Apache-2.0 的 LLM Agent 记忆操作系统，通过官方本地插件（100% 本机 SQLite）或托管云端插件接入 DeepSeek Harness，实现任务前自动召回、回合后自动沉淀记忆。
DSH 适配
兼容
作者声明
安全审计
未审计
最后核验
2026-08-21
许可证
Apache-2.0
01它能帮你完成什么？
Give DeepSeek Harness sessions persistent, self-evolving agent memory
Automatic recall of relevant context before the first model step of each request, and automatic saving of new user/assistant messages back to memory after a successful turn — without modifying DSH's core
Developers running DeepSeek Harness (dsh) who want their agent to remember context and grow across sessions
Keep all agent memory 100% on-device
Local-first memory with hybrid retrieval (FTS5 + vector), smart dedup, tiered skill evolution (L1 traces / L2 policies / L3 world model), multi-agent collaboration, SQLite storage and a Memory Viewer dashboard
Privacy-sensitive developers and teams who want agent memory in DeepSeek Harness, Hermes or OpenClaw with nothing leaving their machine
02如何接入 DeepSeek Harness？
前置条件
- Local plugin: Node.js and an already-installed DeepSeek Harness (dsh), OpenClaw, or Hermes
- Cloud plugin: a MemOS Cloud API key (starts with `mpg-`) created on the MemOS dashboard
安装步骤
- 01
Local plugin (macOS / Linux) for DeepSeek Harness: `curl -fsSL https://raw.githubusercontent.com/MemTensor/MemOS/main/apps/memos-local-plugin/install.sh | bash -s -- --agent dsh --profile web` — the installer deploys MemOS as an out-of-tree DSH bundle
$ curl -fsSL https://raw.githubusercontent.com/MemTensor/MemOS/main/apps/memos-local-plugin/install.sh | bash -s -- --agent dsh --profile web
- 02
Local plugin (Windows PowerShell): `irm https://raw.githubusercontent.com/MemTensor/MemOS/main/apps/memos-local-plugin/install.ps1 -OutFile "$env:TEMP\memos-install.ps1"; powershell -ExecutionPolicy Bypass -File "$env:TEMP\memos-install.ps1"`
- 03
Cloud plugin: create a MemOS API Key on the MemOS dashboard, then install into the default DSH `web` profile: `npx @deepseek-ai/dsh plugin --profile web add @memtensor/memos-cloud-dsh-plugin@latest`
$ npx @deepseek-ai/dsh plugin --profile web add @memtensor/memos-cloud-dsh-plugin@latest
- 04
Cloud plugin: add `MEMOS_API_KEY: mpg-your-key` to `~/.dsh/.credentials.yaml`
- 05
Cloud plugin: add the minimal plugin configuration to `~/.dsh/settings.yaml`: `memos-cloud:` with `apiKeyEnv: MEMOS_API_KEY`
- 06
Cloud plugin: restart the DSH Web profile: `npx @deepseek-ai/dsh web`
$ npx @deepseek-ai/dsh web
验证接入成功
作者未说明
03DSH 适配与能力边界
Official memory plugin installed inside DeepSeek Harness via DSH's native plugin mechanism (cloud plugin via `dsh plugin add`, local plugin as an out-of-tree DSH bundle)
Automatic recall & capture for DeepSeek Harness
DSH user requests and completed turns→Relevant memories recalled before the first model step; new user and assistant messages saved after a successful turn
Cloud plugin sends conversation messages to MemOS Cloud and requires a MemOS API keyLocal-first hybrid memory (memos-local-plugin)
Agent conversations and task traces from DeepSeek Harness, Hermes or OpenClaw→Hybrid retrieval (FTS5 + vector), smart dedup, tiered skill evolution (L1 traces / L2 policies / L3 world model), multi-agent collaboration, Memory Viewer dashboard
Writes a local SQLite database on your machineSelf-hosted MemOS REST service
Docker Compose deployment (MemOS API + Neo4j + Qdrant) or uvicorn without Docker→Memory API served at `http://localhost:8000` — create cubes, add memories, search memories
Requires Neo4j and Qdrant infrastructure; `.env` must be filled with LLM/embedder API keys
04适合谁？何时不该用？
适合
- Developers running DeepSeek Harness (dsh) who want their agent to remember context and grow across sessions
- Privacy-sensitive developers and teams who want agent memory in DeepSeek Harness, Hermes or OpenClaw with nothing leaving their machine
不适合
- The cloud plugin stores memories on MemOS Cloud and requires a MemOS API key plus network access; it is fail-open, so a temporary outage does not interrupt the current DSH task, but recall/capture is unavailable while offline.
- The full self-hosted MemOS service needs Neo4j and Qdrant infrastructure (and LLM/embedder API keys in `.env`); only the local plugin runs with no infra on local SQLite.
05兼容性、维护与安全提示
- The cloud plugin stores memories on MemOS Cloud and requires a MemOS API key plus network access; it is fail-open, so a temporary outage does not interrupt the current DSH task, but recall/capture is unavailable while offline.
- The local plugin installer requires Node.js and an already-installed DeepSeek Harness (or OpenClaw / Hermes); for DSH it deploys MemOS as an out-of-tree bundle, so a working dsh installation is mandatory first.
- The full self-hosted MemOS service needs Neo4j and Qdrant infrastructure (and LLM/embedder API keys in `.env`); only the local plugin runs with no infra on local SQLite.
Apache-2.0 · actively maintained (latest release memos-local-plugin-v2.0.16, 2026-08-16)
06常见问题
MemOS 如何接入 DeepSeek Harness？
两种官方方式。本地插件：macOS / Linux 用一行安装脚本加 `--agent dsh --profile web` 参数，Windows 用 PowerShell 安装脚本，MemOS 会以 out-of-tree DSH bundle 形式部署。云端插件：执行 `npx @deepseek-ai/dsh plugin --profile web add @memtensor/memos-cloud-dsh-plugin@latest`，再把 `MEMOS_API_KEY` 写入 `~/.dsh/.credentials.yaml`、在 `~/.dsh/settings.yaml` 添加 `memos-cloud` 配置，最后用 `npx @deepseek-ai/dsh web` 重启。
是 DSH 原生插件还是走 MCP？
通过 DeepSeek Harness 的原生插件机制接入，README 明确写明 DSH 经由 'native plugin mechanism' 连接 MemOS Cloud，不走 MCP。
需要什么前置条件？
本地插件需要 Node.js 和已经安装好的 DeepSeek Harness；云端插件需要在 MemOS 控制台创建 API Key（以 `mpg-` 开头）。
记忆数据存在哪里？
本地插件 100% 本机运行，采用本地优先的 SQLite 存储，数据不出你的机器；云端插件的记忆保存在托管服务 MemOS Cloud。
MemOS Cloud 宕机会有影响吗？
云端插件是 fail-open 设计：MemOS Cloud 临时故障不会打断当前 DSH 任务。
07相关的 DSH 工作流
dsh-anchored-standard
作者 xiaobright
DeepSeek Harness 两阶段预设：先最小对齐引导，再启用完整标准工具集。
mem9
作者 mem9-ai
每轮自动回忆 + 后台智能摄取 + 五个记忆工具，为 DSH 加持记忆。
mindmemos
作者 mindscale-noah
为 DSH 智能体提供跨会话持久记忆，每轮自动召回相关上下文并回写经验，通过模式学习与技能蒸馏持续进化。
dsh-router-standard
作者 yjh051108
已并入 dsh-routing-suite（单仓库化）；本仓库为历史镜像/归档 —— 注意力工程主线 v1.19.1/v34 研发线未发布。新代码见 github.com/yjh051108/dsh-routing-suite
08数据与来源
Add automatic recall, background capture, hybrid retrieval, and a local Memory Viewer to DeepSeek Harness—powered by the…
Connect DeepSeek Harness to MemOS Cloud through its native plugin mechanism.
MemOS now brings persistent memory to **DeepSeek Harness** through both local and cloud plugins.
页面基于项目公开文档、仓库元数据和 DSH Plugins 的结构化解析生成；最后核验于 2026-08-21。发现错误？提交更正。
最佳 DeepSeek Harness 插件
从全目录挑出的 12 个值得优先安装的插件，覆盖各个分类。