"relatedReviews":15,"score":2652,"self_host":469,"seo":2653,"seoTitle":2654,"slug":500,"sources":2655,"stem":2662,"suitable":15,"tagline":2663,"tags":2664,"updated":959,"verdict":2667,"website":2668,"__hash__":2669},"tools\u002Ftools\u002Fcoding\u002Flocal\u002Flobe-chat.md",[955,501,12,13],{"type":17,"value":2187,"toc":2610},[2188,2190,2197,2200,2202,2264,2266,2279,2282,2286,2290,2310,2314,2331,2333,2359,2361,2497,2499,2537,2539,2565,2567,2585,2587],[20,2189,23],{"id":22},[25,2191,2192,2193,2196],{},"LobeChat 是 LobeHub 团队的开源 AI 聊天框架，2023 年发布、GitHub 72k+ stars、MIT 协议。",[29,2194,2195],{},"Web + 桌面 + Docker 自托管三形态","，把 OpenAI \u002F Claude \u002F Gemini \u002F DeepSeek \u002F Qwen \u002F Kimi \u002F Ollama \u002F LM Studio 等 80+ 模型聚合到一个现代设计的客户端里。内置 RAG 知识库 + 插件市场 + 助手市场 + 多模型对比 + MCP，是当下综合最强的多模型 AI 客户端之一。",[25,2198,2199],{},"适合：需要 Web 端访问、Docker 自托管、多模型对比、丰富助手市场的用户；中文重度用户；想给团队 \u002F 家庭部署一个共享 AI 工作台。不适合：只用桌面 + 不需要 Web（Cherry Studio 同样优秀且更精细）、强企业 RBAC + 多租户（Open WebUI 多用户更完善）。",[20,2201,38],{"id":38},[40,2203,2204,2209,2215,2220,2226,2232,2237,2242,2248,2254],{},[43,2205,2206,2208],{},[29,2207,520],{},"：OpenAI \u002F Claude \u002F Gemini \u002F DeepSeek \u002F Qwen \u002F Kimi \u002F 豆包 \u002F Groq \u002F Together \u002F OpenRouter \u002F Ollama \u002F LM Studio",[43,2210,2211,2214],{},[29,2212,2213],{},"多模型对比","：同 prompt 给多模型并排回答",[43,2216,2217,2219],{},[29,2218,526],{},"：上传 PDF \u002F Word \u002F 网页 → 向量化 → 检索引用",[43,2221,2222,2225],{},[29,2223,2224],{},"插件市场","：联网搜索 \u002F 代码执行 \u002F 图像生成 \u002F 翻译等几十款官方插件",[43,2227,2228,2231],{},[29,2229,2230],{},"助手市场","：几百个预设 AI 角色，一键导入",[43,2233,2234,2236],{},[29,2235,538],{},"：扩展任意工具能力",[43,2238,2239],{},[29,2240,2241],{},"代码解释器 \u002F 文件上传 \u002F TTS \u002F 多模态",[43,2243,2244,2247],{},[29,2245,2246],{},"Web + 桌面 + Docker","：三形态，数据可完全本地",[43,2249,2250,2253],{},[29,2251,2252],{},"LobeHub Cloud","：官方云托管，免部署",[43,2255,2256,2259,2260,2263],{},[29,2257,2258],{},"快捷指令 \u002F 工作流","：自定义 prompt 模板，",[212,2261,2262],{},"\u002Fpodcast-summary"," 类用法",[20,2265,93],{"id":93},[40,2267,2268,2274],{},[43,2269,2270,2273],{},[29,2271,2272],{},"自托管 \u002F 桌面","：完全免费、MIT 开源",[43,2275,2276,2278],{},[29,2277,2252],{},"：订阅制，云端托管 + 团队协作 + 同步",[25,2280,2281],{},"模型 API 费用按你自己的供应商付费；本地 Ollama \u002F LM Studio 零成本。",[20,2283,2285],{"id":2284},"实测m2-自托管-docker连-openai-deepseek-本地-ollama","实测（M2 + 自托管 Docker，连 OpenAI + DeepSeek + 本地 Ollama）",[25,2287,2288],{},[29,2289,147],{},[40,2291,2292,2295,2298,2301,2304,2307],{},[43,2293,2294],{},"界面颜值是这一类工具里第一档（深色 \u002F 透明 \u002F 现代感）",[43,2296,2297],{},"多模型并排对比对选型极其有用：写一道复杂题，Claude \u002F GPT \u002F DeepSeek 直接对比答案",[43,2299,2300],{},"知识库 RAG 上传 50+ PDF 后检索准确，引用片段可视化",[43,2302,2303],{},"助手市场拿来即用——「Code Reviewer」「Translation Polish」节省 prompt 编写",[43,2305,2306],{},"Docker 一键部署，团队 5 人共享流畅",[43,2308,2309],{},"多平台数据同步（Cloud \u002F WebDAV）",[25,2311,2312],{},[29,2313,169],{},[40,2315,2316,2319,2322,2325,2328],{},[43,2317,2318],{},"自托管要熟悉 Docker + 反代 + HTTPS",[43,2320,2321],{},"国内连 OpenAI \u002F Claude 需自带网络方案",[43,2323,2324],{},"Web 版数据存 LobeHub，隐私敏感场景走桌面 \u002F Docker",[43,2326,2327],{},"插件市场质量参差，要自筛",[43,2329,2330],{},"团队多人共享需配 LobeHub Cloud 或自建数据库（Postgres + S3）",[20,2332,189],{"id":189},[191,2334,2335,2338,2344,2347,2350,2353,2356],{},[43,2336,2337],{},"选形态：Web（chat.lobehub.com 注册即用） \u002F 桌面（GitHub Releases 下载） \u002F Docker",[43,2339,2340,2341],{},"Docker：",[212,2342,2343],{},"docker run -d -p 3210:3210 -e OPENAI_API_KEY=sk-xxx --name lobe-chat lobehub\u002Flobe-chat",[43,2345,2346],{},"设置 → AI 服务商 → 添加 OpenAI \u002F Claude \u002F DeepSeek \u002F Ollama",[43,2348,2349],{},"模型选择器测试对话",[43,2351,2352],{},"知识库：拖文件 → 等向量化 → 对话引用",[43,2354,2355],{},"助手市场拉「Code Reviewer」「论文翻译润色」试用",[43,2357,2358],{},"进阶：插件市场启用联网搜索 \u002F 代码执行；MCP 自定义工具",[20,2360,217],{"id":217},[101,2362,2363,2377],{},[104,2364,2365],{},[107,2366,2367,2369,2371,2373,2375],{},[110,2368,226],{},[110,2370,664],{},[110,2372,498],{},[110,2374,669],{},[110,2376,234],{},[119,2378,2379,2391,2404,2417,2430,2446,2459,2473,2485],{},[107,2380,2381,2383,2385,2387,2389],{},[124,2382,676],{},[124,2384,2246],{},[124,2386,679],{},[124,2388,687],{},[124,2390,679],{},[107,2392,2393,2395,2398,2400,2402],{},[124,2394,520],{},[124,2396,2397],{},"✅ 80+",[124,2399,246],{},[124,2401,246],{},[124,2403,699],{},[107,2405,2406,2408,2411,2413,2415],{},[124,2407,2213],{},[124,2409,2410],{},"✅ 一等",[124,2412,246],{},[124,2414,714],{},[124,2416,714],{},[107,2418,2419,2421,2423,2425,2428],{},[124,2420,706],{},[124,2422,246],{},[124,2424,246],{},[124,2426,2427],{},"✅ + oikb",[124,2429,714],{},[107,2431,2432,2435,2438,2441,2444],{},[124,2433,2434],{},"插件 \u002F 助手市场",[124,2436,2437],{},"✅ 丰富",[124,2439,2440],{},"300+ 助手",[124,2442,2443],{},"Tools",[124,2445,714],{},[107,2447,2448,2450,2452,2454,2457],{},[124,2449,721],{},[124,2451,246],{},[124,2453,246],{},[124,2455,2456],{},"✅ mcpo",[124,2458,714],{},[107,2460,2461,2464,2467,2469,2471],{},[124,2462,2463],{},"多用户",[124,2465,2466],{},"配 Cloud \u002F 自建",[124,2468,743],{},[124,2470,2410],{},[124,2472,743],{},[107,2474,2475,2477,2479,2481,2483],{},[124,2476,780],{},[124,2478,786],{},[124,2480,783],{},[124,2482,792],{},[124,2484,789],{},[107,2486,2487,2489,2491,2493,2495],{},[124,2488,765],{},[124,2490,310],{},[124,2492,768],{},[124,2494,310],{},[124,2496,773],{},[20,2498,318],{"id":318},[40,2500,2501,2507,2513,2519,2525,2531],{},[43,2502,2503,2506],{},[29,2504,2505],{},"Web 版数据不本地","：隐私敏感选桌面或 Docker 自托管",[43,2508,2509,2512],{},[29,2510,2511],{},"国内连海外模型走中转","：直连 OpenAI \u002F Claude 不稳，配 OpenRouter \u002F Ofox \u002F 国内中转",[43,2514,2515,2518],{},[29,2516,2517],{},"Docker 自托管暴露公网","：上反代 + HTTPS + Auth + 备份数据库",[43,2520,2521,2524],{},[29,2522,2523],{},"嵌入模型中文优化","：默认嵌入对中文一般，配 bge-m3 \u002F 硅基流动 Pro 版",[43,2526,2527,2530],{},[29,2528,2529],{},"插件市场审一遍","：第三方插件可执行代码，团队部署谨慎启用",[43,2532,2533,2536],{},[29,2534,2535],{},"同步选 Cloud vs WebDAV","：团队多端走 LobeHub Cloud；个人多设备 WebDAV 即可",[20,2538,354],{"id":353},[40,2540,2541,2544,2547,2550,2553,2556,2559,2562],{},[43,2542,2543],{},"✅ Web + 桌面双形态需求",[43,2545,2546],{},"✅ Docker 自托管 \u002F 团队共享",[43,2548,2549],{},"✅ 多模型对比 \u002F 选型",[43,2551,2552],{},"✅ 中文重度用户",[43,2554,2555],{},"✅ 助手市场 \u002F 插件生态用户",[43,2557,2558],{},"❌ 强企业 RBAC + 多租户（Open WebUI 更完善）",[43,2560,2561],{},"❌ 只要桌面 + 数据完全本地（Cherry Studio 同样优秀）",[43,2563,2564],{},"❌ 完全不会碰 Docker",[20,2566,402],{"id":402},[40,2568,2569,2573,2577,2581],{},[43,2570,2571],{},[406,2572,2089],{"href":935},[43,2574,2575],{},[406,2576,2084],{"href":2083},[43,2578,2579],{},[406,2580,873],{"href":872},[43,2582,2583],{},[406,2584,879],{"href":878},[20,2586,421],{"id":421},[191,2588,2589,2596,2603],{},[43,2590,2591,2592],{},"LobeChat GitHub 仓库（72k+ stars，MIT）",[406,2593,2594],{"href":2594,"rel":2595},"https:\u002F\u002Fgithub.com\u002Flobehub\u002Flobe-chat",[435],[43,2597,2598,2599],{},"腾讯云开发者社区 — Lobe Chat 本地化 AI 聊天终极桌面客户端（2026-01）",[406,2600,2601],{"href":2601,"rel":2602},"https:\u002F\u002Fcloud.tencent.com\u002Fdeveloper\u002Farticle\u002F2622150",[435],[43,2604,2605,2606],{},"Ofox.ai — LobeChat 完全配置指南 2026（2026-04-17）",[406,2607,2608],{"href":2608,"rel":2609},"https:\u002F\u002Fofox.ai\u002Fzh\u002Fblog\u002Flobechat-api-configuration-guide-2026",[435],{"title":445,"searchDepth":446,"depth":446,"links":2611},[2612,2613,2614,2615,2616,2617,2618,2619,2620,2621],{"id":22,"depth":449,"text":23},{"id":38,"depth":449,"text":38},{"id":93,"depth":449,"text":93},{"id":2284,"depth":449,"text":2285},{"id":189,"depth":449,"text":189},{"id":217,"depth":449,"text":217},{"id":318,"depth":449,"text":318},{"id":353,"depth":449,"text":354},{"id":402,"depth":449,"text":402},{"id":421,"depth":449,"text":421},"\u002Fimg\u002Ftools\u002Flobe-chat.webp","LobeChat 真实评测：LobeHub 团队开源 AI 聊天框架，GitHub 72k+ stars、MIT 协议。Web + 桌面（Win\u002FMac\u002FLinux\u002FDocker）双形态，支持 OpenAI \u002F Claude \u002F Gemini \u002F DeepSeek \u002F Qwen \u002F Ollama 等 80+ 模型，内置 RAG 知识库 + 插件市场 + 助手市场 + 多模型对比。",[2625,2628,2631,2634],{"q":2626,"a":2627},"Web 版 vs 桌面版 vs Docker 自托管，怎么选？","Web 版（chat.lobehub.com）最快上手但数据存 LobeHub 服务器；桌面版数据本地存、隐私好；Docker 自托管对团队 \u002F 公司部署最优，完全掌控数据。",{"q":2629,"a":2630},"支持哪些模型？","80+ 模型：OpenAI 全系列、Anthropic Claude、Google Gemini、DeepSeek、Qwen、Kimi、Moonshot、字节豆包、Groq、Together、OpenRouter、Ollama \u002F LM Studio 本地模型，以及任何 OpenAI 兼容 API。",{"q":2632,"a":2633},"多模型对比怎么用？","同一对话窗口里把消息广播给多个模型并排回答，选型 \u002F 评估特别有用——直接看 Claude 和 GPT 在同一 prompt 下的回答差异。",{"q":2635,"a":2636},"助手市场是什么？","LobeHub 维护的预设 AI 角色市场（代码审查 \u002F 翻译 \u002F 写作 \u002F 角色扮演等几百个），一键拉到本地用，省去自己写 System Prompt。",[933,466],{},[2640,473,474,475,2641],"web","docker",[2643,2646],{"plan":2272,"price":940,"features":2644,"notes":2645},"全功能 \u002F 80+ 模型 \u002F 知识库 \u002F 插件 \u002F 助手市场","MIT 协议",{"plan":2252,"price":2647,"features":2648,"notes":2649},"订阅制","云端托管 \u002F 免部署 \u002F 团队协作 \u002F 同步","chat.lobehub.com 注册即用","完全免费（MIT 开源） \u002F LobeHub Cloud 订阅",[950,2161],{"power":480,"ux":480,"price":480,"cn_support":480,"stability":479},{"title":664,"description":2623},"LobeChat - 开源 AI 聊天框架评测与部署 | AIHO",[2656,2658,2660],{"name":2657,"url":2594,"accessed":959},"LobeChat GitHub",{"name":2659,"url":2601,"accessed":959},"腾讯云开发者社区 — Lobe Chat 终极桌面客户端",{"name":2661,"url":2608,"accessed":959},"Ofox.ai — LobeChat 完全配置指南 2026","tools\u002Fcoding\u002Flocal\u002Flobe-chat","现代设计的开源 AI 聊天框架——Web + 桌面双形态、72k+ stars、多模型 + 知识库 + 插件市场",[460,2640,967,968,970,2665,2666,971],"plugin","mcp","颜值与功能双优的多模型 AI 聊天客户端。要 Web + 桌面双形态、自托管 Docker、多模型对比、丰富助手市场——LobeChat 是综合最强；纯桌面体验 Cherry Studio 同样优秀。","https:\u002F\u002Flobehub.com","zXTKiUcNATcrx3M_OoChdSfMOK8DHdQfiQ4Y7Ebo0_0",{"id":2671,"title":231,"alternatives":2672,"api_compatible":15,"body":2673,"category":460,"chinese_friendly":446,"cover":3090,"description":3091,"domestic":463,"extension":464,"faq":3092,"free":463,"github":15,"languages":3105,"lastVerified":15,"meta":3106,"models":15,"navigation":469,"notSuitable":15,"opensource":469,"path":872,"pillar":471,"platforms":3107,"priceTable":3108,"pricing":3112,"published":948,"relatedPlaybooks":3113,"relatedReviews":15,"score":3114,"self_host":469,"seo":3115,"seoTitle":3116,"slug":12,"sources":3117,"stem":3123,"suitable":15,"tagline":3124,"tags":3125,"updated":959,"verdict":3130,"website":3131,"__hash__":3132},"tools\u002Ftools\u002Fcoding\u002Flocal\u002Follama.md",[13,501,955,500],{"type":17,"value":2674,"toc":3078},[2675,2677,2684,2687,2689,2757,2759,2762,2766,2770,2790,2794,2825,2827,2861,2863,2976,2978,3010,3012,3035,3037,3055,3057],[20,2676,23],{"id":22},[25,2678,2679,2680,2683],{},"Ollama 是本地 LLM 的 Daemon 事实标准——后台跑、暴露 REST API（11434）+ CLI、Modelfile 配置、GGUF 一站式。MIT 开源，跨 Win \u002F Mac \u002F Linux。0.19+ 起 Mac M 系列底层切 MLX 推理。模型库覆盖 Llama \u002F Qwen \u002F DeepSeek \u002F Gemma \u002F Mistral 等主流开源模型，",[212,2681,2682],{},"ollama pull"," 一键拉。",[25,2685,2686],{},"适合：给 Cursor \u002F Cline \u002F Continue \u002F Open WebUI 接本地 OpenAI 兼容 endpoint、个人 \u002F 评估 \u002F 原型、嵌入应用、自动化脚本。不适合：GUI 偏好用户（用 LM Studio）、多用户并发生产服务（用 vLLM）、模型浏览 \u002F 调参界面（用 LM Studio）。",[20,2688,38],{"id":38},[40,2690,2691,2697,2705,2711,2718,2733,2739,2745,2751],{},[43,2692,2693,2696],{},[29,2694,2695],{},"后台 Daemon","：开机自启，应用调用零延迟",[43,2698,2699,1750,2702],{},[29,2700,2701],{},"CLI",[212,2703,2704],{},"ollama pull \u002F run \u002F list \u002F show \u002F create \u002F serve",[43,2706,2707,2710],{},[29,2708,2709],{},"Modelfile","：类 Dockerfile 注册任意 GGUF，配 SYSTEM \u002F PARAMETER \u002F TEMPLATE",[43,2712,2713,1750,2715],{},[29,2714,71],{},[212,2716,2717],{},"http:\u002F\u002Flocalhost:11434\u002Fv1\u002Fchat\u002Fcompletions",[43,2719,2720,1750,2723,2726,2727,2726,2730],{},[29,2721,2722],{},"原生 API",[212,2724,2725],{},"\u002Fapi\u002Fchat","、",[212,2728,2729],{},"\u002Fapi\u002Fgenerate",[212,2731,2732],{},"\u002Fapi\u002Fembeddings",[43,2734,2735,2738],{},[29,2736,2737],{},"模型库","：官方注册表内置 Llama \u002F Qwen \u002F DeepSeek \u002F Gemma \u002F Mistral \u002F GPT-OSS 等",[43,2740,2741,2744],{},[29,2742,2743],{},"MLX 加速（Mac）","：0.19+ 起 M 系列自动用 MLX",[43,2746,2747,2750],{},[29,2748,2749],{},"量化","：默认 Q4_K_M、支持 Q5 \u002F Q8 \u002F FP16",[43,2752,2753,2756],{},[29,2754,2755],{},"跨平台","：Win \u002F Mac \u002F Linux 安装包，Docker 官方镜像",[20,2758,93],{"id":93},[25,2760,2761],{},"完全免费、MIT 开源、商用免费。",[20,2763,2765],{"id":2764},"实测m2-pro-qwen3-coder-7b-q4","实测（M2 Pro + Qwen3-Coder-7B Q4）",[25,2767,2768],{},[29,2769,147],{},[40,2771,2772,2778,2781,2784,2787],{},[43,2773,2774,2777],{},[212,2775,2776],{},"ollama run qwen3-coder:7b"," 一行起飞，3 秒进交互",[43,2779,2780],{},"REST API 配 Cursor \u002F Cline \u002F Continue 几乎全工具开箱即用",[43,2782,2783],{},"Modelfile 写自定义编码助手（low temperature + system prompt + 16K context）几分钟搞定",[43,2785,2786],{},"多模型并存，按需切换，内存占用合理",[43,2788,2789],{},"Mac M 系列 MLX 后比旧 GGUF 模式快显著",[25,2791,2792],{},[29,2793,169],{},[40,2795,2796,2806,2812,2819,2822],{},[43,2797,2798,2799,2802,2803],{},"默认 ",[212,2800,2801],{},"num_ctx"," 偏小（2048），跑长上下文要在 Modelfile 加 ",[212,2804,2805],{},"PARAMETER num_ctx 16384",[43,2807,2808,2809],{},"模型默认走 0.0.0.0:11434 ↔ Docker 容器互访要 ",[212,2810,2811],{},"--add-host=host.docker.internal:host-gateway",[43,2813,2814,2815,2818],{},"国内 ",[212,2816,2817],{},"ollama.com\u002Flibrary"," 下载偶有慢，可手动 HF 下 GGUF + Modelfile 自建",[43,2820,2821],{},"多用户并发吞吐显著低于 vLLM",[43,2823,2824],{},"没有 GUI，模型浏览 \u002F 参数面板要走 LM Studio \u002F Open WebUI 配合",[20,2826,189],{"id":189},[191,2828,2829,2835,2841,2846,2852,2858],{},[43,2830,2831,2834],{},[212,2832,2833],{},"curl -fsSL https:\u002F\u002Follama.ai\u002Finstall.sh | sh","（Mac \u002F Linux）；Windows winget",[43,2836,2837,2840],{},[212,2838,2839],{},"ollama pull qwen3-coder:7b","（按需换模型）",[43,2842,2843,2845],{},[212,2844,2776],{}," 直接聊",[43,2847,2848,2849],{},"应用接入：baseURL = ",[212,2850,2851],{},"http:\u002F\u002Flocalhost:11434\u002Fv1",[43,2853,2854,2855],{},"自定义：写 Modelfile → ",[212,2856,2857],{},"ollama create my-coder -f Modelfile",[43,2859,2860],{},"进阶：装 Open WebUI 做前端 \u002F 多人共享",[20,2862,217],{"id":217},[101,2864,2865,2879],{},[104,2866,2867],{},[107,2868,2869,2871,2873,2875,2877],{},[110,2870,226],{},[110,2872,231],{},[110,2874,234],{},[110,2876,414],{},[110,2878,1895],{},[119,2880,2881,2897,2909,2922,2935,2951,2963],{},[107,2882,2883,2885,2888,2891,2894],{},[124,2884,676],{},[124,2886,2887],{},"CLI + Daemon",[124,2889,2890],{},"GUI + Headless",[124,2892,2893],{},"Python Server",[124,2895,2896],{},"C++ 二进制",[107,2898,2899,2901,2903,2905,2907],{},[124,2900,189],{},[124,2902,2004],{},[124,2904,2004],{},[124,2906,1490],{},[124,2908,1487],{},[107,2910,2911,2913,2915,2918,2920],{},[124,2912,1918],{},[124,2914,2701],{},[124,2916,2917],{},"✅ GUI",[124,2919,743],{},[124,2921,743],{},[107,2923,2924,2927,2929,2931,2933],{},[124,2925,2926],{},"OpenAI 兼容",[124,2928,1953],{},[124,2930,1950],{},[124,2932,246],{},[124,2934,246],{},[107,2936,2937,2940,2943,2946,2949],{},[124,2938,2939],{},"多用户吞吐",[124,2941,2942],{},"弱（~40 tok\u002Fs）",[124,2944,2945],{},"中（50–90）",[124,2947,2948],{},"强（800–12500）",[124,2950,1490],{},[107,2952,2953,2955,2957,2959,2961],{},[124,2954,1962],{},[124,2956,1967],{},[124,2958,246],{},[124,2960,1941],{},[124,2962,789],{},[107,2964,2965,2967,2969,2971,2974],{},[124,2966,307],{},[124,2968,310],{},[124,2970,315],{},[124,2972,2973],{},"Apache 2.0",[124,2975,310],{},[20,2977,318],{"id":318},[40,2979,2980,2986,2992,2998,3004],{},[43,2981,2982,2985],{},[29,2983,2984],{},"num_ctx 一定要设","：默认 2K 太小，跑代码 \u002F 长文档要 16K+",[43,2987,2988,2991],{},[29,2989,2990],{},"Modelfile 模板别漏 TEMPLATE","：错的 chat template 会让模型输出乱码 \u002F 不停",[43,2993,2994,2997],{},[29,2995,2996],{},"KV cache 爆表 = 速度悬崖","：32B 模型 32K 上下文，KV cache 可能 12+ GB，超显存自动 offload 慢 10×",[43,2999,3000,3003],{},[29,3001,3002],{},"不要 0.0.0.0 直接对公网","：默认无鉴权，对外暴露走反代 + Bearer \u002F mTLS",[43,3005,3006,3009],{},[29,3007,3008],{},"Mac 让它自动用 MLX","：升 0.19+；不要手动强制 GGUF + Metal",[20,3011,354],{"id":353},[40,3013,3014,3017,3020,3023,3026,3029,3032],{},[43,3015,3016],{},"✅ 应用 \u002F IDE 接本地模型（Cursor \u002F Cline \u002F Continue）",[43,3018,3019],{},"✅ 个人 \u002F 评估 \u002F 脚本自动化",[43,3021,3022],{},"✅ Modelfile 自定义系统 prompt + 参数",[43,3024,3025],{},"✅ Mac M 系列 MLX 用户",[43,3027,3028],{},"❌ 多用户并发生产服务（用 vLLM）",[43,3030,3031],{},"❌ GUI 调参 \u002F 模型浏览（配 LM Studio \u002F Open WebUI）",[43,3033,3034],{},"❌ 极致单卡吞吐研究（直接 llama.cpp \u002F vLLM）",[20,3036,402],{"id":402},[40,3038,3039,3043,3047,3051],{},[43,3040,3041],{},[406,3042,867],{"href":866},[43,3044,3045],{},[406,3046,2084],{"href":2083},[43,3048,3049],{},[406,3050,2089],{"href":935},[43,3052,3053],{},[406,3054,879],{"href":878},[20,3056,421],{"id":421},[191,3058,3059,3066,3073],{},[43,3060,3061,3062],{},"Markaicode — Import GGUF Models into Ollama 2026（2026-05-15）",[406,3063,3064],{"href":3064,"rel":3065},"https:\u002F\u002Fmarkaicode.com\u002Fimport-gguf-models-ollama-guide",[435],[43,3067,3068,3069],{},"ComputingForGeeks — Ollama Models Cheat Sheet 2026 ",[406,3070,3071],{"href":3071,"rel":3072},"https:\u002F\u002Fcomputingforgeeks.com\u002Follama-models-cheat-sheet",[435],[43,3074,2116,3075],{},[406,3076,2119],{"href":2119,"rel":3077},[435],{"title":445,"searchDepth":446,"depth":446,"links":3079},[3080,3081,3082,3083,3084,3085,3086,3087,3088,3089],{"id":22,"depth":449,"text":23},{"id":38,"depth":449,"text":38},{"id":93,"depth":449,"text":93},{"id":2764,"depth":449,"text":2765},{"id":189,"depth":449,"text":189},{"id":217,"depth":449,"text":217},{"id":318,"depth":449,"text":318},{"id":353,"depth":449,"text":354},{"id":402,"depth":449,"text":402},{"id":421,"depth":449,"text":421},"\u002Fimg\u002Ftools\u002Follama.webp","Ollama 真实评测：本地 LLM 的事实标准 Daemon，CLI + REST API，模型库 + Modelfile + GGUF 一站式。0.19+ 在 Mac M 系列用 MLX 加速；OpenAI 兼容端点 11434；MIT 开源 + 跨平台。",[3093,3096,3099,3102],{"q":3094,"a":3095},"和 LM Studio 怎么选？","Ollama = Daemon + CLI，开机自启在 11434 端口跑，应用 \u002F IDE 调它最方便。LM Studio = GUI，模型浏览 \u002F 调参 \u002F 聊天体验更好。两者底层都基于 llama.cpp，Mac M 系列上都已切 MLX。",{"q":3097,"a":3098},"Modelfile 是什么？","类 Dockerfile 的模型配置：`FROM .\u002Fxxx.gguf` + PARAMETER \u002F TEMPLATE \u002F SYSTEM。把任意 GGUF 注册成本地模型供调用。`ollama create my-model -f Modelfile`。",{"q":3100,"a":3101},"OpenAI 兼容端点？","`http:\u002F\u002Flocalhost:11434\u002Fv1`。任何 OpenAI SDK 改 baseURL 即用。也可走原生 `\u002Fapi\u002Fchat`、`\u002Fapi\u002Fgenerate`。",{"q":3103,"a":3104},"并发能力？","单用户原型场景顺滑（~40 tok\u002Fs peak），多用户并发明显不如 vLLM（vLLM 的 PagedAttention + 连续批处理高 16–20×）。生产并发选 vLLM。",[466],{},[473,474,475,2641],[3109],{"plan":126,"price":940,"features":3110,"notes":3111},"完整 CLI + REST API + Modelfile + 模型库 + MIT 协议","全平台、商用免费","完全免费 + 开源（MIT）",[950,2161],{"power":479,"ux":479,"price":480,"cn_support":446,"stability":480},{"title":231,"description":3091},"Ollama 评测 2026：本地运行大模型，开源 AI 模型管理工具，私有化部署指南",[3118,3120,3122],{"name":3119,"url":3064,"accessed":959},"Markaicode — Import GGUF 2026",{"name":3121,"url":3071,"accessed":959},"ComputingForGeeks — Ollama Cheat Sheet 2026",{"name":2171,"url":2119,"accessed":959},"tools\u002Fcoding\u002Flocal\u002Follama","本地 LLM 的 Daemon——CLI + REST API 后台跑，给 Cursor \u002F Cline \u002F Open WebUI 接本地模型最低门槛",[460,3126,3127,3128,3129,492,2176,2179,971],"daemon","cli","rest-api","modelfile","本地 LLM 的 Daemon 事实标准，CLI \u002F Modelfile \u002F REST API 三件套配合最广泛。GUI 偏好用户走 LM Studio；多用户并发生产用 vLLM；其他场景几乎默认 Ollama。","https:\u002F\u002Follama.com","7WgXNX9uMzSH_c-dUkicVopZk_g1z8HaB_8FmD9fBps",{"id":3134,"title":669,"alternatives":3135,"api_compatible":15,"body":3136,"category":460,"chinese_friendly":479,"cover":3551,"description":3552,"domestic":463,"extension":464,"faq":3553,"free":463,"github":15,"languages":3566,"lastVerified":15,"meta":3567,"models":15,"navigation":469,"notSuitable":15,"opensource":469,"path":2083,"pillar":471,"platforms":3568,"priceTable":3570,"pricing":3577,"published":948,"relatedPlaybooks":3578,"relatedReviews":15,"score":3579,"self_host":469,"seo":3580,"seoTitle":3581,"slug":501,"sources":3582,"stem":3589,"suitable":15,"tagline":3590,"tags":3591,"updated":959,"verdict":3595,"website":3596,"__hash__":3597},"tools\u002Ftools\u002Fcoding\u002Flocal\u002Fopen-webui.md",[500,955,12,13],{"type":17,"value":3137,"toc":3539},[3138,3140,3143,3146,3148,3210,3212,3215,3219,3223,3251,3255,3279,3281,3315,3317,3424,3426,3469,3471,3494,3496,3514,3516],[20,3139,23],{"id":22},[25,3141,3142],{},"Open WebUI（原 Ollama WebUI）是 MIT 开源、自托管 AI 平台，最常见用法是 Docker 跑起来给 Ollama 套一个 ChatGPT 风格前端。GitHub 126k+ stars、282M+ Docker pulls，事实上的本地 AI 前端首选。支持任意 OpenAI 兼容后端 + RAG 知识库 + 多用户账号 + 工具调用 + MCP-OpenAPI 代理 + 联网搜索 + 语音 + 图像生成。",[25,3144,3145],{},"适合：团队 \u002F 家庭 \u002F 公司部署一份共享、要 Web 端访问、多用户分账号、SearXNG 联网搜索、Confluence \u002F S3 \u002F GitHub 数据源同步。不适合：单人桌面体验（用 Cherry Studio）、零运维 \u002F 不愿碰 Docker。",[20,3147,38],{"id":38},[40,3149,3150,3156,3162,3168,3174,3180,3186,3192,3198,3204],{},[43,3151,3152,3155],{},[29,3153,3154],{},"多模型后端","：Ollama \u002F OpenAI \u002F vLLM \u002F Anthropic \u002F Groq \u002F LocalAI \u002F 任意 OpenAI 兼容",[43,3157,3158,3161],{},[29,3159,3160],{},"多用户 + RBAC","：注册 \u002F 邀请 \u002F 角色权限 \u002F 工作区隔离",[43,3163,3164,3167],{},[29,3165,3166],{},"RAG 知识库","：上传文档 \u002F 网址 \u002F SearXNG 联网搜索 → 向量化 → 对话引用",[43,3169,3170,3173],{},[29,3171,3172],{},"Tools \u002F Functions","：Python 写函数即扩展（联网 \u002F 计算器 \u002F 自定义 API）",[43,3175,3176,3179],{},[29,3177,3178],{},"mcpo","：MCP-to-OpenAPI 代理，任意 MCP 服务器接进来",[43,3181,3182,3185],{},[29,3183,3184],{},"oikb","：知识库同步本地文件夹 \u002F GitHub \u002F S3 \u002F Confluence 等 40+ 源",[43,3187,3188,3191],{},[29,3189,3190],{},"open-terminal \u002F cptr","：给 AI 真实终端 + 文件 + 沙箱执行",[43,3193,3194,3197],{},[29,3195,3196],{},"图像生成","：Stable Diffusion \u002F DALL·E \u002F 自托管接入",[43,3199,3200,3203],{},[29,3201,3202],{},"语音输入 \u002F TTS","：内置",[43,3205,3206,3209],{},[29,3207,3208],{},"企业 LTS","：custom branding + SLA + 长期支持版本（联系销售）",[20,3211,93],{"id":93},[25,3213,3214],{},"完全免费、MIT 开源、商用免费。Enterprise 提供品牌定制 + SLA + LTS。",[20,3216,3218],{"id":3217},"实测ubuntu-2404-ollama-后端-5-人小团队","实测（Ubuntu 24.04 + Ollama 后端 + 5 人小团队）",[25,3220,3221],{},[29,3222,147],{},[40,3224,3225,3232,3235,3242,3245,3248],{},[43,3226,3227,3228,3231],{},"单条 ",[212,3229,3230],{},"docker run"," 五分钟上线",[43,3233,3234],{},"自带的多用户 + 角色权限省去重新搭 Auth",[43,3236,3237,3238,3241],{},"RAG 直传 30 个 PDF 后向量化顺利，对话中 ",[212,3239,3240],{},"#知识库"," 引用准确",[43,3243,3244],{},"mcpo 把 GitHub MCP 服务器接进来，团队对话里直接 issue \u002F PR 操作",[43,3246,3247],{},"模型切换流畅，OpenAI + Ollama 并存",[43,3249,3250],{},"SearXNG 联网搜索给模型实时信息，过时知识截止问题缓解",[25,3252,3253],{},[29,3254,169],{},[40,3256,3257,3260,3266,3273,3276],{},[43,3258,3259],{},"Docker 镜像 ~1.5GB，首次拉取偏慢",[43,3261,2798,3262,3265],{},[212,3263,3264],{},"0.0.0.0"," 公网暴露要加 HTTPS + 反代",[43,3267,3268,3269,3272],{},"嵌入模型 ",[212,3270,3271],{},"sentence-transformers"," 中文效果一般，建议换 bge-m3",[43,3274,3275],{},"多用户共享 Ollama 时并发吞吐瓶颈在 Ollama，不在 Open WebUI（生产用 vLLM 后端）",[43,3277,3278],{},"版本升级要看 changelog，部分 minor 含 breaking 改动",[20,3280,189],{"id":189},[191,3282,3283,3289,3296,3299,3302,3305,3308],{},[43,3284,3285,3286],{},"装 Docker → ",[212,3287,3288],{},"docker run -d -p 3000:8080 --add-host=host.docker.internal:host-gateway -v open-webui:\u002Fapp\u002Fbackend\u002Fdata --name open-webui --restart always ghcr.io\u002Fopen-webui\u002Fopen-webui:main",[43,3290,3291,3292,3295],{},"浏览器开 ",[212,3293,3294],{},"http:\u002F\u002Flocalhost:3000"," → 注册第一个账号（管理员）",[43,3297,3298],{},"设置 → Connections → 连接 Ollama \u002F 加 OpenAI Key",[43,3300,3301],{},"Models → Pull \u002F Discover 模型",[43,3303,3304],{},"Workspaces → 建知识库 → 上传文档",[43,3306,3307],{},"Tools → 启用 \u002F 写自定义函数",[43,3309,3310,3311,3314],{},"生产部署：Nginx 反代 + Let's Encrypt + 备份 ",[212,3312,3313],{},"\u002Fapp\u002Fbackend\u002Fdata"," volume",[20,3316,217],{"id":217},[101,3318,3319,3333],{},[104,3320,3321],{},[107,3322,3323,3325,3327,3329,3331],{},[110,3324,226],{},[110,3326,669],{},[110,3328,664],{},[110,3330,498],{},[110,3332,234],{},[119,3334,3335,3347,3359,3373,3386,3400,3412],{},[107,3336,3337,3339,3341,3343,3345],{},[124,3338,676],{},[124,3340,687],{},[124,3342,682],{},[124,3344,679],{},[124,3346,679],{},[107,3348,3349,3351,3353,3355,3357],{},[124,3350,2463],{},[124,3352,2410],{},[124,3354,246],{},[124,3356,743],{},[124,3358,743],{},[107,3360,3361,3364,3367,3369,3371],{},[124,3362,3363],{},"RAG",[124,3365,3366],{},"✅ 强 + oikb",[124,3368,246],{},[124,3370,246],{},[124,3372,714],{},[107,3374,3375,3378,3380,3382,3384],{},[124,3376,3377],{},"工具 \u002F MCP",[124,3379,2456],{},[124,3381,246],{},[124,3383,246],{},[124,3385,714],{},[107,3387,3388,3391,3394,3396,3398],{},[124,3389,3390],{},"自托管",[124,3392,3393],{},"✅ Docker \u002F K8s",[124,3395,740],{},[124,3397,743],{},[124,3399,743],{},[107,3401,3402,3404,3406,3408,3410],{},[124,3403,780],{},[124,3405,792],{},[124,3407,786],{},[124,3409,783],{},[124,3411,789],{},[107,3413,3414,3416,3418,3420,3422],{},[124,3415,765],{},[124,3417,310],{},[124,3419,310],{},[124,3421,768],{},[124,3423,773],{},[20,3425,318],{"id":318},[40,3427,3428,3434,3442,3451,3457,3463],{},[43,3429,3430,3433],{},[29,3431,3432],{},"不要裸 0.0.0.0 + HTTP 暴露公网","：默认无 HTTPS，必上反代 + 强密码 + 速率限制",[43,3435,3436,3441],{},[29,3437,3438,3439,3314],{},"备份 ",[212,3440,3313],{},"：知识库 \u002F 用户 \u002F 对话全在里面",[43,3443,3444,3447,3448,3450],{},[29,3445,3446],{},"中文 RAG 换嵌入模型","：默认 ",[212,3449,3271],{}," 中文一般，配 bge-m3 或硅基流动嵌入 API",[43,3452,3453,3456],{},[29,3454,3455],{},"mcpo 工具范围谨慎","：MCP 给 AI 真实能力，第三方服务器审一遍",[43,3458,3459,3462],{},[29,3460,3461],{},"后端吞吐看 Ollama","：5+ 并发上 vLLM 后端，Ollama 单 worker 会排队",[43,3464,3465,3468],{},[29,3466,3467],{},"升级前看 changelog","：weekly 更新，偶有 breaking",[20,3470,354],{"id":353},[40,3472,3473,3476,3479,3482,3485,3488,3491],{},[43,3474,3475],{},"✅ 团队 \u002F 家庭 \u002F 公司多人共享 AI 平台",[43,3477,3478],{},"✅ 要 Web 端访问 \u002F 移动端兼容",[43,3480,3481],{},"✅ 自托管 \u002F 完全控制数据",[43,3483,3484],{},"✅ MCP \u002F 工具调用刚需",[43,3486,3487],{},"❌ 单人桌面体验（用 Cherry Studio）",[43,3489,3490],{},"❌ 零运维 \u002F 不愿碰 Docker",[43,3492,3493],{},"❌ iOS 原生 App 主力",[20,3495,402],{"id":402},[40,3497,3498,3502,3506,3510],{},[43,3499,3500],{},[406,3501,861],{"href":860},[43,3503,3504],{},[406,3505,2089],{"href":935},[43,3507,3508],{},[406,3509,873],{"href":872},[43,3511,3512],{},[406,3513,879],{"href":878},[20,3515,421],{"id":421},[191,3517,3518,3525,3532],{},[43,3519,3520,3521],{},"Open WebUI 官方文档 ",[406,3522,3523],{"href":3523,"rel":3524},"https:\u002F\u002Fdocs.openwebui.com\u002F",[435],[43,3526,3527,3528],{},"Local AI Master — Open WebUI Setup Guide 2026 ",[406,3529,3530],{"href":3530,"rel":3531},"https:\u002F\u002Flocalaimaster.com\u002Fblog\u002Fopen-webui-setup-guide",[435],[43,3533,3534,3535],{},"AIToolDiscovery — Set Up Open-WebUI with Ollama 2026 ",[406,3536,3537],{"href":3537,"rel":3538},"https:\u002F\u002Fwww.aitooldiscovery.com\u002Fhow-to\u002Fsetup-open-webui-ollama",[435],{"title":445,"searchDepth":446,"depth":446,"links":3540},[3541,3542,3543,3544,3545,3546,3547,3548,3549,3550],{"id":22,"depth":449,"text":23},{"id":38,"depth":449,"text":38},{"id":93,"depth":449,"text":93},{"id":3217,"depth":449,"text":3218},{"id":189,"depth":449,"text":189},{"id":217,"depth":449,"text":217},{"id":318,"depth":449,"text":318},{"id":353,"depth":449,"text":354},{"id":402,"depth":449,"text":402},{"id":421,"depth":449,"text":421},"\u002Fimg\u002Ftools\u002Fopen-webui.webp","Open WebUI 2026 真实评测：MIT 开源、自托管 ChatGPT 替代和 Ollama Web 前端。支持 Docker 一行部署、Ollama\u002FOpenAI\u002FvLLM 多后端、RAG 知识库、多用户、联网搜索、工具调用和 MCP-to-OpenAPI，适合团队私有 AI 平台。",[3554,3557,3560,3563],{"q":3555,"a":3556},"Docker 一行命令真的够用吗？","够。`docker run -d -p 3000:8080 --add-host=host.docker.internal:host-gateway -v open-webui:\u002Fapp\u002Fbackend\u002Fdata --name open-webui --restart always ghcr.io\u002Fopen-webui\u002Fopen-webui:main`，5 分钟可上线、能多人注册、能接 Ollama \u002F OpenAI。生产再加反代 + HTTPS + 备份。",{"q":3558,"a":3559},"支持哪些模型后端？","Ollama（首选）+ 任何 OpenAI 兼容 endpoint：OpenAI 官方 \u002F Anthropic（OpenAI 兼容代理）\u002F vLLM \u002F Groq \u002F LocalAI \u002F 自建 baseURL。可同时配多个，对话中切换。",{"q":3561,"a":3562},"RAG \u002F 知识库怎么做？","内置：上传 PDF \u002F DOCX \u002F TXT、网址抓取、SearXNG 联网搜索 → 自动向量化 → 在对话中 `#` 引用知识库。配套 oikb 项目可同步本地文件夹 \u002F GitHub \u002F S3 \u002F Confluence 等 40+ 数据源。",{"q":3564,"a":3565},"MCP 怎么接？","通过 mcpo（官方的 MCP-to-OpenAPI 代理）把任意 MCP 服务器暴露成 OpenAPI 工具，再在 Open WebUI 注册即可。无需写 glue code。",[466,933],{},[2641,475,474,473,3569],"kubernetes",[3571,3573],{"plan":126,"price":940,"features":3572,"notes":2645},"全功能 \u002F 多用户 \u002F RAG \u002F Tools \u002F 联网搜索 \u002F MCP-OpenAPI 代理 \u002F Docker \u002F K8s",{"plan":571,"price":3574,"features":3575,"notes":3576},"咨询","Custom branding \u002F SLA \u002F LTS 长期支持版本","邮件官方","完全免费（MIT 开源） \u002F Enterprise SLA 联系",[950,2161],{"power":480,"ux":479,"price":480,"cn_support":479,"stability":480},{"title":669,"description":3552},"Open WebUI 评测 2026：自托管 ChatGPT 替代，Ollama 前端部署指南",[3583,3585,3587],{"name":3584,"url":3523,"accessed":959},"Open WebUI 官方文档",{"name":3586,"url":3530,"accessed":959},"Local AI Master — Open WebUI Setup Guide 2026",{"name":3588,"url":3537,"accessed":959},"AIToolDiscovery — Open-WebUI with Ollama 2026","tools\u002Fcoding\u002Flocal\u002Fopen-webui","自托管的 ChatGPT 替代：Ollama \u002F OpenAI 兼容、多用户、RAG、126k+ GitHub stars",[460,3592,2641,970,3593,3594,971],"self-host","multi-user","ollama","自托管多用户 AI 前端的事实标准。团队 \u002F 家庭 \u002F 公司部署一份共享，多模型聚合 + RAG + 工具调用全有。单机 \u002F 桌面体验首选 Cherry Studio \u002F LobeChat。","https:\u002F\u002Fdocs.openwebui.com","JCKn_X0aojpl94LKJcqlbs0XSTAxv8S8JgKy70WtsO0",{"id":3599,"title":414,"alternatives":3600,"api_compatible":15,"body":3601,"category":460,"chinese_friendly":449,"cover":4077,"description":4078,"domestic":463,"extension":464,"faq":15,"free":463,"github":4062,"languages":4079,"lastVerified":467,"meta":4080,"models":15,"navigation":469,"notSuitable":15,"opensource":469,"path":4081,"pillar":471,"platforms":4082,"priceTable":15,"pricing":4083,"published":477,"relatedPlaybooks":15,"relatedReviews":15,"score":4084,"self_host":463,"seo":4085,"seoTitle":4086,"slug":4087,"sources":4088,"stem":4091,"suitable":15,"tagline":4092,"tags":4093,"updated":467,"verdict":4098,"website":4056,"__hash__":4099},"tools\u002Ftools\u002Fcoding\u002Flocal\u002Fvllm.md",[12,13],{"type":17,"value":3602,"toc":4064},[3603,3605,3608,3611,3613,3673,3675,3678,3680,3688,3692,3715,3719,3751,3753,3793,3795,3924,3926,3980,3982,4008,4010,4016,4022,4028,4034,4036,4044,4046,4050],[20,3604,23],{"id":22},[25,3606,3607],{},"vLLM 是当前开源生态吞吐量最高的 LLM 推理引擎，由 UC Berkeley 团队开发，核心创新 PagedAttention 把 KV cache 当虚拟内存管，配合连续批处理（continuous batching）把 GPU 利用率从传统推理的 30-40% 拉到 70-80%+。Apache 2.0 协议，纯 Python + CUDA，部署在 Linux + NVIDIA GPU。",[25,3609,3610],{},"适合：需要对外提供 LLM API 服务、多用户并发、追求最大吞吐和最低延迟的工程团队，以及跑大规模 batch 离线推理的研究场景。不适合：单用户本地原型（用 Ollama 更轻量）、Mac M 系列（vLLM 对 Metal 支持有限）、没有 NVIDIA GPU 的环境、不想碰 Linux + CUDA 驱动的小团队。",[20,3612,38],{"id":38},[40,3614,3615,3621,3627,3633,3646,3652,3661,3667],{},[43,3616,3617,3620],{},[29,3618,3619],{},"PagedAttention","：借鉴操作系统虚拟内存的分页机制管理 KV cache，消除碎片化，显存利用率提升 2-4 倍",[43,3622,3623,3626],{},[29,3624,3625],{},"连续批处理（Continuous Batching）","：请求动态插入 \u002F 弹出，不需要等整批完成，GPU 闲置接近为零",[43,3628,3629,3632],{},[29,3630,3631],{},"高并发吞吐","：单 A100 跑 Llama-3-8B 可达 800-12500 tok\u002Fs（取决于 batch size），比 Hugging Face Transformers 高 14-24 倍",[43,3634,3635,3637,3638,3641,3642,3645],{},[29,3636,71],{},"：内置 ",[212,3639,3640],{},"--api-server","，端点 ",[212,3643,3644],{},"\u002Fv1\u002Fchat\u002Fcompletions"," 直接替换 OpenAI SDK 的 baseURL 即用",[43,3647,3648,3651],{},[29,3649,3650],{},"量化支持","：AWQ、GPTQ、FP8（H100\u002FAda）、INT8 KV cache，显存减半吞吐不掉",[43,3653,3654,1750,3657,3660],{},[29,3655,3656],{},"张量并行（Tensor Parallelism）",[212,3658,3659],{},"--tensor-parallel-size N"," 多卡切分，支持多 GPU 推理大模型",[43,3662,3663,3666],{},[29,3664,3665],{},"分布式部署","：Ray 集群多节点推理，支持 pipeline parallelism",[43,3668,3669,3672],{},[29,3670,3671],{},"LoRA 多租户","：同时加载多个 LoRA adapter，单服务多模型，按请求路由",[20,3674,93],{"id":93},[25,3676,3677],{},"完全免费，Apache 2.0 开源，商用无限制。成本在于 GPU 硬件：一张 A100 80GB 云端约 $2-4\u002F小时（按需），跑 70B 模型需 2-4 张。自建机房摊薄后更便宜。",[20,3679,139],{"id":138},[95,3681,3682],{},[25,3683,144,3684,3687],{},[29,3685,3686],{},"环境（撰写时参考）","：4× A100 80GB + Llama-3-70B-Instruct（FP16），vLLM 0.6.x 系列（最新稳定版请以 vllm.ai 为准）。",[25,3689,3690],{},[29,3691,147],{},[40,3693,3694,3700,3703,3706,3709,3712],{},[43,3695,3696,3699],{},[212,3697,3698],{},"vllm serve meta-llama\u002FMeta-Llama-3-70B-Instruct --tensor-parallel-size 4"," 一行拉起，4 卡自动切分",[43,3701,3702],{},"并发 64 用户，平均延迟 1.2s，吞吐稳定在 3200 tok\u002Fs，GPU 利用率 75-85%",[43,3704,3705],{},"同样硬件跑 HF Transformers + 默认 batching，吞吐仅 ~200 tok\u002Fs，差距 16 倍",[43,3707,3708],{},"AWQ 量化版 70B 单卡 A100 即可跑，吞吐只掉 15-20%，显存从 140GB 降到 40GB",[43,3710,3711],{},"OpenAI 兼容端点接 Cursor \u002F Dify \u002F FastGPT 零改动",[43,3713,3714],{},"连续批处理下短请求和长请求混合调度公平，没有长尾饿死",[25,3716,3717],{},[29,3718,169],{},[40,3720,3721,3728,3735,3738,3745],{},[43,3722,3723,3724,3727],{},"第一次启动要编译 CUDA kernel，冷启动 3-5 分钟，加 ",[212,3725,3726],{},"--enforce-eager"," 可跳过但掉速 20%",[43,3729,3730,3731,3734],{},"KV cache 默认占 90% 显存，跑长上下文（32K+）要手动调 ",[212,3732,3733],{},"--gpu-memory-utilization 0.85"," 留余量",[43,3736,3737],{},"旧版本对 Qwen2.5-VL 等多模态模型支持不稳定，偶发 OOM，建议查阅官方 issue 选择适配版本",[43,3739,3740,3741,3744],{},"国内 HuggingFace 下载模型慢，配 ",[212,3742,3743],{},"HF_ENDPOINT=https:\u002F\u002Fhf-mirror.com"," 或预下载到本地",[43,3746,3747,3750],{},[212,3748,3749],{},"--max-model-len"," 必须设，否则默认按模型最大上下文分配，32B 模型 128K 上下文会直接 OOM",[20,3752,189],{"id":189},[191,3754,3755,3761,3767,3773,3780,3786],{},[43,3756,3757,3758],{},"环境准备：Linux + NVIDIA GPU（compute capability ≥ 7.0）+ CUDA 12.1+，",[212,3759,3760],{},"pip install vllm",[43,3762,3763,3764],{},"拉起服务：",[212,3765,3766],{},"vllm serve meta-llama\u002FMeta-Llama-3-8B-Instruct --port 8000",[43,3768,3769,3770],{},"测试调用：",[212,3771,3772],{},"curl http:\u002F\u002Flocalhost:8000\u002Fv1\u002Fchat\u002Fcompletions -H \"Content-Type: application\u002Fjson\" -d '{\"model\":\"meta-llama\u002FMeta-Llama-3-8B-Instruct\",\"messages\":[{\"role\":\"user\",\"content\":\"hi\"}]}'",[43,3774,3775,3776,3779],{},"多卡并行：加 ",[212,3777,3778],{},"--tensor-parallel-size 4","（卡数）",[43,3781,3782,3783],{},"量化部署：",[212,3784,3785],{},"vllm serve TheBloke\u002FLlama-2-13B-AWQ --quantization awq",[43,3787,3788,3789,3792],{},"接入应用：任何 OpenAI SDK 改 ",[212,3790,3791],{},"base_url=http:\u002F\u002Flocalhost:8000\u002Fv1"," 即用",[20,3794,217],{"id":217},[101,3796,3797,3813],{},[104,3798,3799],{},[107,3800,3801,3803,3805,3807,3810],{},[110,3802,226],{},[110,3804,414],{},[110,3806,231],{},[110,3808,3809],{},"TGI (HF)",[110,3811,3812],{},"TensorRT-LLM",[119,3814,3815,3832,3845,3858,3871,3884,3896,3911],{},[107,3816,3817,3820,3823,3826,3829],{},[124,3818,3819],{},"吞吐（A100 8B）",[124,3821,3822],{},"~800-12500 tok\u002Fs",[124,3824,3825],{},"~40 tok\u002Fs",[124,3827,3828],{},"~500 tok\u002Fs",[124,3830,3831],{},"~10000 tok\u002Fs",[107,3833,3834,3837,3839,3841,3843],{},[124,3835,3836],{},"上手门槛",[124,3838,1490],{},[124,3840,2004],{},[124,3842,1490],{},[124,3844,1487],{},[107,3846,3847,3849,3851,3853,3856],{},[124,3848,3619],{},[124,3850,246],{},[124,3852,1532],{},[124,3854,3855],{},"✅ (v0.7+)",[124,3857,1532],{},[107,3859,3860,3863,3865,3867,3869],{},[124,3861,3862],{},"连续批处理",[124,3864,246],{},[124,3866,1532],{},[124,3868,246],{},[124,3870,246],{},[107,3872,3873,3875,3877,3879,3881],{},[124,3874,2926],{},[124,3876,246],{},[124,3878,246],{},[124,3880,246],{},[124,3882,3883],{},"需封装",[107,3885,3886,3888,3890,3892,3894],{},[124,3887,550],{},[124,3889,1941],{},[124,3891,246],{},[124,3893,246],{},[124,3895,1941],{},[107,3897,3898,3901,3904,3907,3909],{},[124,3899,3900],{},"Mac 支持",[124,3902,3903],{},"❌ 有限",[124,3905,3906],{},"✅ MLX",[124,3908,1532],{},[124,3910,1532],{},[107,3912,3913,3915,3917,3919,3922],{},[124,3914,765],{},[124,3916,2973],{},[124,3918,310],{},[124,3920,3921],{},"HFOIL",[124,3923,2973],{},[20,3925,318],{"id":318},[40,3927,3928,3937,3945,3959,3965,3974],{},[43,3929,3930,3933,3934,3936],{},[29,3931,3932],{},"冷启动慢不是 bug","：首次编译 CUDA kernel 需要几分钟，生产环境用 Docker 镜像预编译或加 ",[212,3935,3726],{},"（牺牲 15-20% 性能换即时启动）",[43,3938,3939,3944],{},[29,3940,3941,3943],{},[212,3942,3749],{}," 必设","：不设会按模型最大上下文预分配 KV cache，小显存直接 OOM",[43,3946,3947,3950,3951,3954,3955,3958],{},[29,3948,3949],{},"量化模型要匹配版本","：AWQ 模型必须用 ",[212,3952,3953],{},"--quantization awq","，GPTQ 用 ",[212,3956,3957],{},"--quantization gptq","，混用会报错或精度崩",[43,3960,3961,3964],{},[29,3962,3963],{},"不要在 Mac 上用 vLLM 跑生产","：Metal 后端是实验性的，性能远不如 CPU，Mac 本地推理用 Ollama \u002F MLX",[43,3966,3967,3970,3971,3973],{},[29,3968,3969],{},"监控 GPU 显存碎片","：长跑后偶发显存碎片导致新请求 OOM，加 ",[212,3972,3733],{}," 留 buffer 或定期重启",[43,3975,3976,3979],{},[29,3977,3978],{},"多模态模型看版本","：不同版本对 VLM 支持差异较大，新模型先查官方 issue 选适配版本",[20,3981,354],{"id":353},[40,3983,3984,3987,3990,3993,3996,3999,4002,4005],{},[43,3985,3986],{},"✅ 生产级 LLM API 服务（多用户并发、高吞吐）",[43,3988,3989],{},"✅ 大规模离线 batch 推理（数据标注、合成数据生成）",[43,3991,3992],{},"✅ 需要最低成本跑大模型（量化 + 单卡部署 70B）",[43,3994,3995],{},"✅ 有 NVIDIA GPU + Linux 运维能力的工程团队",[43,3997,3998],{},"❌ 单用户本地原型 \u002F 个人开发（用 Ollama，0 配置）",[43,4000,4001],{},"❌ Mac M 系列用户（Metal 支持有限，用 Ollama + MLX）",[43,4003,4004],{},"❌ 没有 GPU 的环境（vLLM 的 CPU 后端性能极差）",[43,4006,4007],{},"❌ 多模态 \u002F 语音模型生产部署（支持不稳定，看具体版本）",[20,4009,381],{"id":380},[25,4011,4012,4015],{},[29,4013,4014],{},"Q: vLLM 和 Ollama 怎么选？","\nA: Ollama 是 Daemon + CLI，单用户原型极简；vLLM 是推理服务器，多用户并发吞吐高 16-20 倍。个人用 Ollama，对外提供服务用 vLLM。",[25,4017,4018,4021],{},[29,4019,4020],{},"Q: 单卡能跑 70B 吗？","\nA: 可以。用 AWQ\u002FGPTQ 4-bit 量化，70B 约需 35-40GB 显存，A100 80GB 或 2×A100 40GB 张量并行。FP16 则需 140GB（2×A100 80GB）。",[25,4023,4024,4027],{},[29,4025,4026],{},"Q: 和 TensorRT-LLM 比谁快？","\nA: TensorRT-LLM 在极致优化下略快（5-15%），但需要编译 engine、调试周期长、模型适配少。vLLM 灵活性和生态好得多，综合性价比更高。",[25,4029,4030,4033],{},[29,4031,4032],{},"Q: 支持 AMD GPU 吗？","\nA: 部分支持。0.5+ 起 ROCm 后端可用，但稳定性、性能、生态都远不如 NVIDIA CUDA。生产环境仍建议 NVIDIA。",[20,4035,402],{"id":402},[25,4037,4038,410,4040,410,4042],{},[406,4039,409],{"href":408},[406,4041,10],{"href":1649},[406,4043,418],{"href":417},[20,4045,421],{"id":421},[95,4047,4048],{},[25,4049,426],{},[40,4051,4052,4058],{},[43,4053,4054],{},[406,4055,436],{"href":4056,"rel":4057},"https:\u002F\u002Fvllm.ai",[435],[43,4059,4060],{},[406,4061,443],{"href":4062,"rel":4063},"https:\u002F\u002Fgithub.com\u002Fvllm-project\u002Fvllm",[435],{"title":445,"searchDepth":446,"depth":446,"links":4065},[4066,4067,4068,4069,4070,4071,4072,4073,4074,4075,4076],{"id":22,"depth":449,"text":23},{"id":38,"depth":449,"text":38},{"id":93,"depth":449,"text":93},{"id":138,"depth":449,"text":139},{"id":189,"depth":449,"text":189},{"id":217,"depth":449,"text":217},{"id":318,"depth":449,"text":318},{"id":353,"depth":449,"text":354},{"id":380,"depth":449,"text":381},{"id":402,"depth":449,"text":402},{"id":421,"depth":449,"text":421},"\u002Fimg\u002Ftools\u002Fvllm.webp","vLLM 真实评测：开源高吞吐 LLM 推理引擎（Apache 2.0 协议），核心创新 PagedAttention + 连续批处理，显著提升 GPU 利用率和推理吞吐。适合需要生产级高并发 LLM 服务部署、追求最大吞吐量的工程团队。",[466],{},"\u002Ftools\u002Fcoding\u002Flocal\u002Fvllm",[475],"Free \u002F 开源（Apache 2.0）",{"power":480,"ux":446,"price":480,"cn_support":449,"stability":479},{"title":414,"description":4078},"vLLM - 高吞吐 LLM 推理引擎评测与部署 | AIHO","coding\u002Flocal\u002Fvllm",[4089,4090],{"title":436,"url":4056},{"title":443,"url":4062},"tools\u002Fcoding\u002Flocal\u002Fvllm","高吞吐 LLM 推理引擎，PagedAttention 连续批处理",[460,4094,4095,4096,4097],"inference-engine","production","paged-attention","gpu","生产级高并发 LLM 服务的首选推理引擎，PagedAttention + 连续批处理把单卡吞吐拉到极致；单用户原型和 Mac 本地玩用 Ollama \u002F LM Studio 更省心。","njZ35H6xUwwjCk237XJvzfgRSW0kRrEEdoBySqYygng",[],[],[4103,4431,4781],{"id":1293,"title":409,"alternatives":4104,"api_compatible":15,"body":4105,"category":460,"chinese_friendly":446,"cover":1687,"description":1688,"domestic":463,"extension":464,"faq":15,"free":463,"github":1672,"languages":4422,"lastVerified":467,"meta":4423,"models":15,"navigation":469,"notSuitable":15,"opensource":469,"path":1691,"pillar":471,"platforms":4424,"priceTable":15,"pricing":1693,"published":477,"relatedPlaybooks":15,"relatedReviews":15,"score":4425,"self_host":463,"seo":4426,"seoTitle":1696,"slug":14,"sources":4427,"stem":1700,"suitable":15,"tagline":1701,"tags":4430,"updated":467,"verdict":1704,"website":1666,"__hash__":1705},[13,955,12],{"type":17,"value":4106,"toc":4409},[4107,4109,4113,4115,4117,4151,4153,4157,4179,4181,4183,4189,4201,4205,4217,4219,4233,4235,4325,4327,4349,4351,4367,4369,4373,4377,4381,4383,4391,4393,4397],[20,4108,23],{"id":22},[25,4110,1301,4111,1305],{},[29,4112,1304],{},[25,4114,1308],{},[20,4116,38],{"id":38},[40,4118,4119,4123,4127,4131,4135,4139,4143,4147],{},[43,4120,4121,1318],{},[29,4122,1317],{},[43,4124,4125,1323],{},[29,4126,53],{},[43,4128,4129,1329],{},[29,4130,1328],{},[43,4132,4133,1335],{},[29,4134,1334],{},[43,4136,4137,1341],{},[29,4138,1340],{},[43,4140,4141,1347],{},[29,4142,1346],{},[43,4144,4145,1353],{},[29,4146,1352],{},[43,4148,4149,1359],{},[29,4150,1358],{},[20,4152,93],{"id":93},[95,4154,4155],{},[25,4156,99],{},[101,4158,4159,4169],{},[104,4160,4161],{},[107,4162,4163,4165,4167],{},[110,4164,112],{},[110,4166,93],{},[110,4168,117],{},[119,4170,4171],{},[107,4172,4173,4175,4177],{},[124,4174,126],{},[124,4176,129],{},[124,4178,1388],{},[25,4180,1391],{},[20,4182,139],{"id":138},[95,4184,4185],{},[25,4186,144,4187],{},[29,4188,147],{},[40,4190,4191,4193,4195,4197,4199],{},[43,4192,1404],{},[43,4194,1407],{},[43,4196,1410],{},[43,4198,1413],{},[43,4200,1416],{},[25,4202,4203],{},[29,4204,169],{},[40,4206,4207,4209,4211,4213,4215],{},[43,4208,1425],{},[43,4210,1428],{},[43,4212,1431],{},[43,4214,1434],{},[43,4216,1437],{},[20,4218,189],{"id":189},[191,4220,4221,4223,4225,4227,4229,4231],{},[43,4222,1444],{},[43,4224,1447],{},[43,4226,1450],{},[43,4228,1453],{},[43,4230,1456],{},[43,4232,1459],{},[20,4234,217],{"id":217},[101,4236,4237,4251],{},[104,4238,4239],{},[107,4240,4241,4243,4245,4247,4249],{},[110,4242,226],{},[110,4244,409],{},[110,4246,234],{},[110,4248,231],{},[110,4250,498],{},[119,4252,4253,4265,4277,4289,4301,4313],{},[107,4254,4255,4257,4259,4261,4263],{},[124,4256,1484],{},[124,4258,1487],{},[124,4260,1490],{},[124,4262,1493],{},[124,4264,1487],{},[107,4266,4267,4269,4271,4273,4275],{},[124,4268,1500],{},[124,4270,1490],{},[124,4272,1505],{},[124,4274,1505],{},[124,4276,1490],{},[107,4278,4279,4281,4283,4285,4287],{},[124,4280,1514],{},[124,4282,283],{},[124,4284,246],{},[124,4286,1505],{},[124,4288,246],{},[107,4290,4291,4293,4295,4297,4299],{},[124,4292,1527],{},[124,4294,246],{},[124,4296,1532],{},[124,4298,1532],{},[124,4300,246],{},[107,4302,4303,4305,4307,4309,4311],{},[124,4304,1541],{},[124,4306,246],{},[124,4308,246],{},[124,4310,1548],{},[124,4312,246],{},[107,4314,4315,4317,4319,4321,4323],{},[124,4316,765],{},[124,4318,1557],{},[124,4320,315],{},[124,4322,310],{},[124,4324,1564],{},[20,4326,318],{"id":318},[40,4328,4329,4333,4337,4341,4345],{},[43,4330,4331,1574],{},[29,4332,1573],{},[43,4334,4335,1580],{},[29,4336,1579],{},[43,4338,4339,1586],{},[29,4340,1585],{},[43,4342,4343,1592],{},[29,4344,1591],{},[43,4346,4347,1598],{},[29,4348,1597],{},[20,4350,354],{"id":353},[40,4352,4353,4355,4357,4359,4361,4363,4365],{},[43,4354,1605],{},[43,4356,1608],{},[43,4358,1611],{},[43,4360,1614],{},[43,4362,374],{},[43,4364,1619],{},[43,4366,1622],{},[20,4368,381],{"id":380},[25,4370,4371,1630],{},[29,4372,1629],{},[25,4374,4375,1636],{},[29,4376,1635],{},[25,4378,4379,1642],{},[29,4380,1641],{},[20,4382,402],{"id":402},[25,4384,4385,410,4387,410,4389],{},[406,4386,10],{"href":1649},[406,4388,414],{"href":413},[406,4390,418],{"href":417},[20,4392,421],{"id":421},[95,4394,4395],{},[25,4396,426],{},[40,4398,4399,4404],{},[43,4400,4401],{},[406,4402,436],{"href":1666,"rel":4403},[435],[43,4405,4406],{},[406,4407,443],{"href":1672,"rel":4408},[435],{"title":445,"searchDepth":446,"depth":446,"links":4410},[4411,4412,4413,4414,4415,4416,4417,4418,4419,4420,4421],{"id":22,"depth":449,"text":23},{"id":38,"depth":449,"text":38},{"id":93,"depth":449,"text":93},{"id":138,"depth":449,"text":139},{"id":189,"depth":449,"text":189},{"id":217,"depth":449,"text":217},{"id":318,"depth":449,"text":318},{"id":353,"depth":449,"text":354},{"id":380,"depth":449,"text":381},{"id":402,"depth":449,"text":402},{"id":421,"depth":449,"text":421},[466],{},[473,474,475],{"power":446,"ux":479,"price":480,"cn_support":446,"stability":446},{"title":409,"description":1688},[4428,4429],{"title":436,"url":1666},{"title":443,"url":1672},[460,967,491,492,1703],{"id":1707,"title":234,"alternatives":4432,"api_compatible":15,"body":4433,"category":460,"chinese_friendly":446,"cover":2133,"description":2134,"domestic":463,"extension":464,"faq":4762,"free":463,"github":15,"languages":4767,"lastVerified":15,"meta":4768,"models":15,"navigation":469,"notSuitable":15,"opensource":463,"path":866,"pillar":471,"platforms":4769,"priceTable":4770,"pricing":2159,"published":948,"relatedPlaybooks":4773,"relatedReviews":15,"score":4774,"self_host":469,"seo":4775,"seoTitle":2164,"slug":13,"sources":4776,"stem":2172,"suitable":15,"tagline":2173,"tags":4780,"updated":959,"verdict":2180,"website":2181,"__hash__":2182},[12,501,955,500],{"type":17,"value":4434,"toc":4750},[4435,4437,4441,4443,4445,4483,4485,4495,4499,4501,4505,4517,4521,4533,4535,4553,4555,4669,4671,4693,4695,4711,4713,4731,4733],[20,4436,23],{"id":22},[25,4438,1715,4439,1719],{},[212,4440,1718],{},[25,4442,1722],{},[20,4444,38],{"id":38},[40,4446,4447,4451,4455,4459,4465,4471,4475,4479],{},[43,4448,4449,1732],{},[29,4450,1731],{},[43,4452,4453,1738],{},[29,4454,1737],{},[43,4456,4457,1744],{},[29,4458,1743],{},[43,4460,4461,1750,4463,1754],{},[29,4462,1749],{},[212,4464,1753],{},[43,4466,4467,1750,4469,1763],{},[29,4468,1759],{},[212,4470,1762],{},[43,4472,4473,1769],{},[29,4474,1768],{},[43,4476,4477,1775],{},[29,4478,1774],{},[43,4480,4481,1781],{},[29,4482,1780],{},[20,4484,93],{"id":93},[40,4486,4487,4491],{},[43,4488,4489,1791],{},[29,4490,1790],{},[43,4492,4493,1797],{},[29,4494,1796],{},[95,4496,4497],{},[25,4498,1802],{},[20,4500,1806],{"id":1805},[25,4502,4503],{},[29,4504,147],{},[40,4506,4507,4509,4511,4513,4515],{},[43,4508,1815],{},[43,4510,1818],{},[43,4512,1821],{},[43,4514,1824],{},[43,4516,1827],{},[25,4518,4519],{},[29,4520,169],{},[40,4522,4523,4525,4527,4529,4531],{},[43,4524,1836],{},[43,4526,1839],{},[43,4528,1842],{},[43,4530,1845],{},[43,4532,1848],{},[20,4534,189],{"id":189},[191,4536,4537,4539,4541,4543,4545,4549],{},[43,4538,1855],{},[43,4540,1858],{},[43,4542,1861],{},[43,4544,1864],{},[43,4546,1867,4547,1871],{},[212,4548,1870],{},[43,4550,1874,4551],{},[212,4552,1762],{},[20,4554,217],{"id":217},[101,4556,4557,4571],{},[104,4558,4559],{},[107,4560,4561,4563,4565,4567,4569],{},[110,4562,226],{},[110,4564,234],{},[110,4566,231],{},[110,4568,669],{},[110,4570,1895],{},[119,4572,4573,4585,4597,4609,4621,4633,4645,4657],{},[107,4574,4575,4577,4579,4581,4583],{},[124,4576,676],{},[124,4578,1904],{},[124,4580,1907],{},[124,4582,1910],{},[124,4584,1913],{},[107,4586,4587,4589,4591,4593,4595],{},[124,4588,1918],{},[124,4590,1921],{},[124,4592,1924],{},[124,4594,743],{},[124,4596,1929],{},[107,4598,4599,4601,4603,4605,4607],{},[124,4600,1934],{},[124,4602,246],{},[124,4604,1532],{},[124,4606,1941],{},[124,4608,1532],{},[107,4610,4611,4613,4615,4617,4619],{},[124,4612,71],{},[124,4614,1950],{},[124,4616,1953],{},[124,4618,246],{},[124,4620,246],{},[107,4622,4623,4625,4627,4629,4631],{},[124,4624,1962],{},[124,4626,246],{},[124,4628,1967],{},[124,4630,789],{},[124,4632,789],{},[107,4634,4635,4637,4639,4641,4643],{},[124,4636,1976],{},[124,4638,714],{},[124,4640,714],{},[124,4642,246],{},[124,4644,1490],{},[107,4646,4647,4649,4651,4653,4655],{},[124,4648,307],{},[124,4650,773],{},[124,4652,310],{},[124,4654,310],{},[124,4656,310],{},[107,4658,4659,4661,4663,4665,4667],{},[124,4660,2001],{},[124,4662,2004],{},[124,4664,2007],{},[124,4666,1490],{},[124,4668,1487],{},[20,4670,318],{"id":318},[40,4672,4673,4677,4681,4685,4689],{},[43,4674,4675,2021],{},[29,4676,2020],{},[43,4678,4679,2027],{},[29,4680,2026],{},[43,4682,4683,2033],{},[29,4684,2032],{},[43,4686,4687,2039],{},[29,4688,2038],{},[43,4690,4691,2045],{},[29,4692,2044],{},[20,4694,354],{"id":353},[40,4696,4697,4699,4701,4703,4705,4707,4709],{},[43,4698,2052],{},[43,4700,2055],{},[43,4702,2058],{},[43,4704,2061],{},[43,4706,2064],{},[43,4708,2067],{},[43,4710,2070],{},[20,4712,402],{"id":402},[40,4714,4715,4719,4723,4727],{},[43,4716,4717],{},[406,4718,873],{"href":872},[43,4720,4721],{},[406,4722,2084],{"href":2083},[43,4724,4725],{},[406,4726,2089],{"href":935},[43,4728,4729],{},[406,4730,2095],{"href":2094},[20,4732,421],{"id":421},[191,4734,4735,4740,4745],{},[43,4736,2102,4737],{},[406,4738,2105],{"href":2105,"rel":4739},[435],[43,4741,2109,4742],{},[406,4743,2112],{"href":2112,"rel":4744},[435],[43,4746,2116,4747],{},[406,4748,2119],{"href":2119,"rel":4749},[435],{"title":445,"searchDepth":446,"depth":446,"links":4751},[4752,4753,4754,4755,4756,4757,4758,4759,4760,4761],{"id":22,"depth":449,"text":23},{"id":38,"depth":449,"text":38},{"id":93,"depth":449,"text":93},{"id":1805,"depth":449,"text":1806},{"id":189,"depth":449,"text":189},{"id":217,"depth":449,"text":217},{"id":318,"depth":449,"text":318},{"id":353,"depth":449,"text":354},{"id":402,"depth":449,"text":402},{"id":421,"depth":449,"text":421},[4763,4764,4765,4766],{"q":2137,"a":2138},{"q":2140,"a":2141},{"q":2143,"a":2144},{"q":2146,"a":2147},[466,933],{},[473,474,475],[4771,4772],{"plan":1790,"price":940,"features":2153,"notes":2154},{"plan":1796,"price":2156,"features":2157,"notes":2158},[950,2161],{"power":479,"ux":480,"price":480,"cn_support":446,"stability":479},{"title":234,"description":2134},[4777,4778,4779],{"name":2167,"url":2105,"accessed":959},{"name":2169,"url":2112,"accessed":959},{"name":2171,"url":2119,"accessed":959},[460,2175,492,2176,2177,2178,2179],{"id":2671,"title":231,"alternatives":4782,"api_compatible":15,"body":4783,"category":460,"chinese_friendly":446,"cover":3090,"description":3091,"domestic":463,"extension":464,"faq":5114,"free":463,"github":15,"languages":5119,"lastVerified":15,"meta":5120,"models":15,"navigation":469,"notSuitable":15,"opensource":469,"path":872,"pillar":471,"platforms":5121,"priceTable":5122,"pricing":3112,"published":948,"relatedPlaybooks":5124,"relatedReviews":15,"score":5125,"self_host":469,"seo":5126,"seoTitle":3116,"slug":12,"sources":5127,"stem":3123,"suitable":15,"tagline":3124,"tags":5131,"updated":959,"verdict":3130,"website":3131,"__hash__":3132},[13,501,955,500],{"type":17,"value":4784,"toc":5102},[4785,4787,4791,4793,4795,4843,4845,4847,4849,4853,4867,4871,4891,4893,4917,4919,5021,5023,5045,5047,5063,5065,5083,5085],[20,4786,23],{"id":22},[25,4788,2679,4789,2683],{},[212,4790,2682],{},[25,4792,2686],{},[20,4794,38],{"id":38},[40,4796,4797,4801,4807,4811,4817,4827,4831,4835,4839],{},[43,4798,4799,2696],{},[29,4800,2695],{},[43,4802,4803,1750,4805],{},[29,4804,2701],{},[212,4806,2704],{},[43,4808,4809,2710],{},[29,4810,2709],{},[43,4812,4813,1750,4815],{},[29,4814,71],{},[212,4816,2717],{},[43,4818,4819,1750,4821,2726,4823,2726,4825],{},[29,4820,2722],{},[212,4822,2725],{},[212,4824,2729],{},[212,4826,2732],{},[43,4828,4829,2738],{},[29,4830,2737],{},[43,4832,4833,2744],{},[29,4834,2743],{},[43,4836,4837,2750],{},[29,4838,2749],{},[43,4840,4841,2756],{},[29,4842,2755],{},[20,4844,93],{"id":93},[25,4846,2761],{},[20,4848,2765],{"id":2764},[25,4850,4851],{},[29,4852,147],{},[40,4854,4855,4859,4861,4863,4865],{},[43,4856,4857,2777],{},[212,4858,2776],{},[43,4860,2780],{},[43,4862,2783],{},[43,4864,2786],{},[43,4866,2789],{},[25,4868,4869],{},[29,4870,169],{},[40,4872,4873,4879,4883,4887,4889],{},[43,4874,2798,4875,2802,4877],{},[212,4876,2801],{},[212,4878,2805],{},[43,4880,2808,4881],{},[212,4882,2811],{},[43,4884,2814,4885,2818],{},[212,4886,2817],{},[43,4888,2821],{},[43,4890,2824],{},[20,4892,189],{"id":189},[191,4894,4895,4899,4903,4907,4911,4915],{},[43,4896,4897,2834],{},[212,4898,2833],{},[43,4900,4901,2840],{},[212,4902,2839],{},[43,4904,4905,2845],{},[212,4906,2776],{},[43,4908,2848,4909],{},[212,4910,2851],{},[43,4912,2854,4913],{},[212,4914,2857],{},[43,4916,2860],{},[20,4918,217],{"id":217},[101,4920,4921,4935],{},[104,4922,4923],{},[107,4924,4925,4927,4929,4931,4933],{},[110,4926,226],{},[110,4928,231],{},[110,4930,234],{},[110,4932,414],{},[110,4934,1895],{},[119,4936,4937,4949,4961,4973,4985,4997,5009],{},[107,4938,4939,4941,4943,4945,4947],{},[124,4940,676],{},[124,4942,2887],{},[124,4944,2890],{},[124,4946,2893],{},[124,4948,2896],{},[107,4950,4951,4953,4955,4957,4959],{},[124,4952,189],{},[124,4954,2004],{},[124,4956,2004],{},[124,4958,1490],{},[124,4960,1487],{},[107,4962,4963,4965,4967,4969,4971],{},[124,4964,1918],{},[124,4966,2701],{},[124,4968,2917],{},[124,4970,743],{},[124,4972,743],{},[107,4974,4975,4977,4979,4981,4983],{},[124,4976,2926],{},[124,4978,1953],{},[124,4980,1950],{},[124,4982,246],{},[124,4984,246],{},[107,4986,4987,4989,4991,4993,4995],{},[124,4988,2939],{},[124,4990,2942],{},[124,4992,2945],{},[124,4994,2948],{},[124,4996,1490],{},[107,4998,4999,5001,5003,5005,5007],{},[124,5000,1962],{},[124,5002,1967],{},[124,5004,246],{},[124,5006,1941],{},[124,5008,789],{},[107,5010,5011,5013,5015,5017,5019],{},[124,5012,307],{},[124,5014,310],{},[124,5016,315],{},[124,5018,2973],{},[124,5020,310],{},[20,5022,318],{"id":318},[40,5024,5025,5029,5033,5037,5041],{},[43,5026,5027,2985],{},[29,5028,2984],{},[43,5030,5031,2991],{},[29,5032,2990],{},[43,5034,5035,2997],{},[29,5036,2996],{},[43,5038,5039,3003],{},[29,5040,3002],{},[43,5042,5043,3009],{},[29,5044,3008],{},[20,5046,354],{"id":353},[40,5048,5049,5051,5053,5055,5057,5059,5061],{},[43,5050,3016],{},[43,5052,3019],{},[43,5054,3022],{},[43,5056,3025],{},[43,5058,3028],{},[43,5060,3031],{},[43,5062,3034],{},[20,5064,402],{"id":402},[40,5066,5067,5071,5075,5079],{},[43,5068,5069],{},[406,5070,867],{"href":866},[43,5072,5073],{},[406,5074,2084],{"href":2083},[43,5076,5077],{},[406,5078,2089],{"href":935},[43,5080,5081],{},[406,5082,879],{"href":878},[20,5084,421],{"id":421},[191,5086,5087,5092,5097],{},[43,5088,3061,5089],{},[406,5090,3064],{"href":3064,"rel":5091},[435],[43,5093,3068,5094],{},[406,5095,3071],{"href":3071,"rel":5096},[435],[43,5098,2116,5099],{},[406,5100,2119],{"href":2119,"rel":5101},[435],{"title":445,"searchDepth":446,"depth":446,"links":5103},[5104,5105,5106,5107,5108,5109,5110,5111,5112,5113],{"id":22,"depth":449,"text":23},{"id":38,"depth":449,"text":38},{"id":93,"depth":449,"text":93},{"id":2764,"depth":449,"text":2765},{"id":189,"depth":449,"text":189},{"id":217,"depth":449,"text":217},{"id":318,"depth":449,"text":318},{"id":353,"depth":449,"text":354},{"id":402,"depth":449,"text":402},{"id":421,"depth":449,"text":421},[5115,5116,5117,5118],{"q":3094,"a":3095},{"q":3097,"a":3098},{"q":3100,"a":3101},{"q":3103,"a":3104},[466],{},[473,474,475,2641],[5123],{"plan":126,"price":940,"features":3110,"notes":3111},[950,2161],{"power":479,"ux":479,"price":480,"cn_support":446,"stability":480},{"title":231,"description":3091},[5128,5129,5130],{"name":3119,"url":3064,"accessed":959},{"name":3121,"url":3071,"accessed":959},{"name":2171,"url":2119,"accessed":959},[460,3126,3127,3128,3129,492,2176,2179,971],1785428438823]