MIT 授權、可自我演化的 AI Agent 記憶作業層；透過 @mindmemos/deepseek-harness-plugin 外掛安裝進 dsh profile 串接 DeepSeek Harness，後端可選官方雲端或自建 FastAPI 服務。
DSH 整合
相容
作者聲明
安全稽核
未稽核
最後核實
2026-08-21
授權條款
MIT
01它能幫你完成什麼？
Give DeepSeek Harness (dsh) agents persistent cross-session agent memory with automatic recall and write-back
Relevant user/task memories recalled and injected before each user turn, and the conversation written back automatically when the turn ends
DeepSeek Harness users who want their dsh agents to remember preferences, project facts and past lessons across sessions
Run MindMemOS as a self-hosted memory service, or connect the official cloud without deploying
A FastAPI memory service on http://127.0.0.1:8000 (via `make dev` with the Docker dependency stack), or the official cloud at https://mindmemos.cn — both speak the same HTTP API / SDK / plugin protocol
Teams that need on-premises or offline agent memory, or want to try the service without deploying anything themselves
Turn accumulated agent experience into reusable skills
Skill candidates distilled from experience memories, with skill execution results, failure traces and user feedback flowing back into the memory system for continuous evolution
Agent builders who want long-term memory to evolve into capability through schema learning and skill distillation
02如何將外掛接入 DeepSeek Harness？
先決條件
- DeepSeek Harness (dsh) installed with a profile — the plugin is added to a profile via `dsh plugin`
- Python SDK installed (`pip install mindmemos-sdk`) — the plugin communicates with the local machine through the `mindmemos` CLI
- A MindMemOS backend: the official cloud service (API key from the website) or a local self-hosted service started with `make dev`
安裝步驟
- 01
Install the Python SDK and confirm the CLI works: `pip install mindmemos-sdk`, then `mindmemos --version`
- 02
Configure base_url, API key and user_id with `mindmemos auth` (pointing at either the cloud or a local service), then check with `mindmemos config show`
- 03
Install the plugin into a dsh profile: `dsh plugin --profile <name> add @mindmemos/deepseek-harness-plugin`
$ dsh plugin --profile <name> add @mindmemos/deepseek-harness-plugin
- 04
Register it by adding an `insert` entry (id: mindmemos-memory, name: '@mindmemos/deepseek-harness-plugin', config: userId / appId) to `~/.dsh/profiles/<name>/cordis.patch.yml`
- 05
Restart dsh with that profile
驗證整合成功
- `mindmemos --version` confirms the CLI is available; `mindmemos config show` confirms the configuration took effect
- For the service itself, the README suggests a curl smoke test against /v1/memory/add and /v1/memory/search: a `code` of `ok` with readable memory content means the access works
03DSH 整合程度與能力邊界
Official DeepSeek Harness plugin (@mindmemos/deepseek-harness-plugin) installed into a dsh profile via `dsh plugin` and registered through cordis.patch.yml; recalls and writes memories around every turn
Per-turn recall & write-back for dsh
every DeepSeek Harness conversation turn→relevant memories injected before each user turn; the conversation written back automatically when the turn ends
sends conversation content to the configured MindMemOS endpoint (cloud https://mindmemos.cn or local http://127.0.0.1:8000)Portable cross-agent memory assets
user profiles, preferences, project facts, tool experience, skill candidates→persistent memory assets that OpenClaw, Hermes, Claude Code, OpenHands and other agents can share or transfer
Self-evolving memory (schema learning & dreaming)
ongoing interactions and stored memories→improved memory quality via frequent-pattern schema learning, offline consolidation (dreaming), and interaction corrections
offline consolidation rewrites and consolidates stored memoriesMemory → Skill distillation
experience memories and skill execution feedback→skill candidates distilled from memory; execution results, failure traces and user feedback flow back into the memory system
04適合誰？何時不該用？
適合
- DeepSeek Harness users who want their dsh agents to remember preferences, project facts and past lessons across sessions
- Teams that need on-premises or offline agent memory, or want to try the service without deploying anything themselves
- Agent builders who want long-term memory to evolve into capability through schema learning and skill distillation
不適合
- The dsh plugin depends on the Python SDK: `mindmemos-sdk` must be installed and `mindmemos auth` completed before installing the plugin, otherwise the logs error out (mindmemos command not found / auth not configured) and memories cannot be read or written.
- Local self-hosting is heavyweight: `make dev` starts a full Docker dependency stack (Qdrant + Neo4j + Kafka) before FastAPI, and at least the chat / embed / rerank model routers must be configured in config/mindmemos/dev.yaml.
05相容性、維護與安全提醒
- The dsh plugin depends on the Python SDK: `mindmemos-sdk` must be installed and `mindmemos auth` completed before installing the plugin, otherwise the logs error out (mindmemos command not found / auth not configured) and memories cannot be read or written.
- Local self-hosting is heavyweight: `make dev` starts a full Docker dependency stack (Qdrant + Neo4j + Kafka) before FastAPI, and at least the chat / embed / rerank model routers must be configured in config/mindmemos/dev.yaml.
- In cloud mode, memory data is sent to the official service at https://mindmemos.cn and requires a website-issued API key; local mode keeps traffic on http://127.0.0.1:8000.
MIT · actively maintained (latest release v0.1.5, 2026-07-22)
06常見問題
MindMemOS 如何串接 DeepSeek Harness？
先安裝 Python SDK（`pip install mindmemos-sdk`）並執行 `mindmemos auth` 設定 base_url、API key 與 user_id；接著執行 `dsh plugin --profile <name> add @mindmemos/deepseek-harness-plugin` 安裝到 dsh profile，在 `~/.dsh/profiles/<name>/cordis.patch.yml` 加入一條 `insert` 註冊項目（id 為 mindmemos-memory），重新啟動該 profile 的 dsh 即可。
它是 dsh 原生外掛還是 MCP？
它是透過 `dsh plugin` 安裝的 npm 外掛套件，藉由 dsh 分層的 `cordis.patch.yml` 檔案組合註冊，並非 MCP。外掛在本機透過 `mindmemos` CLI 與服務通訊，再轉發到雲端或本機服務。
需要哪些前置條件？
必須先安裝 Python SDK 並完成 `mindmemos auth`，否則外掛會因找不到 mindmemos 指令或未設定認證而報錯，無法正常讀寫記憶。後端二選一：官方雲端服務（至官網申請 API key），或用 `make dev` 啟動的本機服務。
記憶資料會存到哪裡？
雲端模式會傳往 https://mindmemos.cn；本地自建則留在 http://127.0.0.1:8000，兩者使用同一套協定。本機部署會用 Docker 啟動 Qdrant + Neo4j + Kafka 等相依服務，並需在 dev.yaml 設定 chat / embed / rerank 模型路由。
07相關的 DSH 工作流程
memos
作者 memtensor
面向 LLM 與 AI 智慧體的自進化記憶系統，提供超持久記憶、混合檢索與跨任務技能複用，可顯著節省 Token。
dsh-anchored-standard
作者 xiaobright
DeepSeek Harness 兩階段預設：先最小對齊引導，再啟用完整標準工具集。
mem9
作者 mem9-ai
每輪自動回憶 + 後臺智慧攝取 + 五個記憶工具，為 DSH 加持記憶。
dsh-router-standard
作者 yjh051108
已併入 dsh-routing-suite（單倉庫化）；本倉庫為歷史映象/歸檔 —— 注意力工程主線 v1.19.1/v34 研發線未釋出。新程式碼見 github.com/yjh051108/dsh-routing-suite
08資料與來源
We released the [DeepSeek Harness Plugin](https://www.npmjs.com/package/@mindmemos/deepseek-harness-plugin), letting Dee…
Connect MindMemOS to different agents and workflows through plugins that retrieve and inject relevant memories before in…
此頁面根據專案公開文件、儲存庫中繼資料與 DSH Plugins 的結構化解析所產生；最後核實於 2026-08-21。發現錯誤？提交更正。
最佳 DeepSeek Harness 外掛
從全目錄挑出的 12 個值得優先安裝的外掛，涵蓋各個分類。