MemOS — OpenClaw 记忆插件 | 本地化 · 智能进化

没有记忆的 Agent，每次都从零开始Without Memory, Every Task Starts from Zero

MemOS 为 OpenClaw 注入持久记忆与自进化技能。MemOS equips OpenClaw with persistent memory and self-evolving skills.

💻

完全本地化Fully Local

记忆、任务、技能全存本机 SQLite，零云依赖。All data stored in local SQLite — zero cloud dependency, complete privacy.

🧠

全量可视化管理Full Visualization

内置管理面板，记忆 / 任务 / 技能完全透明可控。Built-in web dashboard — memories, tasks, and skills fully transparent and controllable.

⚡

任务总结与技能进化Task Summary & Skill Evolution

碎片对话自动归纳为结构化任务，再提炼为可复用技能并持续升级。从「记住」到「学会」，同一个坑不踩两次。Fragmented conversations auto-organized into structured tasks, then distilled into reusable skills that evolve over time. From "remembering" to "mastering" — never repeat the same mistake twice.

💰

分级模型 · 省钱Tiered Models

Embedding 轻量、摘要中等、技能高质量——按需分配，大幅省钱。Lightweight, mid-tier, and high-quality models layered by purpose — maximum performance at minimum cost.

🤝

团队共享 & 多智能体协同Team Sharing & Multi-Agent

Hub-Client 架构跨实例共享记忆、任务与技能。记忆隔离 + 公共记忆 + 团队审批 + 实时通知，多 Agent 协同进化。Hub-Client architecture for cross-instance sharing of memories, tasks, and skills. Memory isolation + public memory + team approval + real-time notifications for collaborative evolution.

🦞

OpenClaw 原生记忆导入Native Memory Import

一键迁移 OpenClaw 内置记忆，智能去重、断点续传、实时进度。你过往的记忆不会丢失，再续前缘。One-click migration from OpenClaw built-in memories. Smart dedup, resume anytime, real-time progress. Your past memories, never lost.

三大引擎，驱动 Agent 协同进化Three Engines That Drive Collaborative Evolution

任务总结与技能自进化Task Summary & Skill Evolution

碎片对话自动归组为结构化任务（目标 → 步骤 → 结果），再由 LLM 评估提炼为可复用技能。遇到相似场景时自动升级——更快、更准、更省 Token。从「能记住」到「会做」，同一个坑不踩两次。任务与技能支持编辑、删除、重试等完整管理。Fragmented conversations are auto-organized into structured tasks (goal → steps → result), then LLM evaluates and distills them into reusable skills. Skills auto-upgrade on similar scenarios — faster, more accurate, lower cost. From "remembering" to "mastering" — never repeat the same mistake. Full CRUD for tasks and skills.

逐轮话题检测Per-Turn Topic Detection结构化摘要Structured Summary自动评估Auto Evaluate版本管理VersioningLLM 降级链LLM Fallback

Task → Skill Evolution

Task: "部署 Nginx 反向代理"  completed
Goal:  配置反向代理到 Node.js
Steps: 1. nginx conf  2. upstream  3. SSL  4. reload
Result: ✓ HTTPS 正常

Evaluating: shouldGenerate=true  conf=0.85
→ SKILL.md + scripts → quality 8.5/10
✓ "nginx-proxy" v1 created

// 再次执行时自动升级
Upgrade: extend → added WebSocket
✓ v2 (score: 9.0)

多智能体协同进化Multi-Agent Collaborative Evolution

每个 Agent 拥有独立的私域记忆，互不可见。但通过「公共记忆」和「技能共享」机制，Agent 之间能够共享决策、经验与能力。一个 Agent 学会的技能，可以发布为公共技能，其他 Agent 搜索并安装后即可复用。多智能体不再各自为战，而是协同进化、共同进步。Each agent has isolated private memory, invisible to others. But through public memory and skill sharing, agents can share decisions, experiences, and capabilities. Skills learned by one agent can be published for others to discover and install. Multi-agent systems no longer work in silos — they evolve collaboratively, growing together.

记忆隔离Memory Isolation公共记忆Public Memory技能共享Skill Sharing

Multi-Agent Collaboration

Agent Alpha:
  memory_search("deploy config")
  → sees own + public memories only
  memory_write_public("shared deploy config")
  skill_publish("nginx-proxy") ✓ now public

Agent Beta:
  skill_search("nginx deployment")
  → Found: nginx-proxy (public)
  skill_install("nginx-proxy") ✓ installed

全量记忆可视化管理Full Memory Visualization

内置 Web 管理面板——记忆、任务、技能、分析、日志、导入、设置共 7 页。任务以对话气泡还原，技能支持版本对比与下载，日志页可查看工具调用输入输出与耗时。Built-in dashboard — 7 pages: memories, tasks, skills, analytics, logs, import, and settings. Task details as chat bubbles. Logs show tool call I/O and duration.

127.0.0.1:18799

MemoriesTasksSkillsAnalyticsLogsImportSettings

总记忆Total

1,284

今日Today

+47

任务Tasks

12

技能Skills

8

user帮我配置 Nginx 反向代理到 3000 端口Set up Nginx proxy to port 30002m

asst好的，创建 nginx 配置文件并写入 upstream 配置。Creating nginx config file and writing upstream block.2m

user还需要加 SSL 证书Also add SSL cert5m

从对话到记忆到技能的智能闭环The Intelligent Loop: Conversation → Memory → Skill

💡 为什么这套架构对 OpenClaw 至关重要💡 Why This Architecture Matters

📋

Task：碎片→知识Tasks: Fragments→Knowledge

多轮对话组织为完整知识单元，检索效率大幅提升。Multi-turn dialogues organized into reusable knowledge units.

⚡

Skill：记住→会做Skills: Remember→Do

实战操作指南，相似任务直接调用，跳过摸索。Battle-tested procedural guides, invoked automatically on similar tasks.

🔄

自动进化：越用越强Auto-Evolution

新经验触发 Skill 升级（refine/extend/fix）。New experiences trigger automatic skill upgrades (refine / extend / fix).

💰

分级模型：按需配算力Tiered Models

轻量/中等/高质量模型分层配置，极致省钱。Purpose-matched models for maximum cost efficiency.

60 秒上手Up and Running in 60 Seconds

无论新安装还是升级，一行命令搞定One-command install. Two configuration methods.

1.一键安装/升级1. Install

一行命令安装/升级，自动检测环境，安装所有依赖和编译工具，自动升级最新版本。
遇到安装问题？查看排查指南 →One command automatically installs all dependencies and build tools, and upgrades to the latest version.
Install issues? See troubleshooting guide →

# Step 1: 安装插件 & 启动# Step 1: Install plugin & start

$ curl -fsSL https://cdn.memtensor.com.cn/memos-local-openclaw/install.sh | bash

2. 配置2. Config

网页面板：http://127.0.0.1:18799 登录后点「设置」。或编辑 openclaw.json。Web panel: http://127.0.0.1:18799 → Settings. Or edit openclaw.json.

127.0.0.1:18799

MemoriesTasksSkillsAnalyticsLogsSettings

Embedding

Provideropenai_compatible Modelbge-m3 Endpointhttps://your-api-endpoint/v1 API Keysk-••••••

Summarizer

Provideropenai_compatible Modelgpt-4o-mini Endpointhttps://your-api-endpoint/v1 API Keysk-••••••

Skill Evolution

Modelclaude-4.6-opus Endpointhttps://your-api-endpoint/v1

Viewer Port18799

保存即生效Save to apply

{
  "plugins": {
    "slots": { "memory": "memos-local-openclaw-plugin" },
    "entries": {
      "memos-local-openclaw-plugin": {
        "config": {
          "embedding": {
            "provider": "openai_compatible",
            "model": "bge-m3",
            "endpoint": "https://your-api-endpoint/v1",
            "apiKey": "sk-••••••"
          },
          "summarizer": {
            "provider": "openai_compatible",
            "model": "gpt-4o-mini",
            "endpoint": "https://your-api-endpoint/v1",
            "apiKey": "sk-••••••"
          },
          "skillEvolution": {
            "summarizer": {
              "provider": "openai_compatible",
              "model": "claude-4.6-opus",
              "endpoint": "https://your-api-endpoint/v1",
              "apiKey": "sk-••••••"
            }
          },
          "viewerPort": 18799
        }
      }
    }
  }
}

适配你的技术栈Works with Your Preferred Stack

OpenAI 兼容 API 即插即用，无配置自动降级本地模型。Any OpenAI-compatible API works out of the box. Automatic fallback to local models when no API key is configured.

OpenAI

Anthropic

Gemini

Bedrock

Cohere

Voyage

Mistral

本地Local

17 个智能工具17 Smart Tools

🧠

auto_recall

每轮自动回忆Auto recall each turn

🔍

memory_search

记忆检索Memory search

📄

memory_get

获取完整记忆Get full memory

📜

memory_timeline

上下文邻居Context neighbors

📢

memory_write_public

写入公共记忆Write public memory

📋

task_summary

任务摘要Task summary

⚡

skill_get

技能指南Skill guide

📦

skill_install

安装技能Install skill

🔎

skill_search

技能发现Skill discovery

🌍

skill_publish

公开技能Publish skill

🔒

skill_unpublish

取消公开Unpublish skill

🌐

memory_viewer

管理面板Dashboard

📤

task_share

共享任务Share task

📥

task_unshare

取消共享Unshare task

🔗

network_memory_detail

团队记忆详情Team memory detail

⬇️

network_skill_pull

拉取团队技能Pull team skill

👥

network_team_info

团队信息Team info

👥 团队共享中心Team Sharing Hub

多实例协作 —
让团队的 Agent 共同进化Multi-Instance Collaboration —
Your Team's Agents Evolve Together

Hub-Client 架构，一个服务端存储共享数据，客户端按需查询。私有数据始终留在本地，只有明确共享的内容才对团队可见。端口自动推导，多实例无冲突。Hub-Client architecture — one Hub stores shared data while clients query on demand. Private data stays local; only explicitly shared content is visible to the team. Auto-derived ports, zero conflicts for multi-instance setups.

Hub-Client 协作架构Hub-Client Architecture

一台机器作为 Hub 服务端管理团队数据，其他实例作为 Client 连接。加入需要管理员审批，角色可随时切换。支持同一台机器运行多个 OpenClaw 实例（如个人 + 工作），端口和数据完全隔离。One machine serves as the Hub managing team data while others connect as Clients. Joining requires admin approval, and roles can be switched anytime. Supports multiple OpenClaw instances on the same machine (e.g., personal + work) with full port and data isolation.

审批流程Approval Flow端口自动推导Auto Port角色切换Role Switch多实例隔离Multi-Instance

Admin Controls & Notifications

Admin Panel:
  approve(bob)       ✓ approved
  promote(alice)     → 🔔 role_promoted
  demote(alice)      → 🔔 role_demoted
  remove(bob)        → confirm → ✓ removed
  remove(self)       → ✗ blocked

Hub Shutdown:
  → 📢 hub_shutdown → all clients

Client Leave:
  → confirm → notify hub → cleanup → ✓

🔄 团队共享支持的操作🔄 Team Sharing Operations

🔍

跨域检索Scoped Search

Local / Group / All 三级搜索范围，本地 + 团队记忆混合召回。Local / Group / All search scopes with hybrid local + team recall.

📤

任务共享Task Sharing

将本地任务及记忆推送到团队，随时取消共享。Push local tasks and memories to team, unshare anytime.

⚡

技能发布与拉取Skill Publish & Pull

发布技能到团队，其他成员拉取完整 bundle 到本地离线使用。Publish skills to team; members pull full bundles locally for offline use.

🛡️

优雅降级Graceful Degradation

Hub 离线时本地记忆和技能正常运作，搜索自动降级为本地结果。Local memory and skills work normally when Hub is offline; search auto-falls back to local results.

🦞 OpenClaw 原生记忆导入OpenClaw Native Memory Import

再续前缘 —
过往的记忆，不会丢失Reconnect —
Your Past Memories, Never Lost

从 OpenClaw 原生 SQLite 和会话记录中无缝迁移，智能去重、自动摘要、技能生成一气呵成。你和 AI 共同积累的每一段对话，都值得被记住。Seamlessly migrate from OpenClaw's native SQLite and session logs. Smart deduplication, auto-summarization, and skill generation — all in one flow. Every conversation you've built with your AI deserves to be preserved.

🚀

一键迁移One-Click Import

自动扫描 OpenClaw 原生记忆文件，一键启动导入，实时显示进度与统计。Automatically scans OpenClaw native memory files. Start import with one click and monitor real-time progress.

🧬

智能去重Smart Dedup

向量相似度 + LLM 判断双重去重，相似内容自动合并，不留冗余。Vector similarity combined with LLM judgment for dual-layer deduplication. Similar content is automatically merged with zero redundancy.

⏸️

断点续传Resume Anytime

支持随时暂停，刷新页面后自动恢复进度。后台持续运行，已处理的自动跳过。Pause anytime and auto-resume on page refresh. Runs in the background, automatically skipping already processed items.

⚡

任务与技能生成Task & Skill Gen

导入后可选生成任务摘要和技能进化，同一 Agent 内串行处理，不同 Agent 之间并行（可配置 1–8 并发度），支持暂停和断点续传。Optionally generate task summaries and evolve skills. Serial within each agent, parallel across agents (configurable 1–8 concurrency), with full pause and resume support.

沉浸体验完整流程Experience the Complete Workflow

从记忆导入到智能检索再到可视化管理，一站式体验 MemOS 的核心能力。From memory import to smart retrieval to visual management — explore MemOS's core capabilities in an interactive demo.

场景一Scene 1

🦞 记忆导入🦞 Memory Import

从 OpenClaw 原生格式无缝迁移，实时进度与智能去重。Seamlessly migrate from OpenClaw's native format with real-time progress and smart deduplication.

1扫描原生记忆文件Scan native memory files

2一键导入与去重One-click import & dedup

3生成任务与技能Generate tasks & skills

🔍 智能检索🔍 Smart Retrieval

FTS5 全文 + 向量相似度 + RRF 融合 + MMR 重排，多策略混合召回。FTS5 full-text search, vector similarity, RRF fusion, and MMR reranking — multi-strategy hybrid recall for precise results.

1输入自然语言查询Enter natural language query

2多路混合检索融合Multi-path hybrid retrieval

3相关度排序展示Relevance-ranked results