[{"data":1,"prerenderedAt":1154},["ShallowReactive",2],{"header-counts":3,"footer-counts":6,"compare-msty-vs-ollama":9,"compare-a-msty":10,"compare-b-ollama":656},{"tools":4,"reviews":5},78,26,{"tools":4,"reviews":5,"playbooks":7,"news":8},22,21,null,{"id":11,"title":12,"alternatives":13,"api_compatible":9,"body":17,"category":580,"chinese_friendly":567,"cover":581,"description":582,"domestic":583,"extension":584,"faq":585,"free":583,"github":9,"languages":598,"lastVerified":9,"meta":601,"models":9,"navigation":602,"notSuitable":9,"opensource":583,"path":603,"pillar":604,"platforms":605,"priceTable":610,"pricing":626,"published":627,"relatedPlaybooks":628,"relatedReviews":9,"score":630,"self_host":602,"seo":633,"seoTitle":9,"slug":634,"sources":635,"stem":645,"suitable":9,"tagline":646,"tags":647,"updated":638,"verdict":653,"website":654,"__hash__":655},"tools\u002Ftools\u002Fagent\u002Fgeneral\u002Fmsty.md","Msty",[14,15,16],"coding\u002Flocal\u002Follama","coding\u002Flocal\u002Flm-studio","coding\u002Flocal\u002Fopen-webui",{"type":18,"value":19,"toc":565},"minimark",[20,25,29,32,35,112,115,141,147,151,156,182,187,213,216,240,243,418,421,477,481,507,510,531,534],[21,22,24],"h2",{"id":23},"tldr","TL;DR",[26,27,28],"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。",[26,30,31],{},"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）。",[21,33,34],{"id":34},"核心能力",[36,37,38,46,52,58,64,70,76,82,88,94,100,106],"ul",{},[39,40,41,45],"li",{},[42,43,44],"strong",{},"三引擎本地推理","：MLX（Apple）\u002F llama.cpp \u002F Ollama 开箱即用",[39,47,48,51],{},[42,49,50],{},"Hosted Models","：OpenAI \u002F Anthropic \u002F Gemini 一窗体接入",[39,53,54,57],{},[42,55,56],{},"Split Chats","：同问题同时跑多模型 + side-by-side 对比",[39,59,60,63],{},[42,61,62],{},"Knowledge Stack","：per-conversation RAG，文档 \u002F URL \u002F Obsidian vault \u002F YouTube transcript",[39,65,66,69],{},[42,67,68],{},"Prompt Studio","：变量 + 模板 + 测试",[39,71,72,75],{},[42,73,74],{},"Persona Studio","：自定义角色 + 工具 + context",[39,77,78,81],{},[42,79,80],{},"Skills Studio","：可复用能力包",[39,83,84,87],{},[42,85,86],{},"Agent Mode","：多步执行 + 工具调用",[39,89,90,93],{},[42,91,92],{},"Flowchart 对话","：分支可视化",[39,95,96,99],{},[42,97,98],{},"Real-time 数据","：实时 web fetch",[39,101,102,105],{},[42,103,104],{},"Offline-first","：零账号 \u002F 零 telemetry \u002F 零云依赖（Free）",[39,107,108,111],{},[42,109,110],{},"Cloud Sync","（Aurum）：跨设备同步对话",[21,113,114],{"id":114},"价格",[36,116,117,123,129,135],{},[39,118,119,122],{},[42,120,121],{},"Free","：$0 永久；本地全功能 + 云 API 接入 + Split Chats + Knowledge Stack",[39,124,125,128],{},[42,126,127],{},"Aurum","：$149\u002F年；cloud sync + 高级 Knowledge + Studio Desktop alpha + 优先支持",[39,130,131,134],{},[42,132,133],{},"Lifetime","：$349 一次；Aurum 全部功能 + 终身更新",[39,136,137,140],{},[42,138,139],{},"Enterprise","：$300\u002Fuser·年；SSO + 私有部署 + 团队管理",[142,143,144],"blockquote",{},[26,145,146],{},"Lifetime 在 2 年用回本，重度用户首选。Free 已经足够个人 90% 场景。",[21,148,150],{"id":149},"实测macos-mac-mini-私人-ai-服务器","实测（macOS + Mac mini 私人 AI 服务器）",[26,152,153],{},[42,154,155],{},"亮点：",[36,157,158,161,164,167,170,173,176,179],{},[39,159,160],{},"装完立刻能用，无需 Ollama \u002F llama.cpp \u002F MLX 任何 CLI",[39,162,163],{},"Split Chats 对比 GPT-5 + Claude + Qwen2.5 + Llama 3.3 一目了然",[39,165,166],{},"Knowledge Stack 的 per-conversation 设计完美：每项目独立 RAG context",[39,168,169],{},"Prompt \u002F Persona \u002F Skills 三 Studio 解决重复 prompt 痛点",[39,171,172],{},"MLX 在 M1\u002FM2\u002FM3\u002FM4 上跑得快",[39,174,175],{},"中文 Qwen2.5 \u002F DeepSeek 走本地路径无外网依赖",[39,177,178],{},"Free 永久免费 + 本地无限 = 真正 zero-cost 路径",[39,180,181],{},"Lifetime $349 比 ChatGPT Plus 18 个月便宜",[26,183,184],{},[42,185,186],{},"踩坑：",[36,188,189,192,195,198,201,204,207,210],{},[39,190,191],{},"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",[39,193,194],{},"插件 \u002F Skill 生态比 Typing Mind \u002F LM Studio 弱",[39,196,197],{},"桌面 only，无 iOS \u002F Android",[39,199,200],{},"llama.cpp 更新滞后官方上游 1-2 版本",[39,202,203],{},"Knowledge Stack 大文档（>100MB）切分偶尔失败",[39,205,206],{},"Agent Mode 仍在打磨，复杂任务稳定性不如 Claude",[39,208,209],{},"Studio Desktop（Aurum alpha）测试中，bug 偶发",[39,211,212],{},"中文 UI 不完整，部分功能仍是英文",[21,214,215],{"id":215},"上手",[217,218,219,222,225,228,231,234,237],"ol",{},[39,220,221],{},"msty.ai → 下载 macOS \u002F Windows \u002F Linux → 安装",[39,223,224],{},"Settings → Model Providers → Ollama 连本地（或直接装 Msty 自带 llama.cpp \u002F MLX）",[39,226,227],{},"装 1-2 个本地模型：Qwen2.5 7B（中文）+ Llama 3.3 8B（英文）",[39,229,230],{},"加云模型：OpenAI Key + Anthropic Key",[39,232,233],{},"新建 chat → 试 Split Chats：+ 第二个模型 → 同问题并行",[39,235,236],{},"Knowledge Stack → 上传项目文档 → 附加到 conversation",[39,238,239],{},"满意后 Free 用着，重度需要同步上 Lifetime $349",[21,241,242],{"id":242},"对比",[244,245,246,267],"table",{},[247,248,249],"thead",{},[250,251,252,256,258,261,264],"tr",{},[253,254,255],"th",{},"维度",[253,257,12],{},[253,259,260],{},"Ollama",[253,262,263],{},"LM Studio",[253,265,266],{},"Jan",[268,269,270,287,302,317,331,344,358,373,387,401],"tbody",{},[250,271,272,276,279,282,285],{},[273,274,275],"td",{},"GUI",[273,277,278],{},"✅ 颜值高",[273,280,281],{},"❌ CLI",[273,283,284],{},"✅",[273,286,284],{},[250,288,289,292,295,297,300],{},[273,290,291],{},"本地引擎",[273,293,294],{},"MLX\u002Fllama.cpp\u002FOllama",[273,296,260],{},[273,298,299],{},"llama.cpp",[273,301,299],{},[250,303,304,307,309,312,314],{},[273,305,306],{},"云模型",[273,308,284],{},[273,310,311],{},"❌",[273,313,311],{},[273,315,316],{},"部分",[250,318,319,321,324,326,328],{},[273,320,56],{},[273,322,323],{},"✅ 旗舰",[273,325,311],{},[273,327,316],{},[273,329,330],{},"–",[250,332,333,335,338,340,342],{},[273,334,62],{},[273,336,337],{},"✅ per-conv",[273,339,311],{},[273,341,311],{},[273,343,316],{},[250,345,346,349,352,354,356],{},[273,347,348],{},"Studios",[273,350,351],{},"✅ Prompt\u002FPersona\u002FSkills",[273,353,330],{},[273,355,330],{},[273,357,330],{},[250,359,360,363,365,368,370],{},[273,361,362],{},"开源",[273,364,311],{},[273,366,367],{},"✅ MIT",[273,369,311],{},[273,371,372],{},"✅ Apache 2.0",[250,374,375,378,381,383,385],{},[273,376,377],{},"起价",[273,379,380],{},"$0",[273,382,380],{},[273,384,380],{},[273,386,380],{},[250,388,389,392,395,397,399],{},[273,390,391],{},"终身",[273,393,394],{},"$349",[273,396,330],{},[273,398,330],{},[273,400,330],{},[250,402,403,406,409,412,415],{},[273,404,405],{},"适合",[273,407,408],{},"桌面颜值 + 多模型",[273,410,411],{},"CLI \u002F Server",[273,413,414],{},"模型市场",[273,416,417],{},"严格开源",[21,419,420],{"id":420},"避坑",[36,422,423,429,435,441,447,453,459,465,471],{},[39,424,425,428],{},[42,426,427],{},"硬件评估","：M1 Air 8GB 只跑 3B-7B，M3 Pro \u002F Max 跑 13B-30B 流畅",[39,430,431,434],{},[42,432,433],{},"Ollama 已装就连","：避免重复下模型，连本地 Ollama 复用 model library",[39,436,437,440],{},[42,438,439],{},"Knowledge Stack 文档","：单文档 \u003C50MB 最稳，大文件先切分",[39,442,443,446],{},[42,444,445],{},"Persona vs Skill","：Persona 是角色（完整 system + 模型）；Skill 是能力包；不要混用",[39,448,449,452],{},[42,450,451],{},"Split Chats 三个模型够","：4 个起每问 token 烧得快",[39,454,455,458],{},[42,456,457],{},"Aurum cloud sync 谨慎","：隐私敏感场景仍用 Free 本地",[39,460,461,464],{},[42,462,463],{},"Studio Desktop alpha","：稳定性不如 main 版本，重要工作不要全押",[39,466,467,470],{},[42,468,469],{},"本地中文模型","：Qwen2.5 7B \u002F DeepSeek 7B 中文最优，Llama 3.3 8B 英文最优",[39,472,473,476],{},[42,474,475],{},"MCP server","：当前不如 Claude Desktop 强，要 MCP 重度场景考虑 Claude Desktop \u002F Crush",[21,478,480],{"id":479},"适合-不适合","适合 \u002F 不适合",[36,482,483,486,489,492,495,498,501,504],{},[39,484,485],{},"✅ 隐私敏感 + 数据零云",[39,487,488],{},"✅ Mac mini \u002F Linux box 私人 AI 服务器",[39,490,491],{},"✅ 不愿学 Ollama CLI 的非工程师",[39,493,494],{},"✅ 多模型对比决策场景",[39,496,497],{},"❌ 硬件不行（8GB RAM）跑不动 7B+",[39,499,500],{},"❌ 要 mobile 移动主力",[39,502,503],{},"❌ 团队协作 + 共享 workspace",[39,505,506],{},"❌ 要 MCP 工具栈深度（用 Claude Desktop \u002F Crush）",[21,508,509],{"id":509},"相关阅读",[36,511,512,519,525],{},[39,513,514],{},[515,516,518],"a",{"href":517},"\u002Ftools\u002Fcoding\u002Flocal\u002Follama","Ollama 评测",[39,520,521],{},[515,522,524],{"href":523},"\u002Ftools\u002Fcoding\u002Flocal\u002Flm-studio","LM Studio 评测",[39,526,527],{},[515,528,530],{"href":529},"\u002Ftools\u002Fcoding\u002Flocal\u002Fopen-webui","Open WebUI 评测",[21,532,533],{"id":533},"来源",[217,535,536,544,551,558],{},[39,537,538,539],{},"Msty 官网 + Features（MLX \u002F llama.cpp \u002F Ollama \u002F Studios）",[515,540,541],{"href":541,"rel":542},"https:\u002F\u002Fmsty.ai\u002Fstudio\u002Ffeatures",[543],"nofollow",[39,545,546,547],{},"AI Chat Daily — Msty Review 2026（4.3\u002F5 评分 + Lifetime）",[515,548,549],{"href":549,"rel":550},"https:\u002F\u002Fwww.aichatdaily.com\u002Ftools\u002Fmsty",[543],[39,552,553,554],{},"ML Journey — Msty Multi-model Comparison Guide ",[515,555,556],{"href":556,"rel":557},"https:\u002F\u002Fmljourney.com\u002Fmsty-the-local-llm-app-that-lets-you-compare-models-side-by-side",[543],[39,559,560,561],{},"AISO Tools — Msty Pricing 2026 ",[515,562,563],{"href":563,"rel":564},"https:\u002F\u002Faisotools.com\u002Fpricing\u002Fmsty",[543],{"title":566,"searchDepth":567,"depth":567,"links":568},"",3,[569,571,572,573,574,575,576,577,578,579],{"id":23,"depth":570,"text":24},2,{"id":34,"depth":570,"text":34},{"id":114,"depth":570,"text":114},{"id":149,"depth":570,"text":150},{"id":215,"depth":570,"text":215},{"id":242,"depth":570,"text":242},{"id":420,"depth":570,"text":420},{"id":479,"depth":570,"text":480},{"id":509,"depth":570,"text":509},{"id":533,"depth":570,"text":533},"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",[586,589,592,595],{"q":587,"a":588},"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":590,"a":591},"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":593,"a":594},"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":596,"a":597},"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。",[599,600],"en","multi",{},true,"\u002Ftools\u002Fagent\u002Fgeneral\u002Fmsty","agent",[606,607,608,609],"macos","windows","linux","web",[611,614,618,622],{"plan":121,"price":380,"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":127,"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":133,"price":619,"features":620,"notes":621},"$349 一次","Aurum 终身授权","永久 + 所有更新",{"plan":139,"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":567,"stability":631},4,5,{"title":12,"description":582},"agent\u002Fgeneral\u002Fmsty",[636,639,641,643],{"name":637,"url":541,"accessed":638},"Msty 官网 + Features","2026-06-24",{"name":640,"url":549,"accessed":638},"AI Chat Daily — Msty Review 2026 4.3\u002F5",{"name":642,"url":556,"accessed":638},"ML Journey — Msty Local LLM Comparison Guide",{"name":644,"url":563,"accessed":638},"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",[648,649,650,651,652],"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","niNqR8c2HfHsRBNs01ip9Crt9cywkIdrVITosmvDYyA",{"id":657,"title":260,"alternatives":658,"api_compatible":9,"body":661,"category":1100,"chinese_friendly":567,"cover":1101,"description":1102,"domestic":583,"extension":584,"faq":1103,"free":583,"github":9,"languages":1116,"lastVerified":9,"meta":1117,"models":9,"navigation":602,"notSuitable":9,"opensource":602,"path":517,"pillar":1118,"platforms":1119,"priceTable":1121,"pricing":1127,"published":627,"relatedPlaybooks":1128,"relatedReviews":9,"score":1131,"self_host":602,"seo":1132,"seoTitle":9,"slug":14,"sources":1133,"stem":1140,"suitable":9,"tagline":1141,"tags":1142,"updated":638,"verdict":1151,"website":1152,"__hash__":1153},"tools\u002Ftools\u002Fcoding\u002Flocal\u002Follama.md",[15,16,659,660],"coding\u002Flocal\u002Fcherry-studio","coding\u002Flocal\u002Flobe-chat",{"type":18,"value":662,"toc":1088},[663,665,673,676,678,748,750,753,757,761,781,785,816,818,852,854,980,982,1014,1016,1039,1041,1063,1065],[21,664,24],{"id":23},[26,666,667,668,672],{},"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，",[669,670,671],"code",{},"ollama pull"," 一键拉。",[26,674,675],{},"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）。",[21,677,34],{"id":34},[36,679,680,686,695,701,709,724,730,736,742],{},[39,681,682,685],{},[42,683,684],{},"后台 Daemon","：开机自启，应用调用零延迟",[39,687,688,691,692],{},[42,689,690],{},"CLI","：",[669,693,694],{},"ollama pull \u002F run \u002F list \u002F show \u002F create \u002F serve",[39,696,697,700],{},[42,698,699],{},"Modelfile","：类 Dockerfile 注册任意 GGUF，配 SYSTEM \u002F PARAMETER \u002F TEMPLATE",[39,702,703,691,706],{},[42,704,705],{},"OpenAI 兼容 API",[669,707,708],{},"http:\u002F\u002Flocalhost:11434\u002Fv1\u002Fchat\u002Fcompletions",[39,710,711,691,714,717,718,717,721],{},[42,712,713],{},"原生 API",[669,715,716],{},"\u002Fapi\u002Fchat","、",[669,719,720],{},"\u002Fapi\u002Fgenerate",[669,722,723],{},"\u002Fapi\u002Fembeddings",[39,725,726,729],{},[42,727,728],{},"模型库","：官方注册表内置 Llama \u002F Qwen \u002F DeepSeek \u002F Gemma \u002F Mistral \u002F GPT-OSS 等",[39,731,732,735],{},[42,733,734],{},"MLX 加速（Mac）","：0.19+ 起 M 系列自动用 MLX",[39,737,738,741],{},[42,739,740],{},"量化","：默认 Q4_K_M、支持 Q5 \u002F Q8 \u002F FP16",[39,743,744,747],{},[42,745,746],{},"跨平台","：Win \u002F Mac \u002F Linux 安装包，Docker 官方镜像",[21,749,114],{"id":114},[26,751,752],{},"完全免费、MIT 开源、商用免费。",[21,754,756],{"id":755},"实测m2-pro-qwen3-coder-7b-q4","实测（M2 Pro + Qwen3-Coder-7B Q4）",[26,758,759],{},[42,760,155],{},[36,762,763,769,772,775,778],{},[39,764,765,768],{},[669,766,767],{},"ollama run qwen3-coder:7b"," 一行起飞，3 秒进交互",[39,770,771],{},"REST API 配 Cursor \u002F Cline \u002F Continue 几乎全工具开箱即用",[39,773,774],{},"Modelfile 写自定义编码助手（low temperature + system prompt + 16K context）几分钟搞定",[39,776,777],{},"多模型并存，按需切换，内存占用合理",[39,779,780],{},"Mac M 系列 MLX 后比旧 GGUF 模式快显著",[26,782,783],{},[42,784,186],{},[36,786,787,797,803,810,813],{},[39,788,789,790,793,794],{},"默认 ",[669,791,792],{},"num_ctx"," 偏小（2048），跑长上下文要在 Modelfile 加 ",[669,795,796],{},"PARAMETER num_ctx 16384",[39,798,799,800],{},"模型默认走 0.0.0.0:11434 ↔ Docker 容器互访要 ",[669,801,802],{},"--add-host=host.docker.internal:host-gateway",[39,804,805,806,809],{},"国内 ",[669,807,808],{},"ollama.com\u002Flibrary"," 下载偶有慢，可手动 HF 下 GGUF + Modelfile 自建",[39,811,812],{},"多用户并发吞吐显著低于 vLLM",[39,814,815],{},"没有 GUI，模型浏览 \u002F 参数面板要走 LM Studio \u002F Open WebUI 配合",[21,817,215],{"id":215},[217,819,820,826,832,837,843,849],{},[39,821,822,825],{},[669,823,824],{},"curl -fsSL https:\u002F\u002Follama.ai\u002Finstall.sh | sh","（Mac \u002F Linux）；Windows winget",[39,827,828,831],{},[669,829,830],{},"ollama pull qwen3-coder:7b","（按需换模型）",[39,833,834,836],{},[669,835,767],{}," 直接聊",[39,838,839,840],{},"应用接入：baseURL = ",[669,841,842],{},"http:\u002F\u002Flocalhost:11434\u002Fv1",[39,844,845,846],{},"自定义：写 Modelfile → ",[669,847,848],{},"ollama create my-coder -f Modelfile",[39,850,851],{},"进阶：装 Open WebUI 做前端 \u002F 多人共享",[21,853,242],{"id":242},[244,855,856,871],{},[247,857,858],{},[250,859,860,862,864,866,869],{},[253,861,255],{},[253,863,260],{},[253,865,263],{},[253,867,868],{},"vLLM",[253,870,299],{},[268,872,873,890,905,920,935,951,965],{},[250,874,875,878,881,884,887],{},[273,876,877],{},"形态",[273,879,880],{},"CLI + Daemon",[273,882,883],{},"GUI + Headless",[273,885,886],{},"Python Server",[273,888,889],{},"C++ 二进制",[250,891,892,894,897,899,902],{},[273,893,215],{},[273,895,896],{},"极低",[273,898,896],{},[273,900,901],{},"中",[273,903,904],{},"高",[250,906,907,910,912,915,918],{},[273,908,909],{},"模型浏览",[273,911,690],{},[273,913,914],{},"✅ GUI",[273,916,917],{},"无",[273,919,917],{},[250,921,922,925,928,931,933],{},[273,923,924],{},"OpenAI 兼容",[273,926,927],{},"✅ :11434",[273,929,930],{},"✅ :1234",[273,932,284],{},[273,934,284],{},[250,936,937,940,943,946,949],{},[273,938,939],{},"多用户吞吐",[273,941,942],{},"弱（~40 tok\u002Fs）",[273,944,945],{},"中（50–90）",[273,947,948],{},"强（800–12500）",[273,950,901],{},[250,952,953,956,959,961,963],{},[273,954,955],{},"MLX (Mac)",[273,957,958],{},"✅ 0.19+",[273,960,284],{},[273,962,316],{},[273,964,330],{},[250,966,967,969,972,975,978],{},[273,968,362],{},[273,970,971],{},"MIT",[273,973,974],{},"闭源",[273,976,977],{},"Apache 2.0",[273,979,971],{},[21,981,420],{"id":420},[36,983,984,990,996,1002,1008],{},[39,985,986,989],{},[42,987,988],{},"num_ctx 一定要设","：默认 2K 太小，跑代码 \u002F 长文档要 16K+",[39,991,992,995],{},[42,993,994],{},"Modelfile 模板别漏 TEMPLATE","：错的 chat template 会让模型输出乱码 \u002F 不停",[39,997,998,1001],{},[42,999,1000],{},"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×",[39,1003,1004,1007],{},[42,1005,1006],{},"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",[39,1009,1010,1013],{},[42,1011,1012],{},"Mac 让它自动用 MLX","：升 0.19+；不要手动强制 GGUF + Metal",[21,1015,480],{"id":479},[36,1017,1018,1021,1024,1027,1030,1033,1036],{},[39,1019,1020],{},"✅ 应用 \u002F IDE 接本地模型（Cursor \u002F Cline \u002F Continue）",[39,1022,1023],{},"✅ 个人 \u002F 评估 \u002F 脚本自动化",[39,1025,1026],{},"✅ Modelfile 自定义系统 prompt + 参数",[39,1028,1029],{},"✅ Mac M 系列 MLX 用户",[39,1031,1032],{},"❌ 多用户并发生产服务（用 vLLM）",[39,1034,1035],{},"❌ GUI 调参 \u002F 模型浏览（配 LM Studio \u002F Open WebUI）",[39,1037,1038],{},"❌ 极致单卡吞吐研究（直接 llama.cpp \u002F vLLM）",[21,1040,509],{"id":509},[36,1042,1043,1047,1051,1057],{},[39,1044,1045],{},[515,1046,524],{"href":523},[39,1048,1049],{},[515,1050,530],{"href":529},[39,1052,1053],{},[515,1054,1056],{"href":1055},"\u002Ftools\u002Fcoding\u002Flocal\u002Fcherry-studio","Cherry Studio 评测",[39,1058,1059],{},[515,1060,1062],{"href":1061},"\u002Fplaybook\u002Fonboarding\u002Frag-pipeline-build","RAG Pipeline 搭建 Playbook",[21,1064,533],{"id":533},[217,1066,1067,1074,1081],{},[39,1068,1069,1070],{},"Markaicode — Import GGUF Models into Ollama 2026（2026-05-15）",[515,1071,1072],{"href":1072,"rel":1073},"https:\u002F\u002Fmarkaicode.com\u002Fimport-gguf-models-ollama-guide",[543],[39,1075,1076,1077],{},"ComputingForGeeks — Ollama Models Cheat Sheet 2026 ",[515,1078,1079],{"href":1079,"rel":1080},"https:\u002F\u002Fcomputingforgeeks.com\u002Follama-models-cheat-sheet",[543],[39,1082,1083,1084],{},"Codersera — Ollama vs LM Studio vs vLLM vs llama.cpp vs MLX 2026 ",[515,1085,1086],{"href":1086,"rel":1087},"https:\u002F\u002Fcodersera.com\u002Fblog\u002Follama-vs-lm-studio-vs-vllm-vs-llama-cpp-vs-mlx-2026\u002F",[543],{"title":566,"searchDepth":567,"depth":567,"links":1089},[1090,1091,1092,1093,1094,1095,1096,1097,1098,1099],{"id":23,"depth":570,"text":24},{"id":34,"depth":570,"text":34},{"id":114,"depth":570,"text":114},{"id":755,"depth":570,"text":756},{"id":215,"depth":570,"text":215},{"id":242,"depth":570,"text":242},{"id":420,"depth":570,"text":420},{"id":479,"depth":570,"text":480},{"id":509,"depth":570,"text":509},{"id":533,"depth":570,"text":533},"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。",[1104,1107,1110,1113],{"q":1105,"a":1106},"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":1108,"a":1109},"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":1111,"a":1112},"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":1114,"a":1115},"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。",[599],{},"coding",[607,606,608,1120],"docker",[1122],{"plan":1123,"price":1124,"features":1125,"notes":1126},"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）",[1129,1130],"onboarding\u002Frag-pipeline-build","onboarding\u002Fclaude-code-getting-started",{"power":631,"ux":631,"price":632,"cn_support":567,"stability":632},{"title":260,"description":1102},[1134,1136,1138],{"name":1135,"url":1072,"accessed":638},"Markaicode — Import GGUF 2026",{"name":1137,"url":1079,"accessed":638},"ComputingForGeeks — Ollama Cheat Sheet 2026",{"name":1139,"url":1086,"accessed":638},"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",[1100,1143,1144,1145,1146,1147,1148,1149,1150],"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","shirZzL900qiCQXzrJS1r3bv7XatM1q8hPc2mEoq88Y",1784565440389]