[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"package:cask:swama:zh":3,"releases:stats:cask:swama:zh":205,"related:cask:swama:zh":206,"markdown:3:800:597hmd":253},{"artifacts":4,"autoUpdates":25,"categories":26,"conflictsWith":32,"dependsOn":35,"deprecated":25,"description":49,"descriptionEn":50,"disabled":25,"displayName":51,"downloadSha256":52,"downloadSize":53,"downloadUrl":54,"editorChoice":25,"formulaeUrl":55,"homepage":56,"installCommand":57,"installs":58,"installs30d":59,"isFont":25,"isLibrary":25,"kegOnly":25,"kind":62,"latestRelease":63,"machineTranslated":25,"minMacos":48,"name":51,"names":93,"platforms":95,"primaryCategory":182,"rank30d":183,"releaseCount":184,"releaseStats":185,"repoUrl":56,"screenshots":192,"sourceLocale":87,"sourceUrl":193,"summary":194,"summaryTranslation":195,"supports":196,"tags":198,"tap":203,"token":94,"version":92,"versionChangedAt":204},{"apps":5,"binaries":7,"entries":8,"pkgs":24},[6],"Swama.app",[],[9,16],{"declaration":10,"phase":13,"sources":14,"target":12,"type":15},{"app":11,"target":12},[6],"\u002FApplications\u002FSwama.app","install",[6],"app",{"declaration":17,"phase":21,"sources":22,"type":23},{"zap":18},[19],{"trash":20},"~\u002FLibrary\u002FPreferences\u002Ftrans-n.ai.Swama.plist","cleanup",[],"zap",[],false,[27],{"icon":28,"machineTranslated":25,"name":29,"slug":30,"sourceLocale":31},"lucide:sparkles","AI 工具","ai","zh-CN",{"casks":33,"formulae":34},[],[],{"arch":36,"casks":38,"formulae":39,"macos":40,"requirements":41},[37],"arm64",[],[],">= 15",{"arch":42,"macos":46},[43],{"bits":44,"type":45},64,"arm",{">=":47},[48],"15","Swama 是面向 Apple Silicon Mac 的本地 AI 运行时，使用 Swift 和 Apple 的 MLX 实现。它可在 Mac 上运行受支持的模型，并通过兼容 OpenAI 的 API、命令行工具和菜单栏应用提供服务。\n\n- 在本地运行语言、视觉、嵌入和语音识别模型；文本转语音功能仍属实验性。\n- 通过 API 提供聊天补全、嵌入、音频转录等受支持的请求。聊天补全支持流式传输、工具调用，以及使用视觉模型处理图像输入。\n- 使用 `\u002Fv1\u002Fdecisions` 端点对谓词、选项和评分进行打分，无需生成文本。\n- 使用简短别名启动模型；Swama 会在首次使用时从 Hugging Face 下载模型。API 服务器默认监听端口 `28100`。\n\n使用 `brew install --cask swama` 安装。该软件包会安装 `Swama.app`。官方列出的要求是 Apple Silicon Mac 和 macOS 15.4 或更高版本。安装应用后，如需使用命令行工具，请在菜单栏应用中选择 **Install Command Line Tool…**，将 `swama` 添加到 PATH。可运行 `swama run qwen3.5 \"Hello!\"` 开始使用；如需仅供本机访问的 API 服务器，可运行 `swama serve --host 127.0.0.1`。\n\nResponses API 是无状态子集，文本转语音为实验性功能。模型会在首次使用时下载。\n\n来源：[Swama 项目](https:\u002F\u002Fgithub.com\u002FTrans-N-ai\u002Fswama) · [Homebrew cask](https:\u002F\u002Fraw.githubusercontent.com\u002FHomebrew\u002Fhomebrew-cask\u002FHEAD\u002FCasks\u002Fs\u002Fswama.rb)","Machine-learning runtime","Swama","eaa84fe25979df8523d2334e6c491240d0bb55723cc950fa2b151eebcdbc328f",38082542,"https:\u002F\u002Fgithub.com\u002FTrans-N-ai\u002Fswama\u002Freleases\u002Fdownload\u002Fv2.5.1\u002FSwama.dmg","https:\u002F\u002Fformulae.brew.sh\u002Fcask\u002Fswama","https:\u002F\u002Fgithub.com\u002FTrans-N-ai\u002Fswama","brew install --cask swama",{"d30":59,"d365":60,"d90":61},5,455,31,"cask",{"bodyMarkdown":64,"brewCommittedAt":65,"hasNotes":66,"id":67,"isLatest":66,"isPrerelease":25,"machineTranslated":25,"publishedAt":68,"sections":69,"source":86,"sourceLocale":87,"summary":88,"title":89,"translation":90,"version":92},"Swama 2.5.1 修复了 v2.4.0 引入的问题：嵌入式 swama CLI 无法加载 SwamaCore.bundle，因此在构建该应用的 Mac 以外的设备上首次推理时可能退出。现在应用会将该 bundle 与 CLI 一同提供。应用内置服务器不受影响。建议 v2.4.0 用户升级。\n\n本版本还包含 v2.5.0 的更改；该版本的资源在发布后几分钟内即被撤回。新增遵循 OpenAI Decisions API 的本地决策 API POST \u002Fv1\u002Fdecisions，以及 SystemOne 适配器 POST \u002Fv1\u002Fsystemone。决策支持 predicate、choice 和 score 问题、可选名称、字符串或 user-message 输入，以及最多四张内联 data-URL 图像。回答按问题顺序返回，并包含类型化的 choice 值和候选项置信度。旧版 SGLang 格式的 \u002Fv1\u002Fdecisions 请求现在会收到指向 OpenAI 格式的提示。SystemOne 适配器支持 Clef 风格图像、空状态，以及超过 26 个选项时使用双字母标签。\n\n图像处理调整包括使用 detail: \"low\" 将图像缩放到 512 px，以及不再将图像放大到超出模型需求。其他更改涉及在设备上进行决策后处理、缓存推理开场内容、调整分词与评分、按到达顺序接纳等待中的模型操作、更新 mlx-swift-lm，以及修正 README。客户端断开时会取消未完成的生成；Swift 6.4 Release 构建中的模型 ID 验证不再拒绝所有 ID。应用要求使用 Apple Silicon 和 macOS 15.6 或更高版本。\n\n来源：https:\u002F\u002Fgithub.com\u002FTrans-N-ai\u002Fswama\u002Freleases\u002Ftag\u002Fv2.5.1","2026-10-07T07:47:52Z",true,8654,"2026-10-07T07:26:15Z",[70,75,81],{"area":71,"items":72},"修复",[73,74],"在嵌入式 CLI 旁附带 SwamaCore.bundle。","修复 CLI 在其他 Mac 上首次推理时退出的问题。",{"area":76,"items":77},"API",[78,79,80],"新增 POST \u002Fv1\u002Fdecisions 和 POST \u002Fv1\u002Fsystemone。","决策请求接受最多四张内联 data-URL 图像。","旧 SGLang 格式请求会收到提示，指向 OpenAI 格式。",{"area":82,"items":83},"其他",[84,85],"客户端断开连接时取消生成任务。","要求使用 Apple Silicon 和 macOS 15.6 或更高版本。","editorial","en-US","修复嵌入式 CLI 在非构建用 Mac 上首次推理时退出的问题。还包含 v2.5.0 的更改；该版本资源在发布后不久被撤回。","Swama 2.5.1",{"locale":31,"machineTranslated":25,"sourceLocale":87,"status":91},"manual","2.5.1",[51,94],"swama",[96,125,154],{"arch":37,"artifacts":97,"conflictsWith":111,"dependsOn":114,"downloadSha256":52,"downloadUrl":54,"macos":123,"minMacos":48,"requiresRosetta":25,"tag":124,"version":92},{"apps":98,"binaries":99,"entries":100,"pkgs":110},[6],[],[101,105],{"declaration":102,"phase":13,"sources":104,"target":12,"type":15},{"app":103,"target":12},[6],[6],{"declaration":106,"phase":21,"sources":109,"type":23},{"zap":107},[108],{"trash":20},[],[],{"casks":112,"formulae":113},[],[],{"arch":115,"casks":116,"formulae":117,"macos":40,"requirements":118},[37],[],[],{"arch":119,"macos":121},[120],{"bits":44,"type":45},{">=":122},[48],"27","arm64_golden_gate",{"arch":37,"artifacts":126,"conflictsWith":140,"dependsOn":143,"downloadSha256":52,"downloadUrl":54,"macos":152,"minMacos":48,"requiresRosetta":25,"tag":153,"version":92},{"apps":127,"binaries":128,"entries":129,"pkgs":139},[6],[],[130,134],{"declaration":131,"phase":13,"sources":133,"target":12,"type":15},{"app":132,"target":12},[6],[6],{"declaration":135,"phase":21,"sources":138,"type":23},{"zap":136},[137],{"trash":20},[],[],{"casks":141,"formulae":142},[],[],{"arch":144,"casks":145,"formulae":146,"macos":40,"requirements":147},[37],[],[],{"arch":148,"macos":150},[149],{"bits":44,"type":45},{">=":151},[48],"26","arm64_tahoe",{"arch":37,"artifacts":155,"conflictsWith":169,"dependsOn":172,"downloadSha256":52,"downloadUrl":54,"macos":48,"minMacos":48,"requiresRosetta":25,"tag":181,"version":92},{"apps":156,"binaries":157,"entries":158,"pkgs":168},[6],[],[159,163],{"declaration":160,"phase":13,"sources":162,"target":12,"type":15},{"app":161,"target":12},[6],[6],{"declaration":164,"phase":21,"sources":167,"type":23},{"zap":165},[166],{"trash":20},[],[],{"casks":170,"formulae":171},[],[],{"arch":173,"casks":174,"formulae":175,"macos":40,"requirements":176},[37],[],[],{"arch":177,"macos":179},[178],{"bits":44,"type":45},{">=":180},[48],"arm64_sequoia",{"icon":28,"machineTranslated":25,"name":29,"slug":30,"sourceLocale":31},2888,13,{"brewLag":186,"cadence":190,"count30d":191},{"compared":187,"earlierCount":188,"medianMinutes":189},3,0,21,"monthly",1,[],"https:\u002F\u002Fgithub.com\u002Fhomebrew\u002Fhomebrew-cask\u002Fblob\u002FHEAD\u002FCasks\u002Fs\u002Fswama.rb","面向 Apple Silicon Mac 的本地 AI 运行时，可运行 MLX 模型，并提供 API、命令行工具和菜单栏应用。",{"locale":31,"machineTranslated":25,"sourceLocale":87,"status":91},{"arm64":66,"requiresRosetta":25,"status":197,"x86_64":25},"known",[199,200,201,202],"local AI","model inference","OpenAI API","Apple Silicon","homebrew\u002Fcask","2026-10-07T09:38:42.722922Z",null,[207,217,228,237,245],{"accentColor":208,"autoUpdates":25,"deprecated":25,"disabled":25,"displayName":209,"editorChoice":25,"iconUrl":210,"installs30d":211,"isFont":25,"isLibrary":25,"kind":62,"machineTranslated":25,"name":209,"primaryCategory":212,"rank30d":213,"sourceLocale":87,"summary":214,"token":215,"version":216,"versionChangedAt":204},"#6BCED7","TauriTavern","https:\u002F\u002Fcdn.opennavo.com\u002Ficons\u002Fuploads\u002Fc68c3a5c397a\u002Ffc618101081d-256.png",14,{"icon":28,"machineTranslated":25,"name":29,"slug":30,"sourceLocale":31},2006,"一款原生 macOS 客户端，旨在提供兼容 SillyTavern 的使用体验。","tauritavern","2.3.0",{"accentColor":218,"autoUpdates":25,"deprecated":25,"disabled":25,"displayName":219,"editorChoice":25,"iconUrl":220,"installs30d":221,"isFont":25,"isLibrary":25,"kind":62,"machineTranslated":25,"name":219,"primaryCategory":222,"rank30d":223,"sourceLocale":87,"summary":224,"token":225,"version":226,"versionChangedAt":227},"#7AB901","NVIDIA Personal AI Router","https:\u002F\u002Fcdn.opennavo.com\u002Ficons\u002Fuploads\u002F7579a7810f4e\u002F0362e2d97c7e-256.png",20,{"icon":28,"machineTranslated":25,"name":29,"slug":30,"sourceLocale":31},1719,"在同一网络中的兼容计算机之间路由彼此独立的本地推理请求。","nvidia-pair","0.1.1","2026-10-07T09:38:40.131182Z",{"autoUpdates":66,"deprecated":25,"disabled":25,"displayName":229,"editorChoice":25,"installs30d":230,"isFont":25,"isLibrary":25,"kind":62,"machineTranslated":25,"name":229,"primaryCategory":231,"rank30d":232,"sourceLocale":87,"summary":233,"token":234,"version":235,"versionChangedAt":236},"MindMac",6,{"icon":28,"machineTranslated":25,"name":29,"slug":30,"sourceLocale":31},2687,"一款适用于 macOS 的 ChatGPT 客户端，支持 macOS 13 或更高版本。","mindmac","1.9.28","2026-10-07T09:38:38.931734Z",{"autoUpdates":66,"deprecated":25,"disabled":25,"displayName":238,"editorChoice":25,"installs30d":59,"isFont":25,"isLibrary":25,"kind":62,"machineTranslated":25,"name":238,"primaryCategory":239,"rank30d":240,"sourceLocale":87,"summary":241,"token":242,"version":243,"versionChangedAt":244},"LlamaChat",{"icon":28,"machineTranslated":25,"name":29,"slug":30,"sourceLocale":31},2831,"一款 macOS 应用，可在 Mac 本机与 LLaMA、Alpaca 和 GPT4All 模型聊天。","llamachat","1.2.0","2026-10-07T09:38:37.849603Z",{"autoUpdates":25,"deprecated":25,"disabled":25,"displayName":246,"editorChoice":25,"installs30d":230,"isFont":25,"isLibrary":25,"kind":62,"machineTranslated":25,"name":246,"primaryCategory":247,"rank30d":248,"sourceLocale":87,"summary":249,"token":250,"version":251,"versionChangedAt":252},"Poe",{"icon":28,"machineTranslated":25,"name":29,"slug":30,"sourceLocale":31},2711,"Poe 是面向 macOS 的 AI 聊天客户端，以 Poe 桌面应用的形式提供。","poe","1.1.45","2026-10-07T09:38:41.545736Z","\u003Cp>Swama 是面向 Apple Silicon Mac 的本地 AI 运行时，使用 Swift 和 Apple 的 MLX 实现。它可在 Mac 上运行受支持的模型，并通过兼容 OpenAI 的 API、命令行工具和菜单栏应用提供服务。\u003C\u002Fp>\n\u003Cul>\n\u003Cli>在本地运行语言、视觉、嵌入和语音识别模型；文本转语音功能仍属实验性。\u003C\u002Fli>\n\u003Cli>通过 API 提供聊天补全、嵌入、音频转录等受支持的请求。聊天补全支持流式传输、工具调用，以及使用视觉模型处理图像输入。\u003C\u002Fli>\n\u003Cli>使用 \u003Ccode>\u002Fv1\u002Fdecisions\u003C\u002Fcode> 端点对谓词、选项和评分进行打分，无需生成文本。\u003C\u002Fli>\n\u003Cli>使用简短别名启动模型；Swama 会在首次使用时从 Hugging Face 下载模型。API 服务器默认监听端口 \u003Ccode>28100\u003C\u002Fcode>。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>使用 \u003Ccode>brew install --cask swama\u003C\u002Fcode> 安装。该软件包会安装 \u003Ccode>Swama.app\u003C\u002Fcode>。官方列出的要求是 Apple Silicon Mac 和 macOS 15.4 或更高版本。安装应用后，如需使用命令行工具，请在菜单栏应用中选择 \u003Cstrong>Install Command Line Tool…\u003C\u002Fstrong>，将 \u003Ccode>swama\u003C\u002Fcode> 添加到 PATH。可运行 \u003Ccode>swama run qwen3.5 \"Hello!\"\u003C\u002Fcode> 开始使用；如需仅供本机访问的 API 服务器，可运行 \u003Ccode>swama serve --host 127.0.0.1\u003C\u002Fcode>。\u003C\u002Fp>\n\u003Cp>Responses API 是无状态子集，文本转语音为实验性功能。模型会在首次使用时下载。\u003C\u002Fp>\n\u003Cp>来源：\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FTrans-N-ai\u002Fswama\" target=\"_blank\" rel=\"noopener noreferrer\">Swama 项目\u003C\u002Fa> · \u003Ca href=\"https:\u002F\u002Fraw.githubusercontent.com\u002FHomebrew\u002Fhomebrew-cask\u002FHEAD\u002FCasks\u002Fs\u002Fswama.rb\" target=\"_blank\" rel=\"noopener noreferrer\">Homebrew cask\u003C\u002Fa>\u003C\u002Fp>\n"]