Apache-2.0 的持久 agent memory 伺服器，為 DeepSeek Harness 提供原生 DSH/Cordis 外掛包，可託管或自架，支援多執行環境共享記憶與混合檢索。
DSH 整合
原生執行環境
作者聲明
安全稽核
未稽核
最後核實
2026-08-21
授權條款
Apache-2.0
01它能幫你完成什麼？
Give DeepSeek Harness agents persistent memory across sessions and machines
Recall, smart ingest, and memory tools that store, search, get, update, and delete memories against the mem9 server, so agent context survives restarts, laptop switches, and long-running projects
DeepSeek Harness (dsh) users whose agents lose context between sessions
Share one agent memory layer across multiple runtimes and workflow platforms
A single mem9 space that DeepSeek Harness, OpenClaw, Claude Code, OpenCode, Codex, Dify apps, and custom HTTP clients read and write together via the same store/search/get/update/delete API
Teams running multi-agent workflows across different agent runtimes who want one shared memory instead of separate local notebooks
02如何將外掛接入 DeepSeek Harness？
先決條件
- A mem9 endpoint — the hosted API at https://api.mem9.ai, or a self-hosted mnemo-server backed by TiDB, PostgreSQL, or db9
- A mem9 API key for your space (returned or configured by your provisioning flow)
安裝步驟
- 01
Choose your mem9 endpoint: use the hosted API (https://api.mem9.ai), or self-host by applying the matching control-plane schema and starting mnemo-server with MNEMO_DSN
- 02
Follow the DeepSeek Harness integration guide in dsh-plugin/README.md to install the native DSH/Cordis bundle
- 03
Set your credentials: export MEM9_API_URL (hosted or self-hosted server URL) and export MEM9_API_KEY
驗證整合成功
作者未說明
03DSH 整合程度與能力邊界
Native DSH/Cordis bundle with recall, smart ingest, and memory tools, backed by the mem9 memory server (hosted API or self-hosted)
Persistent memory across sessions and machines
agent conversations and durable facts→memories that survive restarts, laptop switches, and long-running projects
memories are written to the mem9 server (hosted API or self-hosted database), not kept purely localShared multi-agent memory space
one mem9 space with an X-API-Key (optionally X-Mnemo-Agent-Id per agent)→store, search, get, update, and delete operations shared across OpenClaw, Hermes Agent, Claude Code, OpenCode, Codex, DeepSeek Harness, Dify and custom clients
all connected runtimes read and write the same memory spaceHybrid recall and visual dashboard
natural-language queries against the memory space→semantic search, keyword search, and inspection workflows in one system, plus a dashboard to inspect and manage memory
Hosted or self-hosted deployment
the hosted mem9 API (https://api.mem9.ai), or self-hosted mnemo-server on tidb / postgres / db9→the same API contract either way — switching deployment is a base-URL and credential change, not a plugin rewrite
self-hosting requires applying the control-plane schema and running mnemo-server with MNEMO_DSN
04適合誰？何時不該用？
適合
- DeepSeek Harness (dsh) users whose agents lose context between sessions
- Teams running multi-agent workflows across different agent runtimes who want one shared memory instead of separate local notebooks
不適合
- mem9 is not a purely local plugin — it requires a running mem9 endpoint, either the hosted API at https://api.mem9.ai or a self-hosted mnemo-server with a TiDB, PostgreSQL, or db9 backend and the matching control-plane schema applied.
- Memory recall and ingest are server-side operations sent over the network: with the hosted API, memories live in mem9's TiDB Cloud Starter-based service, so sensitive context may leave the local machine unless you self-host.
05相容性、維護與安全提醒
- mem9 is not a purely local plugin — it requires a running mem9 endpoint, either the hosted API at https://api.mem9.ai or a self-hosted mnemo-server with a TiDB, PostgreSQL, or db9 backend and the matching control-plane schema applied.
- Memory recall and ingest are server-side operations sent over the network: with the hosted API, memories live in mem9's TiDB Cloud Starter-based service, so sensitive context may leave the local machine unless you self-host.
- Self-hosted smart ingest depends on an LLM provider API key (MNEMO_LLM_API_KEY); without it, ingest silently falls back to raw ingest behavior, which may lower memory quality.
Apache-2.0 · actively maintained (last push 2026-08-19)
06常見問題
如何把 mem9 接入 DeepSeek Harness？
mem9 提供原生 DSH/Cordis 外掛包，內建召回、智慧寫入與記憶工具。請依 dsh-plugin/README.md 中的 DeepSeek Harness 指南安裝外掛，再把 MEM9_API_URL 與 MEM9_API_KEY 指向託管 API 或你的自架伺服器即可。
一定要自架 mem9 伺服器嗎？
不必。最快的方式是直接使用託管 API（https://api.mem9.ai）。自架也完整支援：套用對應後端的控制層 schema（TiDB、PostgreSQL 或 db9），再用 MNEMO_DSN 啟動 mnemo-server。兩者 API 契約相同，之後切換只需變更 base URL 與憑證。
我的 agent memory 儲存在哪裡？
儲存在 MEM9_API_URL 指向的 mem9 空間：託管 API 建於 TiDB Cloud Starter 之上；自架時則儲存在你自己的資料庫。記憶的寫入、搜尋、更新與刪除都透過 mem9 伺服器 API 完成。
多個智能體可以共享同一份記憶嗎？
可以。單一空間可同時服務 OpenClaw、Hermes Agent、Claude Code、OpenCode、Codex、DeepSeek Harness、Dify 應用與自訂用戶端，所有執行環境都能召回同一批事實。設定 X-Mnemo-Agent-Id 可區分共享空間內的寫入來源。
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。
mindmemos
作者 mindscale-noah
為 DSH 智慧體提供跨會話持久記憶，每輪自動召回相關上下文並回寫經驗，通過模式學習與技能蒸餾持續進化。
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
| DeepSeek Harness | Native DSH/Cordis bundle with recall, smart ingest, and memory tools | [`dsh-plugin/README.md`](dsh…
Hermes Agent, Claude Code, OpenCode, Codex, DeepSeek Harness, and Dify guides are below.
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。