[{"data":1,"prerenderedAt":2122},["ShallowReactive",2],{"header-counts":3,"footer-counts":6,"alt-main-msty":8,"alt-list-msty":657},{"tools":4,"reviews":5},98,27,{"tools":4,"reviews":5,"playbooks":7,"news":5},23,{"id":9,"title":10,"alternatives":11,"api_compatible":15,"body":16,"category":579,"chinese_friendly":566,"cover":580,"description":581,"domestic":582,"extension":583,"faq":584,"free":597,"github":15,"languages":598,"lastVerified":601,"meta":602,"models":15,"navigation":597,"notSuitable":15,"opensource":582,"path":603,"pillar":604,"platforms":605,"priceTable":610,"pricing":626,"published":627,"relatedPlaybooks":628,"relatedReviews":15,"score":630,"self_host":597,"seo":633,"seoTitle":634,"slug":635,"sources":636,"stem":646,"suitable":15,"tagline":647,"tags":648,"updated":639,"verdict":654,"website":655,"__hash__":656},"tools\u002Ftools\u002Fagent\u002Fgeneral\u002Fmsty.md","Msty",[12,13,14],"coding\u002Flocal\u002Follama","coding\u002Flocal\u002Flm-studio","coding\u002Flocal\u002Fopen-webui",null,{"type":17,"value":18,"toc":564},"minimark",[19,24,28,31,34,111,114,140,146,150,155,181,186,212,215,239,242,417,420,476,480,506,509,530,533],[20,21,23],"h2",{"id":22},"tldr","TL;DR",[25,26,27],"p",{},"Msty 是 privacy-first 桌面 AI 工作站，macOS \u002F Windows \u002F Linux 原生 app + 浏览器，独立开发者出品。差异点：内置 MLX (Apple) \u002F llama.cpp \u002F Ollama 三种本地推理引擎，无需 CLI + Hosted Models（OpenAI \u002F Anthropic \u002F Gemini）一窗体并存 + Split Chats 同问题多模型并行 + Knowledge Stack（per-conversation RAG）+ Prompt \u002F Persona \u002F Skills 三个 Studios + Agent Mode 多步执行。Free 本地无限 \u002F Aurum $149·年 \u002F Lifetime $349 \u002F Enterprise $300\u002Fuser·年。",[25,29,30],{},"适合：隐私敏感 + 不愿数据上云；Mac mini \u002F Linux box 当私人 AI 服务器；想避开 Ollama CLI 的非工程师；多模型对比决策场景。不适合：硬件不行（7B+ 本地跑不动）；要 BYO API + 极简（用 Typing Mind）；要 mobile（无 iOS \u002F Android 移动 app）；团队协作（更适合 Claude Team）。",[20,32,33],{"id":33},"核心能力",[35,36,37,45,51,57,63,69,75,81,87,93,99,105],"ul",{},[38,39,40,44],"li",{},[41,42,43],"strong",{},"三引擎本地推理","：MLX（Apple）\u002F llama.cpp \u002F Ollama 开箱即用",[38,46,47,50],{},[41,48,49],{},"Hosted Models","：OpenAI \u002F Anthropic \u002F Gemini 一窗体接入",[38,52,53,56],{},[41,54,55],{},"Split Chats","：同问题同时跑多模型 + side-by-side 对比",[38,58,59,62],{},[41,60,61],{},"Knowledge Stack","：per-conversation RAG，文档 \u002F URL \u002F Obsidian vault \u002F YouTube transcript",[38,64,65,68],{},[41,66,67],{},"Prompt Studio","：变量 + 模板 + 测试",[38,70,71,74],{},[41,72,73],{},"Persona Studio","：自定义角色 + 工具 + context",[38,76,77,80],{},[41,78,79],{},"Skills Studio","：可复用能力包",[38,82,83,86],{},[41,84,85],{},"Agent Mode","：多步执行 + 工具调用",[38,88,89,92],{},[41,90,91],{},"Flowchart 对话","：分支可视化",[38,94,95,98],{},[41,96,97],{},"Real-time 数据","：实时 web fetch",[38,100,101,104],{},[41,102,103],{},"Offline-first","：零账号 \u002F 零 telemetry \u002F 零云依赖（Free）",[38,106,107,110],{},[41,108,109],{},"Cloud Sync","（Aurum）：跨设备同步对话",[20,112,113],{"id":113},"价格",[35,115,116,122,128,134],{},[38,117,118,121],{},[41,119,120],{},"Free","：$0 永久；本地全功能 + 云 API 接入 + Split Chats + Knowledge Stack",[38,123,124,127],{},[41,125,126],{},"Aurum","：$149\u002F年；cloud sync + 高级 Knowledge + Studio Desktop alpha + 优先支持",[38,129,130,133],{},[41,131,132],{},"Lifetime","：$349 一次；Aurum 全部功能 + 终身更新",[38,135,136,139],{},[41,137,138],{},"Enterprise","：$300\u002Fuser·年；SSO + 私有部署 + 团队管理",[141,142,143],"blockquote",{},[25,144,145],{},"Lifetime 在 2 年用回本，重度用户首选。Free 已经足够个人 90% 场景。",[20,147,149],{"id":148},"实测macos-mac-mini-私人-ai-服务器","实测（macOS + Mac mini 私人 AI 服务器）",[25,151,152],{},[41,153,154],{},"亮点：",[35,156,157,160,163,166,169,172,175,178],{},[38,158,159],{},"装完立刻能用，无需 Ollama \u002F llama.cpp \u002F MLX 任何 CLI",[38,161,162],{},"Split Chats 对比 GPT-5 + Claude + Qwen2.5 + Llama 3.3 一目了然",[38,164,165],{},"Knowledge Stack 的 per-conversation 设计完美：每项目独立 RAG context",[38,167,168],{},"Prompt \u002F Persona \u002F Skills 三 Studio 解决重复 prompt 痛点",[38,170,171],{},"MLX 在 M1\u002FM2\u002FM3\u002FM4 上跑得快",[38,173,174],{},"中文 Qwen2.5 \u002F DeepSeek 走本地路径无外网依赖",[38,176,177],{},"Free 永久免费 + 本地无限 = 真正 zero-cost 路径",[38,179,180],{},"Lifetime $349 比 ChatGPT Plus 18 个月便宜",[25,182,183],{},[41,184,185],{},"踩坑：",[35,187,188,191,194,197,200,203,206,209],{},[38,189,190],{},"本地性能受硬件限制：M1 Air 跑 7B 慢，M3 Max \u002F Mac Studio \u002F 高端 GPU 更适合",[38,192,193],{},"插件 \u002F Skill 生态比 Typing Mind \u002F LM Studio 弱",[38,195,196],{},"桌面 only，无 iOS \u002F Android",[38,198,199],{},"llama.cpp 更新滞后官方上游 1-2 版本",[38,201,202],{},"Knowledge Stack 大文档（>100MB）切分偶尔失败",[38,204,205],{},"Agent Mode 仍在打磨，复杂任务稳定性不如 Claude",[38,207,208],{},"Studio Desktop（Aurum alpha）测试中，bug 偶发",[38,210,211],{},"中文 UI 不完整，部分功能仍是英文",[20,213,214],{"id":214},"上手",[216,217,218,221,224,227,230,233,236],"ol",{},[38,219,220],{},"msty.ai → 下载 macOS \u002F Windows \u002F Linux → 安装",[38,222,223],{},"Settings → Model Providers → Ollama 连本地（或直接装 Msty 自带 llama.cpp \u002F MLX）",[38,225,226],{},"装 1-2 个本地模型：Qwen2.5 7B（中文）+ Llama 3.3 8B（英文）",[38,228,229],{},"加云模型：OpenAI Key + Anthropic Key",[38,231,232],{},"新建 chat → 试 Split Chats：+ 第二个模型 → 同问题并行",[38,234,235],{},"Knowledge Stack → 上传项目文档 → 附加到 conversation",[38,237,238],{},"满意后 Free 用着，重度需要同步上 Lifetime $349",[20,240,241],{"id":241},"对比",[243,244,245,266],"table",{},[246,247,248],"thead",{},[249,250,251,255,257,260,263],"tr",{},[252,253,254],"th",{},"维度",[252,256,10],{},[252,258,259],{},"Ollama",[252,261,262],{},"LM Studio",[252,264,265],{},"Jan",[267,268,269,286,301,316,330,343,357,372,386,400],"tbody",{},[249,270,271,275,278,281,284],{},[272,273,274],"td",{},"GUI",[272,276,277],{},"✅ 颜值高",[272,279,280],{},"❌ CLI",[272,282,283],{},"✅",[272,285,283],{},[249,287,288,291,294,296,299],{},[272,289,290],{},"本地引擎",[272,292,293],{},"MLX\u002Fllama.cpp\u002FOllama",[272,295,259],{},[272,297,298],{},"llama.cpp",[272,300,298],{},[249,302,303,306,308,311,313],{},[272,304,305],{},"云模型",[272,307,283],{},[272,309,310],{},"❌",[272,312,310],{},[272,314,315],{},"部分",[249,317,318,320,323,325,327],{},[272,319,55],{},[272,321,322],{},"✅ 旗舰",[272,324,310],{},[272,326,315],{},[272,328,329],{},"–",[249,331,332,334,337,339,341],{},[272,333,61],{},[272,335,336],{},"✅ per-conv",[272,338,310],{},[272,340,310],{},[272,342,315],{},[249,344,345,348,351,353,355],{},[272,346,347],{},"Studios",[272,349,350],{},"✅ Prompt\u002FPersona\u002FSkills",[272,352,329],{},[272,354,329],{},[272,356,329],{},[249,358,359,362,364,367,369],{},[272,360,361],{},"开源",[272,363,310],{},[272,365,366],{},"✅ MIT",[272,368,310],{},[272,370,371],{},"✅ Apache 2.0",[249,373,374,377,380,382,384],{},[272,375,376],{},"起价",[272,378,379],{},"$0",[272,381,379],{},[272,383,379],{},[272,385,379],{},[249,387,388,391,394,396,398],{},[272,389,390],{},"终身",[272,392,393],{},"$349",[272,395,329],{},[272,397,329],{},[272,399,329],{},[249,401,402,405,408,411,414],{},[272,403,404],{},"适合",[272,406,407],{},"桌面颜值 + 多模型",[272,409,410],{},"CLI \u002F Server",[272,412,413],{},"模型市场",[272,415,416],{},"严格开源",[20,418,419],{"id":419},"避坑",[35,421,422,428,434,440,446,452,458,464,470],{},[38,423,424,427],{},[41,425,426],{},"硬件评估","：M1 Air 8GB 只跑 3B-7B，M3 Pro \u002F Max 跑 13B-30B 流畅",[38,429,430,433],{},[41,431,432],{},"Ollama 已装就连","：避免重复下模型，连本地 Ollama 复用 model library",[38,435,436,439],{},[41,437,438],{},"Knowledge Stack 文档","：单文档 \u003C50MB 最稳，大文件先切分",[38,441,442,445],{},[41,443,444],{},"Persona vs Skill","：Persona 是角色（完整 system + 模型）；Skill 是能力包；不要混用",[38,447,448,451],{},[41,449,450],{},"Split Chats 三个模型够","：4 个起每问 token 烧得快",[38,453,454,457],{},[41,455,456],{},"Aurum cloud sync 谨慎","：隐私敏感场景仍用 Free 本地",[38,459,460,463],{},[41,461,462],{},"Studio Desktop alpha","：稳定性不如 main 版本，重要工作不要全押",[38,465,466,469],{},[41,467,468],{},"本地中文模型","：Qwen2.5 7B \u002F DeepSeek 7B 中文最优，Llama 3.3 8B 英文最优",[38,471,472,475],{},[41,473,474],{},"MCP server","：当前不如 Claude Desktop 强，要 MCP 重度场景考虑 Claude Desktop \u002F Crush",[20,477,479],{"id":478},"适合-不适合","适合 \u002F 不适合",[35,481,482,485,488,491,494,497,500,503],{},[38,483,484],{},"✅ 隐私敏感 + 数据零云",[38,486,487],{},"✅ Mac mini \u002F Linux box 私人 AI 服务器",[38,489,490],{},"✅ 不愿学 Ollama CLI 的非工程师",[38,492,493],{},"✅ 多模型对比决策场景",[38,495,496],{},"❌ 硬件不行（8GB RAM）跑不动 7B+",[38,498,499],{},"❌ 要 mobile 移动主力",[38,501,502],{},"❌ 团队协作 + 共享 workspace",[38,504,505],{},"❌ 要 MCP 工具栈深度（用 Claude Desktop \u002F Crush）",[20,507,508],{"id":508},"相关阅读",[35,510,511,518,524],{},[38,512,513],{},[514,515,517],"a",{"href":516},"\u002Ftools\u002Fcoding\u002Flocal\u002Follama","Ollama 评测",[38,519,520],{},[514,521,523],{"href":522},"\u002Ftools\u002Fcoding\u002Flocal\u002Flm-studio","LM Studio 评测",[38,525,526],{},[514,527,529],{"href":528},"\u002Ftools\u002Fcoding\u002Flocal\u002Fopen-webui","Open WebUI 评测",[20,531,532],{"id":532},"来源",[216,534,535,543,550,557],{},[38,536,537,538],{},"Msty 官网 + Features（MLX \u002F llama.cpp \u002F Ollama \u002F Studios）",[514,539,540],{"href":540,"rel":541},"https:\u002F\u002Fmsty.ai\u002Fstudio\u002Ffeatures",[542],"nofollow",[38,544,545,546],{},"AI Chat Daily — Msty Review 2026（4.3\u002F5 评分 + Lifetime）",[514,547,548],{"href":548,"rel":549},"https:\u002F\u002Fwww.aichatdaily.com\u002Ftools\u002Fmsty",[542],[38,551,552,553],{},"ML Journey — Msty Multi-model Comparison Guide ",[514,554,555],{"href":555,"rel":556},"https:\u002F\u002Fmljourney.com\u002Fmsty-the-local-llm-app-that-lets-you-compare-models-side-by-side",[542],[38,558,559,560],{},"AISO Tools — Msty Pricing 2026 ",[514,561,562],{"href":562,"rel":563},"https:\u002F\u002Faisotools.com\u002Fpricing\u002Fmsty",[542],{"title":565,"searchDepth":566,"depth":566,"links":567},"",3,[568,570,571,572,573,574,575,576,577,578],{"id":22,"depth":569,"text":23},2,{"id":33,"depth":569,"text":33},{"id":113,"depth":569,"text":113},{"id":148,"depth":569,"text":149},{"id":214,"depth":569,"text":214},{"id":241,"depth":569,"text":241},{"id":419,"depth":569,"text":419},{"id":478,"depth":569,"text":479},{"id":508,"depth":569,"text":508},{"id":532,"depth":569,"text":532},"general","\u002Fimg\u002Ftools\u002Fmsty.webp","Msty 真实评测：privacy-first 桌面 AI 工作站，macOS \u002F Windows \u002F Linux 原生 app + 浏览器。差异点：内置 MLX (Apple) \u002F llama.cpp \u002F Ollama 三种本地推理引擎 + Hosted Models（OpenAI \u002F Anthropic \u002F Gemini）一窗体管理 + Split Chats 多模型同时跑同问题 + Knowledge Stack（per-conversation RAG，区别于 AnythingLLM workspace）+ Prompt \u002F Persona \u002F Skills 三个 Studios + Agent Mode 多步执行。Free 本地全功能 \u002F Aurum $5-149 \u002F Lifetime $349 \u002F Enterprise $300\u002Fuser·年。",false,"md",[585,588,591,594],{"q":586,"a":587},"和 Ollama \u002F LM Studio \u002F Jan \u002F AnythingLLM 怎么选？","Msty 强在『多模型 split chat 对比 + Knowledge Stack 灵活 per-conversation + UI 颜值』。Ollama 是 CLI + server，无 GUI 适合开发者。LM Studio 强在『模型市场 + 性能 profiling』。Jan 开源 + Apache 2.0 协议自由度高。AnythingLLM 强在 workspace + RAG agent。要桌面颜值 + 多模型对比 + 简单 RAG → Msty；要 CLI \u002F server → Ollama；要模型市场 → LM Studio；要严格开源 → Jan。",{"q":589,"a":590},"Knowledge Stack 怎么用？","上传文档 \u002F URL \u002F 文本到 Knowledge collection，per-conversation 附加。和 AnythingLLM 的 workspace 区别：Msty 是 conversation 级，每对话独立 context，不会跨对话泄露。多项目并行场景非常顺。Aurum 解锁更大 \u002F 更高级 Knowledge。",{"q":592,"a":593},"Split Chats 真的实用吗？","对，多模型决策场景非常有用：决定哪个模型适合任务（同问题看 GPT-5 \u002F Claude \u002F Llama 3 输出）；本地 vs 云模型质量评估；事实问题模型分歧检测（多个模型给同样答案 = 更可信）。日常使用确实降低选错模型成本。",{"q":595,"a":596},"中国大陆能用吗？","本地模式完全离线可用（Ollama \u002F llama.cpp \u002F MLX 本地模型 + Qwen \u002F DeepSeek 中文模型）。云模式接 OpenAI \u002F Anthropic 需要海外网络 + 卡。Aurum 订阅需海外支付。最佳路径：本地 Free + Ollama + Qwen2.5\u002FDeepSeek 中文，零订阅 + 零外网依赖。",true,[599,600],"en","multi","2026-08-02",{},"\u002Ftools\u002Fagent\u002Fgeneral\u002Fmsty","agent",[606,607,608,609],"macos","windows","linux","web",[611,614,618,622],{"plan":120,"price":379,"features":612,"notes":613},"本地模型 + 云 API 接入 + Split Chats + Knowledge Stack（无限）+ 无账号","永久免费",{"plan":126,"price":615,"features":616,"notes":617},"$149\u002F年","Free 全部 + cloud sync + 高级 Knowledge + 优先支持 + Studio Desktop alpha","每用户",{"plan":132,"price":619,"features":620,"notes":621},"$349 一次","Aurum 终身授权","永久 + 所有更新",{"plan":138,"price":623,"features":624,"notes":625},"$300\u002Fuser·年","SSO + 团队管理 + 私有部署支持","需联系销售","Free 本地无限 \u002F Aurum $149·年（cloud sync）\u002F Lifetime $349 一次买断 \u002F Enterprise $300·user·年","2026-06-19",[629],"onboarding\u002Flocal-ai-workstation",{"power":631,"ux":632,"price":632,"cn_support":566,"stability":631},4,5,{"title":10,"description":581},"Msty - 本地多模型桌面 AI 评测与指南 | AIHO","agent\u002Fgeneral\u002Fmsty",[637,640,642,644],{"name":638,"url":540,"accessed":639},"Msty 官网 + Features","2026-06-24",{"name":641,"url":548,"accessed":639},"AI Chat Daily — Msty Review 2026 4.3\u002F5",{"name":643,"url":555,"accessed":639},"ML Journey — Msty Local LLM Comparison Guide",{"name":645,"url":562,"accessed":639},"AISO Tools — Msty Pricing 2026","tools\u002Fagent\u002Fgeneral\u002Fmsty","本地优先 + 多模型并行的桌面 AI——Split Chats \u002F Knowledge Stack \u002F Agent Mode 三件套",[649,650,651,652,653],"local-first","multi-model","privacy","knowledge-base","msty","Ollama \u002F LM Studio 的『精品桌面应用版』——拒绝 CLI + 拒绝云上传 + 多模型同窗对比的最佳选择。Mac mini \u002F Linux 私人 AI 服务器场景神器。要团队协作 + 云同步建议 Claude Team \u002F ChatGPT Team。","https:\u002F\u002Fmsty.ai","ZLfDTWh2UVeooVSc0WqNWW3UXpXxbDW2Z_X_7iE3yls",[658,1173,1636],{"id":659,"title":259,"alternatives":660,"api_compatible":663,"body":678,"category":1117,"chinese_friendly":566,"cover":1118,"description":1119,"domestic":582,"extension":583,"faq":1120,"free":597,"github":1133,"languages":1134,"lastVerified":601,"meta":1135,"models":15,"navigation":597,"notSuitable":15,"opensource":597,"path":516,"pillar":1136,"platforms":1137,"priceTable":1139,"pricing":1145,"published":627,"relatedPlaybooks":1146,"relatedReviews":15,"score":1149,"self_host":597,"seo":1150,"seoTitle":1151,"slug":12,"sources":1152,"stem":1159,"suitable":15,"tagline":1160,"tags":1161,"updated":639,"verdict":1170,"website":1171,"__hash__":1172},"tools\u002Ftools\u002Fcoding\u002Flocal\u002Follama.md",[13,14,661,662],"coding\u002Flocal\u002Fcherry-studio","coding\u002Flocal\u002Flobe-chat",[664,665,666,667,668,669,670,671,672,673,674,675,676,259,677],"OpenAI","Anthropic","Google","Grok","Mistral","Cohere","阿里通义","百度文心","腾讯混元","Moonshot Kimi","字节豆包","DeepSeek","智谱 GLM","Hugging Face",{"type":17,"value":679,"toc":1105},[680,682,690,693,695,765,767,770,774,778,798,802,833,835,869,871,997,999,1031,1033,1056,1058,1080,1082],[20,681,23],{"id":22},[25,683,684,685,689],{},"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，",[686,687,688],"code",{},"ollama pull"," 一键拉。",[25,691,692],{},"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）。",[20,694,33],{"id":33},[35,696,697,703,712,718,726,741,747,753,759],{},[38,698,699,702],{},[41,700,701],{},"后台 Daemon","：开机自启，应用调用零延迟",[38,704,705,708,709],{},[41,706,707],{},"CLI","：",[686,710,711],{},"ollama pull \u002F run \u002F list \u002F show \u002F create \u002F serve",[38,713,714,717],{},[41,715,716],{},"Modelfile","：类 Dockerfile 注册任意 GGUF，配 SYSTEM \u002F PARAMETER \u002F TEMPLATE",[38,719,720,708,723],{},[41,721,722],{},"OpenAI 兼容 API",[686,724,725],{},"http:\u002F\u002Flocalhost:11434\u002Fv1\u002Fchat\u002Fcompletions",[38,727,728,708,731,734,735,734,738],{},[41,729,730],{},"原生 API",[686,732,733],{},"\u002Fapi\u002Fchat","、",[686,736,737],{},"\u002Fapi\u002Fgenerate",[686,739,740],{},"\u002Fapi\u002Fembeddings",[38,742,743,746],{},[41,744,745],{},"模型库","：官方注册表内置 Llama \u002F Qwen \u002F DeepSeek \u002F Gemma \u002F Mistral \u002F GPT-OSS 等",[38,748,749,752],{},[41,750,751],{},"MLX 加速（Mac）","：0.19+ 起 M 系列自动用 MLX",[38,754,755,758],{},[41,756,757],{},"量化","：默认 Q4_K_M、支持 Q5 \u002F Q8 \u002F FP16",[38,760,761,764],{},[41,762,763],{},"跨平台","：Win \u002F Mac \u002F Linux 安装包，Docker 官方镜像",[20,766,113],{"id":113},[25,768,769],{},"完全免费、MIT 开源、商用免费。",[20,771,773],{"id":772},"实测m2-pro-qwen3-coder-7b-q4","实测（M2 Pro + Qwen3-Coder-7B Q4）",[25,775,776],{},[41,777,154],{},[35,779,780,786,789,792,795],{},[38,781,782,785],{},[686,783,784],{},"ollama run qwen3-coder:7b"," 一行起飞，3 秒进交互",[38,787,788],{},"REST API 配 Cursor \u002F Cline \u002F Continue 几乎全工具开箱即用",[38,790,791],{},"Modelfile 写自定义编码助手（low temperature + system prompt + 16K context）几分钟搞定",[38,793,794],{},"多模型并存，按需切换，内存占用合理",[38,796,797],{},"Mac M 系列 MLX 后比旧 GGUF 模式快显著",[25,799,800],{},[41,801,185],{},[35,803,804,814,820,827,830],{},[38,805,806,807,810,811],{},"默认 ",[686,808,809],{},"num_ctx"," 偏小（2048），跑长上下文要在 Modelfile 加 ",[686,812,813],{},"PARAMETER num_ctx 16384",[38,815,816,817],{},"模型默认走 0.0.0.0:11434 ↔ Docker 容器互访要 ",[686,818,819],{},"--add-host=host.docker.internal:host-gateway",[38,821,822,823,826],{},"国内 ",[686,824,825],{},"ollama.com\u002Flibrary"," 下载偶有慢，可手动 HF 下 GGUF + Modelfile 自建",[38,828,829],{},"多用户并发吞吐显著低于 vLLM",[38,831,832],{},"没有 GUI，模型浏览 \u002F 参数面板要走 LM Studio \u002F Open WebUI 配合",[20,834,214],{"id":214},[216,836,837,843,849,854,860,866],{},[38,838,839,842],{},[686,840,841],{},"curl -fsSL https:\u002F\u002Follama.ai\u002Finstall.sh | sh","（Mac \u002F Linux）；Windows winget",[38,844,845,848],{},[686,846,847],{},"ollama pull qwen3-coder:7b","（按需换模型）",[38,850,851,853],{},[686,852,784],{}," 直接聊",[38,855,856,857],{},"应用接入：baseURL = ",[686,858,859],{},"http:\u002F\u002Flocalhost:11434\u002Fv1",[38,861,862,863],{},"自定义：写 Modelfile → ",[686,864,865],{},"ollama create my-coder -f Modelfile",[38,867,868],{},"进阶：装 Open WebUI 做前端 \u002F 多人共享",[20,870,241],{"id":241},[243,872,873,888],{},[246,874,875],{},[249,876,877,879,881,883,886],{},[252,878,254],{},[252,880,259],{},[252,882,262],{},[252,884,885],{},"vLLM",[252,887,298],{},[267,889,890,907,922,937,952,968,982],{},[249,891,892,895,898,901,904],{},[272,893,894],{},"形态",[272,896,897],{},"CLI + Daemon",[272,899,900],{},"GUI + Headless",[272,902,903],{},"Python Server",[272,905,906],{},"C++ 二进制",[249,908,909,911,914,916,919],{},[272,910,214],{},[272,912,913],{},"极低",[272,915,913],{},[272,917,918],{},"中",[272,920,921],{},"高",[249,923,924,927,929,932,935],{},[272,925,926],{},"模型浏览",[272,928,707],{},[272,930,931],{},"✅ GUI",[272,933,934],{},"无",[272,936,934],{},[249,938,939,942,945,948,950],{},[272,940,941],{},"OpenAI 兼容",[272,943,944],{},"✅ :11434",[272,946,947],{},"✅ :1234",[272,949,283],{},[272,951,283],{},[249,953,954,957,960,963,966],{},[272,955,956],{},"多用户吞吐",[272,958,959],{},"弱（~40 tok\u002Fs）",[272,961,962],{},"中（50–90）",[272,964,965],{},"强（800–12500）",[272,967,918],{},[249,969,970,973,976,978,980],{},[272,971,972],{},"MLX (Mac)",[272,974,975],{},"✅ 0.19+",[272,977,283],{},[272,979,315],{},[272,981,329],{},[249,983,984,986,989,992,995],{},[272,985,361],{},[272,987,988],{},"MIT",[272,990,991],{},"闭源",[272,993,994],{},"Apache 2.0",[272,996,988],{},[20,998,419],{"id":419},[35,1000,1001,1007,1013,1019,1025],{},[38,1002,1003,1006],{},[41,1004,1005],{},"num_ctx 一定要设","：默认 2K 太小，跑代码 \u002F 长文档要 16K+",[38,1008,1009,1012],{},[41,1010,1011],{},"Modelfile 模板别漏 TEMPLATE","：错的 chat template 会让模型输出乱码 \u002F 不停",[38,1014,1015,1018],{},[41,1016,1017],{},"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×",[38,1020,1021,1024],{},[41,1022,1023],{},"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",[38,1026,1027,1030],{},[41,1028,1029],{},"Mac 让它自动用 MLX","：升 0.19+；不要手动强制 GGUF + Metal",[20,1032,479],{"id":478},[35,1034,1035,1038,1041,1044,1047,1050,1053],{},[38,1036,1037],{},"✅ 应用 \u002F IDE 接本地模型（Cursor \u002F Cline \u002F Continue）",[38,1039,1040],{},"✅ 个人 \u002F 评估 \u002F 脚本自动化",[38,1042,1043],{},"✅ Modelfile 自定义系统 prompt + 参数",[38,1045,1046],{},"✅ Mac M 系列 MLX 用户",[38,1048,1049],{},"❌ 多用户并发生产服务（用 vLLM）",[38,1051,1052],{},"❌ GUI 调参 \u002F 模型浏览（配 LM Studio \u002F Open WebUI）",[38,1054,1055],{},"❌ 极致单卡吞吐研究（直接 llama.cpp \u002F vLLM）",[20,1057,508],{"id":508},[35,1059,1060,1064,1068,1074],{},[38,1061,1062],{},[514,1063,523],{"href":522},[38,1065,1066],{},[514,1067,529],{"href":528},[38,1069,1070],{},[514,1071,1073],{"href":1072},"\u002Ftools\u002Fcoding\u002Flocal\u002Fcherry-studio","Cherry Studio 评测",[38,1075,1076],{},[514,1077,1079],{"href":1078},"\u002Fplaybook\u002Fonboarding\u002Frag-pipeline-build","RAG Pipeline 搭建 Playbook",[20,1081,532],{"id":532},[216,1083,1084,1091,1098],{},[38,1085,1086,1087],{},"Markaicode — Import GGUF Models into Ollama 2026（2026-05-15）",[514,1088,1089],{"href":1089,"rel":1090},"https:\u002F\u002Fmarkaicode.com\u002Fimport-gguf-models-ollama-guide",[542],[38,1092,1093,1094],{},"ComputingForGeeks — Ollama Models Cheat Sheet 2026 ",[514,1095,1096],{"href":1096,"rel":1097},"https:\u002F\u002Fcomputingforgeeks.com\u002Follama-models-cheat-sheet",[542],[38,1099,1100,1101],{},"Codersera — Ollama vs LM Studio vs vLLM vs llama.cpp vs MLX 2026 ",[514,1102,1103],{"href":1103,"rel":1104},"https:\u002F\u002Fcodersera.com\u002Fblog\u002Follama-vs-lm-studio-vs-vllm-vs-llama-cpp-vs-mlx-2026\u002F",[542],{"title":565,"searchDepth":566,"depth":566,"links":1106},[1107,1108,1109,1110,1111,1112,1113,1114,1115,1116],{"id":22,"depth":569,"text":23},{"id":33,"depth":569,"text":33},{"id":113,"depth":569,"text":113},{"id":772,"depth":569,"text":773},{"id":214,"depth":569,"text":214},{"id":241,"depth":569,"text":241},{"id":419,"depth":569,"text":419},{"id":478,"depth":569,"text":479},{"id":508,"depth":569,"text":508},{"id":532,"depth":569,"text":532},"local","\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。",[1121,1124,1127,1130],{"q":1122,"a":1123},"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":1125,"a":1126},"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":1128,"a":1129},"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":1131,"a":1132},"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。","https:\u002F\u002Fgithub.com\u002Follama\u002Follama",[599],{},"coding",[607,606,608,1138],"docker",[1140],{"plan":1141,"price":1142,"features":1143,"notes":1144},"开源版","免费","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）",[1147,1148],"onboarding\u002Frag-pipeline-build","onboarding\u002Fclaude-code-getting-started",{"power":631,"ux":631,"price":632,"cn_support":566,"stability":632},{"title":259,"description":1119},"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",[1153,1155,1157],{"name":1154,"url":1089,"accessed":639},"Markaicode — Import GGUF 2026",{"name":1156,"url":1096,"accessed":639},"ComputingForGeeks — Ollama Cheat Sheet 2026",{"name":1158,"url":1103,"accessed":639},"Codersera — Ollama vs LM Studio vs vLLM 2026","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",[1117,1162,1163,1164,1165,1166,1167,1168,1169],"daemon","cli","rest-api","modelfile","gguf","mlx","openai-compatible","open-source","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","yL3ZqN3rlWsImgvBFSYlFraTqj9ki9gSJlMbXVmruyg",{"id":1174,"title":262,"alternatives":1175,"api_compatible":15,"body":1176,"category":1117,"chinese_friendly":566,"cover":1589,"description":1590,"domestic":582,"extension":583,"faq":1591,"free":597,"github":15,"languages":1604,"lastVerified":601,"meta":1606,"models":15,"navigation":597,"notSuitable":15,"opensource":582,"path":522,"pillar":1136,"platforms":1607,"priceTable":1608,"pricing":1616,"published":627,"relatedPlaybooks":1617,"relatedReviews":15,"score":1618,"self_host":597,"seo":1619,"seoTitle":1620,"slug":13,"sources":1621,"stem":1627,"suitable":15,"tagline":1628,"tags":1629,"updated":639,"verdict":1633,"website":1634,"__hash__":1635},"tools\u002Ftools\u002Fcoding\u002Flocal\u002Flm-studio.md",[12,14,661,662],{"type":17,"value":1177,"toc":1577},[1178,1180,1187,1190,1192,1248,1250,1264,1269,1273,1277,1294,1298,1315,1317,1343,1345,1473,1475,1507,1509,1532,1534,1554,1556],[20,1179,23],{"id":22},[25,1181,1182,1183,1186],{},"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 + ",[686,1184,1185],{},"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。",[25,1188,1189],{},"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）。",[20,1191,33],{"id":33},[35,1193,1194,1200,1206,1212,1221,1230,1236,1242],{},[38,1195,1196,1199],{},[41,1197,1198],{},"模型浏览器","：内置 Hugging Face 检索，按 GGUF \u002F MLX \u002F 大小筛选、一键下载",[38,1201,1202,1205],{},[41,1203,1204],{},"聊天界面","：System Prompt \u002F temperature \u002F top-p \u002F context size 可视化调参",[38,1207,1208,1211],{},[41,1209,1210],{},"多模型并存 \u002F 切换","：同时加载多模型在不同会话中比较",[38,1213,1214,708,1217,1220],{},[41,1215,1216],{},"OpenAI 兼容 Local Server",[686,1218,1219],{},"http:\u002F\u002Flocalhost:1234\u002Fv1","，任何 SDK 即接即用",[38,1222,1223,708,1226,1229],{},[41,1224,1225],{},"Headless \u002F CLI",[686,1227,1228],{},"lms server start --port 1234","，无 GUI 可跑",[38,1231,1232,1235],{},[41,1233,1234],{},"PDF \u002F 文档对话","：内置基础 RAG，丢文件就能聊",[38,1237,1238,1241],{},[41,1239,1240],{},"MLX 原生支持（Mac）","：M1+ 上比 GGUF + Metal 快 30–50%",[38,1243,1244,1247],{},[41,1245,1246],{},"持续批处理","：Codersera 2026 测得 50–90 tok\u002Fs（消费级 GPU + 中等模型）",[20,1249,113],{"id":113},[35,1251,1252,1258],{},[38,1253,1254,1257],{},[41,1255,1256],{},"个人 \u002F 评估","：免费，全功能可用",[38,1259,1260,1263],{},[41,1261,1262],{},"商用","：邮件 \u002F 官网联系 LM Studio 团队",[141,1265,1266],{},[25,1267,1268],{},"模型本身免费（开源权重），LM Studio 不抽水任何 token 费用。",[20,1270,1272],{"id":1271},"实测mac-m2-pro-qwen3-coder-7b-gguf-q4_k_m","实测（Mac M2 Pro + Qwen3-Coder-7B GGUF Q4_K_M）",[25,1274,1275],{},[41,1276,154],{},[35,1278,1279,1282,1285,1288,1291],{},[38,1280,1281],{},"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",[38,1283,1284],{},"加载 7B Q4 模型 \u003C 3 秒，生成 ~75 tok\u002Fs",[38,1286,1287],{},"Local Server 开了 Cursor 直接接 baseURL → 本地代码补全零成本",[38,1289,1290],{},"MLX 版同模型 ~110 tok\u002Fs，差距显著",[38,1292,1293],{},"多窗口加载 2 个模型并排测，调 prompt 直观",[25,1295,1296],{},[41,1297,185],{},[35,1299,1300,1303,1306,1309,1312],{},[38,1301,1302],{},"模型库依赖 Hugging Face，国内访问要镜像 \u002F 代理",[38,1304,1305],{},"GPU 显存吃满后会自动 offload 到 CPU，无提示就慢下来",[38,1307,1308],{},"Headless 模式相对 Ollama 偏新，文档稍少",[38,1310,1311],{},"闭源应用（虽免费），不适合企业合规挂钩",[38,1313,1314],{},"中文 UI 可用但部分菜单仍英文",[20,1316,214],{"id":214},[216,1318,1319,1322,1325,1328,1331,1338],{},[38,1320,1321],{},"lmstudio.ai 下载（Mac \u002F Windows \u002F Linux）",[38,1323,1324],{},"打开 → Discover 标签 → 搜模型（如 qwen3-coder、deepseek-v3 GGUF\u002FMLX）→ Download",[38,1326,1327],{},"Chat 标签 → 选模型 → 调参聊天",[38,1329,1330],{},"Local Server 标签 → Start Server → 默认端口 1234",[38,1332,1333,1334,1337],{},"在你的应用里：",[686,1335,1336],{},"baseURL = \"http:\u002F\u002Flocalhost:1234\u002Fv1\"","，API Key 任意",[38,1339,1340,1341],{},"Headless：",[686,1342,1228],{},[20,1344,241],{"id":241},[243,1346,1347,1362],{},[246,1348,1349],{},[249,1350,1351,1353,1355,1357,1360],{},[252,1352,254],{},[252,1354,262],{},[252,1356,259],{},[252,1358,1359],{},"Open WebUI",[252,1361,298],{},[267,1363,1364,1380,1395,1408,1420,1432,1446,1459],{},[249,1365,1366,1368,1371,1374,1377],{},[272,1367,894],{},[272,1369,1370],{},"GUI + CLI",[272,1372,1373],{},"CLI Daemon",[272,1375,1376],{},"Docker UI",[272,1378,1379],{},"二进制",[249,1381,1382,1384,1387,1390,1392],{},[272,1383,926],{},[272,1385,1386],{},"✅ 内置",[272,1388,1389],{},"CLI pull",[272,1391,934],{},[272,1393,1394],{},"手动",[249,1396,1397,1400,1402,1404,1406],{},[272,1398,1399],{},"参数调优 GUI",[272,1401,283],{},[272,1403,310],{},[272,1405,315],{},[272,1407,310],{},[249,1409,1410,1412,1414,1416,1418],{},[272,1411,722],{},[272,1413,947],{},[272,1415,944],{},[272,1417,283],{},[272,1419,283],{},[249,1421,1422,1424,1426,1428,1430],{},[272,1423,972],{},[272,1425,283],{},[272,1427,975],{},[272,1429,329],{},[272,1431,329],{},[249,1433,1434,1437,1440,1442,1444],{},[272,1435,1436],{},"多用户并发",[272,1438,1439],{},"弱",[272,1441,1439],{},[272,1443,283],{},[272,1445,918],{},[249,1447,1448,1450,1453,1455,1457],{},[272,1449,361],{},[272,1451,1452],{},"闭源（免费）",[272,1454,988],{},[272,1456,988],{},[272,1458,988],{},[249,1460,1461,1464,1466,1469,1471],{},[272,1462,1463],{},"上手难度",[272,1465,913],{},[272,1467,1468],{},"低",[272,1470,918],{},[272,1472,921],{},[20,1474,419],{"id":419},[35,1476,1477,1483,1489,1495,1501],{},[38,1478,1479,1482],{},[41,1480,1481],{},"国内下模型走镜像","：HF 直连慢 \u002F 卡，配 HF_ENDPOINT=hf-mirror.com",[38,1484,1485,1488],{},[41,1486,1487],{},"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",[38,1490,1491,1494],{},[41,1492,1493],{},"MLX 优先（Mac M 系列）","：能下 MLX 版就别下 GGUF，速度差距明显",[38,1496,1497,1500],{},[41,1498,1499],{},"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",[38,1502,1503,1506],{},[41,1504,1505],{},"闭源合规要核","：企业内部使用前查 license；商用必须联系官方",[20,1508,479],{"id":478},[35,1510,1511,1514,1517,1520,1523,1526,1529],{},[38,1512,1513],{},"✅ 本地 LLM 入门 \u002F 评估",[38,1515,1516],{},"✅ Mac M 系列用户",[38,1518,1519],{},"✅ 想给 Cursor \u002F Cline 接本地 OpenAI 兼容 endpoint",[38,1521,1522],{},"✅ 需要 GUI 调参 \u002F 模型比较",[38,1524,1525],{},"❌ 多用户并发生产服务",[38,1527,1528],{},"❌ 嵌入式 \u002F 边缘设备",[38,1530,1531],{},"❌ 强合规 \u002F 必须开源审计",[20,1533,508],{"id":508},[35,1535,1536,1540,1544,1548],{},[38,1537,1538],{},[514,1539,517],{"href":516},[38,1541,1542],{},[514,1543,529],{"href":528},[38,1545,1546],{},[514,1547,1073],{"href":1072},[38,1549,1550],{},[514,1551,1553],{"href":1552},"\u002Fplaybook\u002Fonboarding\u002Fclaude-code-getting-started","Claude Code 上手 Playbook",[20,1555,532],{"id":532},[216,1557,1558,1565,1572],{},[38,1559,1560,1561],{},"LM Studio 官网 ",[514,1562,1563],{"href":1563,"rel":1564},"https:\u002F\u002Flmstudio.ai\u002F",[542],[38,1566,1567,1568],{},"Codersera — LM Studio Complete Guide 2026 ",[514,1569,1570],{"href":1570,"rel":1571},"https:\u002F\u002Fcodersera.com\u002Fblog\u002Flm-studio-complete-guide-2026\u002F",[542],[38,1573,1100,1574],{},[514,1575,1103],{"href":1103,"rel":1576},[542],{"title":565,"searchDepth":566,"depth":566,"links":1578},[1579,1580,1581,1582,1583,1584,1585,1586,1587,1588],{"id":22,"depth":569,"text":23},{"id":33,"depth":569,"text":33},{"id":113,"depth":569,"text":113},{"id":1271,"depth":569,"text":1272},{"id":214,"depth":569,"text":214},{"id":241,"depth":569,"text":241},{"id":419,"depth":569,"text":419},{"id":478,"depth":569,"text":479},{"id":508,"depth":569,"text":508},{"id":532,"depth":569,"text":532},"\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。",[1592,1595,1598,1601],{"q":1593,"a":1594},"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":1596,"a":1597},"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":1599,"a":1600},"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":1602,"a":1603},"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。",[599,1605],"zh",{},[607,606,608],[1609,1612],{"plan":1256,"price":1142,"features":1610,"notes":1611},"全功能 GUI + Headless API + GGUF\u002FMLX","供个人 \u002F 评估使用",{"plan":1262,"price":1613,"features":1614,"notes":1615},"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",[1147,1148],{"power":631,"ux":632,"price":632,"cn_support":566,"stability":631},{"title":262,"description":1590},"LM Studio 评测 2026：本地运行开源大模型，图形化界面，AI 模型管理",[1622,1624,1626],{"name":1623,"url":1563,"accessed":639},"LM Studio 官网",{"name":1625,"url":1570,"accessed":639},"Codersera — LM Studio Complete Guide 2026",{"name":1158,"url":1103,"accessed":639},"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",[1117,1630,1166,1167,1631,1632,1168],"gui","llama-cpp","mac","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","Pj3jNb1Z4S55e91UkW1ZMgI86ps_Wm7yCc7zeXPF05U",{"id":1637,"title":1359,"alternatives":1638,"api_compatible":1639,"body":1640,"category":1117,"chinese_friendly":631,"cover":2072,"description":2073,"domestic":582,"extension":583,"faq":2074,"free":597,"github":2087,"languages":2088,"lastVerified":601,"meta":2089,"models":15,"navigation":597,"notSuitable":15,"opensource":597,"path":528,"pillar":1136,"platforms":2090,"priceTable":2092,"pricing":2100,"published":627,"relatedPlaybooks":2101,"relatedReviews":15,"score":2102,"self_host":597,"seo":2103,"seoTitle":2104,"slug":14,"sources":2105,"stem":2112,"suitable":15,"tagline":2113,"tags":2114,"updated":639,"verdict":2119,"website":2120,"__hash__":2121},"tools\u002Ftools\u002Fcoding\u002Flocal\u002Fopen-webui.md",[662,661,12,13],[664,665,666,667,668,669,670,671,672,673,674,675,676,259,677],{"type":17,"value":1641,"toc":2060},[1642,1644,1647,1650,1652,1714,1716,1719,1723,1727,1755,1759,1783,1785,1819,1821,1943,1945,1988,1990,2013,2015,2035,2037],[20,1643,23],{"id":22},[25,1645,1646],{},"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。",[25,1648,1649],{},"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。",[20,1651,33],{"id":33},[35,1653,1654,1660,1666,1672,1678,1684,1690,1696,1702,1708],{},[38,1655,1656,1659],{},[41,1657,1658],{},"多模型后端","：Ollama \u002F OpenAI \u002F vLLM \u002F Anthropic \u002F Groq \u002F LocalAI \u002F 任意 OpenAI 兼容",[38,1661,1662,1665],{},[41,1663,1664],{},"多用户 + RBAC","：注册 \u002F 邀请 \u002F 角色权限 \u002F 工作区隔离",[38,1667,1668,1671],{},[41,1669,1670],{},"RAG 知识库","：上传文档 \u002F 网址 \u002F SearXNG 联网搜索 → 向量化 → 对话引用",[38,1673,1674,1677],{},[41,1675,1676],{},"Tools \u002F Functions","：Python 写函数即扩展（联网 \u002F 计算器 \u002F 自定义 API）",[38,1679,1680,1683],{},[41,1681,1682],{},"mcpo","：MCP-to-OpenAPI 代理，任意 MCP 服务器接进来",[38,1685,1686,1689],{},[41,1687,1688],{},"oikb","：知识库同步本地文件夹 \u002F GitHub \u002F S3 \u002F Confluence 等 40+ 源",[38,1691,1692,1695],{},[41,1693,1694],{},"open-terminal \u002F cptr","：给 AI 真实终端 + 文件 + 沙箱执行",[38,1697,1698,1701],{},[41,1699,1700],{},"图像生成","：Stable Diffusion \u002F DALL·E \u002F 自托管接入",[38,1703,1704,1707],{},[41,1705,1706],{},"语音输入 \u002F TTS","：内置",[38,1709,1710,1713],{},[41,1711,1712],{},"企业 LTS","：custom branding + SLA + 长期支持版本（联系销售）",[20,1715,113],{"id":113},[25,1717,1718],{},"完全免费、MIT 开源、商用免费。Enterprise 提供品牌定制 + SLA + LTS。",[20,1720,1722],{"id":1721},"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）",[25,1724,1725],{},[41,1726,154],{},[35,1728,1729,1736,1739,1746,1749,1752],{},[38,1730,1731,1732,1735],{},"单条 ",[686,1733,1734],{},"docker run"," 五分钟上线",[38,1737,1738],{},"自带的多用户 + 角色权限省去重新搭 Auth",[38,1740,1741,1742,1745],{},"RAG 直传 30 个 PDF 后向量化顺利，对话中 ",[686,1743,1744],{},"#知识库"," 引用准确",[38,1747,1748],{},"mcpo 把 GitHub MCP 服务器接进来，团队对话里直接 issue \u002F PR 操作",[38,1750,1751],{},"模型切换流畅，OpenAI + Ollama 并存",[38,1753,1754],{},"SearXNG 联网搜索给模型实时信息，过时知识截止问题缓解",[25,1756,1757],{},[41,1758,185],{},[35,1760,1761,1764,1770,1777,1780],{},[38,1762,1763],{},"Docker 镜像 ~1.5GB，首次拉取偏慢",[38,1765,806,1766,1769],{},[686,1767,1768],{},"0.0.0.0"," 公网暴露要加 HTTPS + 反代",[38,1771,1772,1773,1776],{},"嵌入模型 ",[686,1774,1775],{},"sentence-transformers"," 中文效果一般，建议换 bge-m3",[38,1778,1779],{},"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）",[38,1781,1782],{},"版本升级要看 changelog，部分 minor 含 breaking 改动",[20,1784,214],{"id":214},[216,1786,1787,1793,1800,1803,1806,1809,1812],{},[38,1788,1789,1790],{},"装 Docker → ",[686,1791,1792],{},"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",[38,1794,1795,1796,1799],{},"浏览器开 ",[686,1797,1798],{},"http:\u002F\u002Flocalhost:3000"," → 注册第一个账号（管理员）",[38,1801,1802],{},"设置 → Connections → 连接 Ollama \u002F 加 OpenAI Key",[38,1804,1805],{},"Models → Pull \u002F Discover 模型",[38,1807,1808],{},"Workspaces → 建知识库 → 上传文档",[38,1810,1811],{},"Tools → 启用 \u002F 写自定义函数",[38,1813,1814,1815,1818],{},"生产部署：Nginx 反代 + Let's Encrypt + 备份 ",[686,1816,1817],{},"\u002Fapp\u002Fbackend\u002Fdata"," volume",[20,1820,241],{"id":241},[243,1822,1823,1839],{},[246,1824,1825],{},[249,1826,1827,1829,1831,1834,1837],{},[252,1828,254],{},[252,1830,1359],{},[252,1832,1833],{},"LobeChat",[252,1835,1836],{},"Cherry Studio",[252,1838,262],{},[267,1840,1841,1856,1870,1884,1898,1913,1929],{},[249,1842,1843,1845,1848,1851,1854],{},[272,1844,894],{},[272,1846,1847],{},"Docker \u002F 桌面",[272,1849,1850],{},"Web + 桌面",[272,1852,1853],{},"桌面",[272,1855,1853],{},[249,1857,1858,1861,1864,1866,1868],{},[272,1859,1860],{},"多用户",[272,1862,1863],{},"✅ 一等",[272,1865,283],{},[272,1867,934],{},[272,1869,934],{},[249,1871,1872,1875,1878,1880,1882],{},[272,1873,1874],{},"RAG",[272,1876,1877],{},"✅ 强 + oikb",[272,1879,283],{},[272,1881,283],{},[272,1883,1439],{},[249,1885,1886,1889,1892,1894,1896],{},[272,1887,1888],{},"工具 \u002F MCP",[272,1890,1891],{},"✅ mcpo",[272,1893,283],{},[272,1895,283],{},[272,1897,1439],{},[249,1899,1900,1903,1906,1909,1911],{},[272,1901,1902],{},"自托管",[272,1904,1905],{},"✅ Docker \u002F K8s",[272,1907,1908],{},"✅ Docker",[272,1910,934],{},[272,1912,934],{},[249,1914,1915,1918,1921,1924,1927],{},[272,1916,1917],{},"GitHub Stars",[272,1919,1920],{},"126k+",[272,1922,1923],{},"72k+",[272,1925,1926],{},"60k+",[272,1928,329],{},[249,1930,1931,1934,1936,1938,1941],{},[272,1932,1933],{},"开源协议",[272,1935,988],{},[272,1937,988],{},[272,1939,1940],{},"AGPL-3.0",[272,1942,1452],{},[20,1944,419],{"id":419},[35,1946,1947,1953,1961,1970,1976,1982],{},[38,1948,1949,1952],{},[41,1950,1951],{},"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",[38,1954,1955,1960],{},[41,1956,1957,1958,1818],{},"备份 ",[686,1959,1817],{},"：知识库 \u002F 用户 \u002F 对话全在里面",[38,1962,1963,1966,1967,1969],{},[41,1964,1965],{},"中文 RAG 换嵌入模型","：默认 ",[686,1968,1775],{}," 中文一般，配 bge-m3 或硅基流动嵌入 API",[38,1971,1972,1975],{},[41,1973,1974],{},"mcpo 工具范围谨慎","：MCP 给 AI 真实能力，第三方服务器审一遍",[38,1977,1978,1981],{},[41,1979,1980],{},"后端吞吐看 Ollama","：5+ 并发上 vLLM 后端，Ollama 单 worker 会排队",[38,1983,1984,1987],{},[41,1985,1986],{},"升级前看 changelog","：weekly 更新，偶有 breaking",[20,1989,479],{"id":478},[35,1991,1992,1995,1998,2001,2004,2007,2010],{},[38,1993,1994],{},"✅ 团队 \u002F 家庭 \u002F 公司多人共享 AI 平台",[38,1996,1997],{},"✅ 要 Web 端访问 \u002F 移动端兼容",[38,1999,2000],{},"✅ 自托管 \u002F 完全控制数据",[38,2002,2003],{},"✅ MCP \u002F 工具调用刚需",[38,2005,2006],{},"❌ 单人桌面体验（用 Cherry Studio）",[38,2008,2009],{},"❌ 零运维 \u002F 不愿碰 Docker",[38,2011,2012],{},"❌ iOS 原生 App 主力",[20,2014,508],{"id":508},[35,2016,2017,2023,2027,2031],{},[38,2018,2019],{},[514,2020,2022],{"href":2021},"\u002Ftools\u002Fcoding\u002Flocal\u002Flobe-chat","LobeChat 评测",[38,2024,2025],{},[514,2026,1073],{"href":1072},[38,2028,2029],{},[514,2030,517],{"href":516},[38,2032,2033],{},[514,2034,1079],{"href":1078},[20,2036,532],{"id":532},[216,2038,2039,2046,2053],{},[38,2040,2041,2042],{},"Open WebUI 官方文档 ",[514,2043,2044],{"href":2044,"rel":2045},"https:\u002F\u002Fdocs.openwebui.com\u002F",[542],[38,2047,2048,2049],{},"Local AI Master — Open WebUI Setup Guide 2026 ",[514,2050,2051],{"href":2051,"rel":2052},"https:\u002F\u002Flocalaimaster.com\u002Fblog\u002Fopen-webui-setup-guide",[542],[38,2054,2055,2056],{},"AIToolDiscovery — Set Up Open-WebUI with Ollama 2026 ",[514,2057,2058],{"href":2058,"rel":2059},"https:\u002F\u002Fwww.aitooldiscovery.com\u002Fhow-to\u002Fsetup-open-webui-ollama",[542],{"title":565,"searchDepth":566,"depth":566,"links":2061},[2062,2063,2064,2065,2066,2067,2068,2069,2070,2071],{"id":22,"depth":569,"text":23},{"id":33,"depth":569,"text":33},{"id":113,"depth":569,"text":113},{"id":1721,"depth":569,"text":1722},{"id":214,"depth":569,"text":214},{"id":241,"depth":569,"text":241},{"id":419,"depth":569,"text":419},{"id":478,"depth":569,"text":479},{"id":508,"depth":569,"text":508},{"id":532,"depth":569,"text":532},"\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。",[2075,2078,2081,2084],{"q":2076,"a":2077},"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":2079,"a":2080},"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":2082,"a":2083},"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":2085,"a":2086},"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。","https:\u002F\u002Fgithub.com\u002Fopen-webui\u002Fopen-webui",[599,1605],{},[1138,608,606,607,2091],"kubernetes",[2093,2096],{"plan":1141,"price":1142,"features":2094,"notes":2095},"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","MIT 协议",{"plan":138,"price":2097,"features":2098,"notes":2099},"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",[1147,1148],{"power":632,"ux":631,"price":632,"cn_support":631,"stability":632},{"title":1359,"description":2073},"Open WebUI 评测 2026：自托管 ChatGPT 替代，Ollama 前端部署指南",[2106,2108,2110],{"name":2107,"url":2044,"accessed":639},"Open WebUI 官方文档",{"name":2109,"url":2051,"accessed":639},"Local AI Master — Open WebUI Setup Guide 2026",{"name":2111,"url":2058,"accessed":639},"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",[1117,2115,1138,2116,2117,2118,1169],"self-host","rag","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","5Y-zy_XMSVV_gsS0aa6V7zk268r2TvaIcssIp-HXM_U",1785660639079]