[{"data":1,"prerenderedAt":756},["ShallowReactive",2],{"docs-page-cn-\u002Fcn\u002Fopenclaw\u002Flocal_plugin":3,"surround-cn-\u002Fcn\u002Fopenclaw\u002Flocal_plugin":742},{"id":4,"title":5,"avatar":6,"banner":6,"body":7,"category":6,"desc":735,"description":244,"extension":736,"links":6,"meta":737,"navigation":6,"path":738,"seo":739,"stem":740,"__hash__":741},"docs\u002Fcn\u002Fopenclaw\u002Flocal_plugin.md","本地插件",null,{"type":8,"value":9,"toc":718},"minimark",[10,21,36,45,48,52,140,142,145,148,225,227,230,235,238,271,274,284,286,290,293,326,329,331,335,338,360,363,365,369,372,413,416,418,422,425,475,477,481,548,555,557,560,563,637,640,642,645,693,695,698,714],[11,12,13],"note",{},[14,15,16,20],"p",{},[17,18,19],"strong",{},"兼容性建议","\n为避免插件加载或运行报错，建议在 MemOS 完成对应 Agent 最新版本适配后，再升级 Agent。",[14,22,23,27,28,31,32,35],{},[24,25,26],"code",{},"@memtensor\u002Fmemos-local-plugin"," 是 MemOS 新一代本地插件：一套本地优先的记忆核心，同时适配 ",[17,29,30],{},"OpenClaw"," 与 ",[17,33,34],{},"Hermes Agent","。它不会把记忆数据托管到云端，而是在你的机器上维护 SQLite 数据库、技能包和日志，让 Agent 在本地持续积累可复用经验。",[14,37,38,39,44],{},"如果你只想为 OpenClaw 快速接入云端托管记忆，请查看 ",[40,41,43],"a",{"href":42},"\u002Fcn\u002Fopenclaw\u002Fguide","OpenClaw 云插件","。如果你更看重隐私、本机运行、可观测性，或希望 OpenClaw \u002F Hermes 都使用同一套本地记忆能力，请使用本页面的本地插件。",[46,47],"br",{},[49,50,51],"h2",{"id":51},"核心能力",[53,54,55,68],"table",{},[56,57,58],"thead",{},[59,60,61,65],"tr",{},[62,63,64],"th",{},"能力",[62,66,67],{},"说明",[69,70,71,80,88,96,104,112,120,132],"tbody",{},[59,72,73,77],{},[74,75,76],"td",{},"本地优先",[74,78,79],{},"OpenClaw 与 Hermes 各自拥有独立运行目录，SQLite、Skill、日志和配置都保留在本机。",[59,81,82,85],{},[74,83,84],{},"双 Agent 适配",[74,86,87],{},"OpenClaw 通过 TypeScript 插件进程内接入；Hermes 通过 Python Provider + JSON-RPC 桥接到同一套 Node.js 记忆核心。",[59,89,90,93],{},[74,91,92],{},"四层记忆",[74,94,95],{},"L1 Trace 记录每一步执行，L2 Policy 归纳跨任务策略，L3 World Model 压缩环境认知，Skill 将高价值经验结晶为可调用能力。",[59,97,98,101],{},[74,99,100],{},"三层检索",[74,102,103],{},"按 Skill → Trace\u002FEpisode → World Model 检索，并融合向量、FTS5、关键词 pattern 与错误特征，使用 RRF + MMR 控制相关性和多样性。",[59,105,106,109],{},[74,107,108],{},"反馈驱动进化",[74,110,111],{},"工具结果、环境反馈、用户显式反馈会更新记忆价值，推动策略归纳、技能结晶和 decision repair。",[59,113,114,117],{},[74,115,116],{},"本地 Viewer",[74,118,119],{},"提供 Overview、Memories、Tasks、Policies、World Models、Skills、Analytics、Logs、Import、Settings、Help 等页面。",[59,121,122,125],{},[74,123,124],{},"导入与迁移",[74,126,127,128,131],{},"支持 JSON 导入导出、旧版插件数据迁移，以及按当前 Agent 导入 OpenClaw 会话 JSONL 或 Hermes ",[24,129,130],{},"MEMORY.md","。",[59,133,134,137],{},[74,135,136],{},"可选团队共享",[74,138,139],{},"默认完全隔离；如需协作，可在 Memory Viewer 的 Team Sharing 面板中开启局域网 \u002F VPN 内 Skill 和可选 Trace 摘要共享。",[46,141],{},[49,143,144],{"id":144},"工作原理",[14,146,147],{},"插件在每轮任务开始前检索相关上下文，并把结果注入给 Agent；任务结束后，它会把对话、工具调用、观察结果和反馈写入本地流水线。高价值模式会从原始 Trace 逐步沉淀为 Policy、World Model 和可调用 Skill。下次遇到相似任务时，Agent 可以直接得到“该怎么做”和“哪些坑要避开”的上下文。",[53,149,150,163],{},[56,151,152],{},[59,153,154,157,160],{},[62,155,156],{},"阶段",[62,158,159],{},"发生了什么",[62,161,162],{},"产物",[69,164,165,179,192,203,214],{},[59,166,167,170,176],{},[74,168,169],{},"1. Agent 适配",[74,171,172,173,131],{},"OpenClaw \u002F Hermes 通过各自 Adapter 把会话、工具调用和反馈交给统一的 ",[24,174,175],{},"MemoryCore",[74,177,178],{},"标准化的 turn、tool outcome、feedback",[59,180,181,184,189],{},[74,182,183],{},"2. 本地写入",[74,185,186,188],{},[24,187,175],{}," 把执行过程拆成可追溯的步骤记录。",[74,190,191],{},"L1 Trace",[59,193,194,197,200],{},[74,195,196],{},"3. 经验归纳",[74,198,199],{},"多个相似 Trace 会归纳为跨任务策略，并进一步压缩为环境认知。",[74,201,202],{},"L2 Policy、L3 World Model",[59,204,205,208,211],{},[74,206,207],{},"4. 技能结晶",[74,209,210],{},"高价值策略会生成可调用 Skill，并根据后续反馈更新可靠性。",[74,212,213],{},"Skill、η、生命周期状态",[59,215,216,219,222],{},[74,217,218],{},"5. 检索注入",[74,220,221],{},"下一轮任务开始前，Retriever 从 Skill、Trace\u002FEpisode、World Model 三层召回上下文。",[74,223,224],{},"注入给 Agent 的本地记忆上下文",[46,226],{},[49,228,229],{"id":229},"快速开始",[231,232,234],"h3",{"id":233},"step-1一行命令安装或升级","Step 1：一行命令安装或升级",[14,236,237],{},"安装与升级使用同一条命令。当前安装脚本面向 macOS \u002F Linux：",[239,240,245],"pre",{"className":241,"code":242,"language":243,"meta":244,"style":244},"language-bash shiki shiki-themes material-theme-lighter github-light-high-contrast github-dark-default","curl -fsSL https:\u002F\u002Fraw.githubusercontent.com\u002FMemTensor\u002FMemOS\u002Fmain\u002Fapps\u002Fmemos-local-plugin\u002Finstall.sh | bash\n","bash","",[24,246,247],{"__ignoreMap":244},[248,249,252,256,260,264,268],"span",{"class":250,"line":251},"line",1,[248,253,255],{"class":254},"saonL","curl",[248,257,259],{"class":258},"sp2SP"," -fsSL",[248,261,263],{"class":262},"sp1uZ"," https:\u002F\u002Fraw.githubusercontent.com\u002FMemTensor\u002FMemOS\u002Fmain\u002Fapps\u002Fmemos-local-plugin\u002Finstall.sh",[248,265,267],{"class":266},"saN0X"," |",[248,269,270],{"class":254}," bash\n",[14,272,273],{},"安装器会自动检测系统中是否已安装 OpenClaw \u002F Hermes。交互式终端会询问安装到哪个 Agent；非交互环境会自动安装到检测到的 Agent。安装器会部署插件代码、安装生产依赖，并在需要时重启对应运行时。",[275,276,277],"blockquote",{},[14,278,279,280,283],{},"不建议直接 ",[24,281,282],{},"npm install"," 这个包。安装脚本会处理 Agent 检测、目录布局、配置初始化和运行时重启。",[46,285],{},[231,287,289],{"id":288},"step-2打开-memory-viewer","Step 2：打开 Memory Viewer",[14,291,292],{},"安装完成后，打开对应的 Memory Viewer：",[53,294,295,305],{},[56,296,297],{},[59,298,299,302],{},[62,300,301],{},"Agent",[62,303,304],{},"Memory Viewer",[69,306,307,316],{},[59,308,309,311],{},[74,310,30],{},[74,312,313],{},[24,314,315],{},"http:\u002F\u002F127.0.0.1:18799",[59,317,318,321],{},[74,319,320],{},"Hermes",[74,322,323],{},[24,324,325],{},"http:\u002F\u002F127.0.0.1:18800",[14,327,328],{},"如果你同时安装了 OpenClaw 和 Hermes，它们会使用各自独立的 Viewer 和本地数据目录。",[46,330],{},[231,332,334],{"id":333},"step-3在面板里完成配置","Step 3：在面板里完成配置",[14,336,337],{},"所有用户可见配置都从 Memory Viewer 修改：",[339,340,341,348,354],"ul",{},[342,343,344,347],"li",{},[17,345,346],{},"Settings → AI Models","：配置 Embedding、LLM、Skill Evolver，并用 Test 按钮确认可用。",[342,349,350,353],{},[17,351,352],{},"Settings → Team Sharing","：开启或关闭团队共享，配置团队地址与 token。",[342,355,356,359],{},[17,357,358],{},"Settings → General","：配置语言、日志详细程度、匿名 telemetry 等。",[14,361,362],{},"保存后，Viewer 会自动重启插件并加载新设置。",[46,364],{},[231,366,368],{"id":367},"step-4启动对应-agent","Step 4：启动对应 Agent",[14,370,371],{},"安装完成后，按你选择的 Agent 正常启动即可。插件会在 Agent 构建 prompt 前检索本地上下文，并在本轮任务结束后把对话、工具调用、观察结果和反馈写入本地记忆。",[53,373,374,386],{},[56,375,376],{},[59,377,378,380,383],{},[62,379,301],{},[62,381,382],{},"启动方式",[62,384,385],{},"插件接入方式",[69,387,388,400],{},[59,389,390,392,395],{},[74,391,30],{},[74,393,394],{},"正常启动或重启 OpenClaw gateway",[74,396,397,398],{},"TypeScript 插件在 OpenClaw 进程内调用 ",[24,399,175],{},[59,401,402,404,410],{},[74,403,320],{},[74,405,406,407],{},"运行 ",[24,408,409],{},"hermes chat",[74,411,412],{},"Python Provider 通过 JSON-RPC 调用 Node.js 记忆核心",[14,414,415],{},"如果 Hermes 所在机器无法运行 Node.js，Hermes Provider 会报告不可用，并回退到 Hermes 自身的内存模式。",[46,417],{},[231,419,421],{"id":420},"step-5验证记忆功能","Step 5：验证记忆功能",[14,423,424],{},"回到 Memory Viewer，建议检查以下页面：",[426,427,428,434,440,446,455],"ol",{},[342,429,430,433],{},[17,431,432],{},"Overview","：确认核心状态、版本、事件流正常。",[342,435,436,439],{},[17,437,438],{},"Memories","：确认对话和工具步骤被写入为 Trace。",[342,441,442,445],{},[17,443,444],{},"Tasks \u002F Policies \u002F World Models \u002F Skills","：查看经验如何逐步归纳和结晶。",[342,447,448,451,452,454],{},[17,449,450],{},"Import","：导入旧版数据、OpenClaw 会话 JSONL、Hermes ",[24,453,130],{},"，或导入 \u002F 导出 JSON 备份。",[342,456,457,460,461,464,465,464,468,464,471,474],{},[17,458,459],{},"Help","：查看每个字段含义，例如 ",[24,462,463],{},"V","、",[24,466,467],{},"α",[24,469,470],{},"R_human",[24,472,473],{},"η","、support、gain 等。",[46,476],{},[49,478,480],{"id":479},"agent-差异","Agent 差异",[53,482,483,494],{},[56,484,485],{},[59,486,487,490,492],{},[62,488,489],{},"项目",[62,491,30],{},[62,493,320],{},[69,495,496,513,526,537],{},[59,497,498,501,506],{},[74,499,500],{},"接入方式",[74,502,503,504],{},"TypeScript 插件，进程内调用 ",[24,505,175],{},[74,507,508,509,512],{},"Python ",[24,510,511],{},"MemoryProvider","，通过 stdio JSON-RPC 调用 Node bridge",[59,514,515,518,522],{},[74,516,517],{},"默认 Viewer",[74,519,520],{},[24,521,315],{},[74,523,524],{},[24,525,325],{},[59,527,528,531,534],{},[74,529,530],{},"模型配置",[74,532,533],{},"在 OpenClaw Viewer 的 Settings → AI Models 中配置",[74,535,536],{},"在 Hermes Viewer 的 Settings → AI Models 中配置",[59,538,539,542,545],{},[74,540,541],{},"数据共享",[74,543,544],{},"默认与 Hermes 隔离",[74,546,547],{},"默认与 OpenClaw 隔离",[14,549,550,551,554],{},"两个 Agent 即使安装在同一台机器上，也会使用各自的数据库和 Viewer。只有显式开启 ",[24,552,553],{},"hub:"," 后，才会进行团队共享。",[46,556],{},[49,558,559],{"id":559},"可用工具",[14,561,562],{},"OpenClaw 与 Hermes 会通过各自宿主暴露记忆工具，常见能力包括：",[53,564,565,575],{},[56,566,567],{},[59,568,569,572],{},[62,570,571],{},"工具",[62,573,574],{},"用途",[69,576,577,587,597,607,617,627],{},[59,578,579,584],{},[74,580,581],{},[24,582,583],{},"memory_search",[74,585,586],{},"按查询检索相关 Skill、Trace\u002FEpisode、World Model。",[59,588,589,594],{},[74,590,591],{},[24,592,593],{},"memory_get",[74,595,596],{},"获取某条记忆详情。",[59,598,599,604],{},[74,600,601],{},[24,602,603],{},"memory_timeline",[74,605,606],{},"查看某个 episode \u002F task 的时间线。",[59,608,609,614],{},[74,610,611],{},[24,612,613],{},"skill_list",[74,615,616],{},"列出可调用 Skill。",[59,618,619,624],{},[74,620,621],{},[24,622,623],{},"skill_get",[74,625,626],{},"获取某个 Skill 的调用指南。",[59,628,629,634],{},[74,630,631],{},[24,632,633],{},"memory_environment",[74,635,636],{},"查询 L3 World Model，了解项目结构、环境规律和约束。",[14,638,639],{},"插件也会记录工具调用成功 \u002F 失败结果，用于后续 decision repair。",[46,641],{},[49,643,644],{"id":644},"数据管理",[339,646,647,657,670,684],{},[342,648,649,652,653,656],{},[17,650,651],{},"备份","：在 Viewer 的 Import 页面导出 JSON，或备份当前 Agent 的 ",[24,654,655],{},"~\u002F.\u003Cagent>\u002Fmemos-plugin\u002F"," 目录。",[342,658,659,662,663,666,667,131],{},[17,660,661],{},"仅清空记忆","：在确认已备份后删除运行目录下的 ",[24,664,665],{},"data\u002F"," 和 ",[24,668,669],{},"skills\u002F",[342,671,672,675,676,679,680,683],{},[17,673,674],{},"清空日志","：删除 ",[24,677,678],{},"logs\u002F"," 下普通日志。",[24,681,682],{},"audit.log"," 会按月 gzip 保留。",[342,685,686,689,690,692],{},[17,687,688],{},"彻底重置","：删除整个 ",[24,691,655],{},"，下次启动会重新创建空目录。",[46,694],{},[49,696,697],{"id":697},"更多资料",[339,699,700,708],{},[342,701,702],{},[40,703,707],{"href":704,"rel":705},"https:\u002F\u002Fgithub.com\u002FMemTensor\u002FMemOS\u002Ftree\u002Fmain\u002Fapps\u002Fmemos-local-plugin",[706],"nofollow","MemOS 本地插件项目",[342,709,710],{},[40,711,713],{"href":712},"\u002Fcn\u002Fopenclaw\u002Fplugin_compare","云插件 vs 本地插件",[715,716,717],"style",{},"html pre.shiki code .saonL, html code.shiki .saonL{--shiki-light:#E2931D;--shiki-default:#702C00;--shiki-dark:#FFA657}html pre.shiki code .sp2SP, html code.shiki .sp2SP{--shiki-light:#91B859;--shiki-default:#023B95;--shiki-dark:#79C0FF}html pre.shiki code .sp1uZ, html code.shiki .sp1uZ{--shiki-light:#91B859;--shiki-default:#032563;--shiki-dark:#A5D6FF}html pre.shiki code .saN0X, html code.shiki .saN0X{--shiki-light:#39ADB5;--shiki-default:#A0111F;--shiki-dark:#FF7B72}html .light .shiki span {color: var(--shiki-light);background: var(--shiki-light-bg);font-style: var(--shiki-light-font-style);font-weight: var(--shiki-light-font-weight);text-decoration: var(--shiki-light-text-decoration);}html.light .shiki span {color: var(--shiki-light);background: var(--shiki-light-bg);font-style: var(--shiki-light-font-style);font-weight: var(--shiki-light-font-weight);text-decoration: var(--shiki-light-text-decoration);}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"title":244,"searchDepth":719,"depth":719,"links":720},2,[721,722,723,731,732,733,734],{"id":51,"depth":719,"text":51},{"id":144,"depth":719,"text":144},{"id":229,"depth":719,"text":229,"children":724},[725,727,728,729,730],{"id":233,"depth":726,"text":234},3,{"id":288,"depth":726,"text":289},{"id":333,"depth":726,"text":334},{"id":367,"depth":726,"text":368},{"id":420,"depth":726,"text":421},{"id":479,"depth":719,"text":480},{"id":559,"depth":719,"text":559},{"id":644,"depth":719,"text":644},{"id":697,"depth":719,"text":697},"使用 @memtensor\u002Fmemos-local-plugin 为 OpenClaw 与 Hermes Agent 提供本地优先的长期记忆、三层检索、技能结晶和可观测管理面板。","md",{},"\u002Fcn\u002Fopenclaw\u002Flocal_plugin",{"title":5,"description":244},"cn\u002Fopenclaw\u002Flocal_plugin","j1WqhFvSGcrCtKTmmfgK8n8f3KtiNYQ4c5-u9Qs968s",[743,749],{"title":43,"path":42,"stem":744,"icon":745,"framework":6,"module":6,"class":746,"target":-1,"active":747,"defaultOpen":747,"children":-1,"description":748},"openclaw\u002Fguide","i-ri-cloud-line",[],false,"增强 OpenClaw 的记忆能力并减少 72% 的 Token 消耗：MemOS OpenClaw 插件现已上线！",{"title":750,"path":751,"stem":752,"icon":753,"framework":6,"module":6,"class":754,"target":-1,"active":747,"defaultOpen":747,"children":-1,"description":755},"知识库使用","\u002Fcn\u002Fopenclaw\u002Fexamples\u002Fknowledge_base","openclaw\u002Fexamples\u002Fknowledge_base","i-ri-database-2-line",[],"在 MemOS Cloud 控制台创建知识库，并在 OpenClaw 云插件中配置知识库 ID 进行检索。",1781767990601]