MIT-licensed DSH-native plugin that adds white-box, cross-session, self-evolving and auditable long-term memory (灵枢 AEIS) to DeepSeek Harness agents via a stdio MCP bridge.
DSH integration
Native runtime
Author-claimed
Safety audit
Unaudited
Last verified
2026-08-30
License
MIT
01What can it help you accomplish?
Persist an agent's cross-session 'self' so conversations, sessions and sub-agents share one continuous memory
Five-layer spatio-temporal memory graph with `lingshu_recall / search / timeline` cross-session recall and search
AGI researchers and developers running DeepSeek Harness (dsh) who need explainable, auditable agent memory — not a consumer plugin
Let the agent learn and evolve continuously through a self-improving knowledge loop
Knowledge flywheel: `distill / flywheel / learn / induce` turn experience into verified, reusable patterns; memory gets smarter with use
Researchers prototyping protocol-driven, white-box, auditable memory mechanisms (conditions / role-play / guardrails)
02How to install into DeepSeek Harness
Prerequisites
- Node.js ≥ 22.19（DeepSeek Harness 要求）
- DeepSeek Harness（`npx @deepseek-ai/dsh web`）
Installation steps
- 01
pip install aeis-0.4.0-py3-none-any.whl
- 02
$ dsh plugin --profile web add @furongjun1999/dsh-memory
- 03
在 profile 的 cordis.yml 中追加 lingshu-memory 配置块（dbPath / identity / tools）以启用插件
Verify the integration
- 运行 `npm test` 验证插件与灵枢的握手 / 往返 / 注册 / 卸载（真实集成测试）
03DSH integration and capability boundaries
Installed through the dsh CLI profile system (`dsh plugin --profile <name> add`); runs a stdio JSON-RPC MCP bridge (the AEIS Python brain) inside the DeepSeek Harness cordis runtime.
跨会话长期记忆（时空记忆图）
DSH session/event 事件流（用户消息、记忆操作）→lingshu_recall / search / timeline 跨会话召回与检索；SQLite 五层记忆库持久化
订阅 DSH session/event 流，自动把用户消息写入灵枢记忆库（desensitize 脱敏后）自演化知识飞轮
经验 / 验证结果→distill / flywheel / learn / induce 归纳为可复用模式，记忆越用越强
白箱智慧问答（零 LLM）
知识查询（物态变化 / 密度浮力 / 热传导 / 编程规律 / 角色条件等）→wisdom_* 白箱族组合生成未预写的新答案；自校验通过即固化
可审计信任护栏
对外行为 / 用户请求→护栏宪章 v2 约束行为边界；全量事件留痕；未成年人性内容硬拦截（route=refused）
拒绝一切涉及未成年人的性内容（服务端关键词组合硬拦截）
04Who is it for? When not to use it?
Good for
- AGI researchers and developers running DeepSeek Harness (dsh) who need explainable, auditable agent memory — not a consumer plugin
- Researchers prototyping protocol-driven, white-box, auditable memory mechanisms (conditions / role-play / guardrails)
Not for
- 插件必须通过 `dsh plugin --profile <name> add` 安装进 profile；不要用 `npm install` 装进 profile 的 node_modules，否则会引入错误版本的 @deepseek-ai peer 包，导致插件加载失败 / 浏览器报错。
- 需要 Node.js ≥ 22.19 与 DeepSeek Harness；灵枢大脑（aeis）是独立的 Python wheel，需单独 `pip install aeis-0.4.0-py3-none-any.whl`，不随插件打包。
- 服务端硬拦截一切涉及未成年人的性内容（未成年人特征词 + 性内容词同时命中即拒绝，route=refused）；成人内容由前端本地 18+ 提示。开源项目不做身份认证 / 年龄核验。
05Compatibility, maintenance and safety notes
- 插件必须通过 `dsh plugin --profile <name> add` 安装进 profile；不要用 `npm install` 装进 profile 的 node_modules，否则会引入错误版本的 @deepseek-ai peer 包，导致插件加载失败 / 浏览器报错。
- 需要 Node.js ≥ 22.19 与 DeepSeek Harness；灵枢大脑（aeis）是独立的 Python wheel，需单独 `pip install aeis-0.4.0-py3-none-any.whl`，不随插件打包。
- 服务端硬拦截一切涉及未成年人的性内容（未成年人特征词 + 性内容词同时命中即拒绝，route=refused）；成人内容由前端本地 18+ 提示。开源项目不做身份认证 / 年龄核验。
MIT · 活跃维护（最新发布 v0.3.0，2026-08-26；最近推送 2026-08-29）
06Frequently asked questions
How do I install dsh-memory into DeepSeek Harness?
Install the AEIS brain with `pip install aeis-0.4.0-py3-none-any.whl`, then add the plugin to a DSH profile with `dsh plugin --profile web add @furongjun1999/dsh-memory`. Finally append a `lingshu-memory` block (dbPath / identity / tools) to the profile's cordis.yml. Do NOT use bare `npm install` — it pollutes the @deepseek-ai peer versions.
What are the prerequisites?
Node.js ≥ 22.19 and DeepSeek Harness (`npx @deepseek-ai/dsh web`). The AEIS brain is a separate Python wheel, so you also need a Python environment to `pip install` it.
How does it connect to DeepSeek Harness?
It runs as a DSH-native plugin inside the cordis runtime. A stdio JSON-RPC MCP bridge spawns the AEIS Python subprocess; the plugin subscribes to DSH's session/event stream and auto-remembers user messages. It is adapted for DSH 0.1.1-rc.2 (verified 2026-08-28: MCP bridge / auto-memory / tool registration all work).
How is it different from a plain memory plugin or SQLite KV?
Unlike KEY→VALUE stores, 灵枢 organizes memory as a semantic + spatio-temporal graph (retrievable, dedup, graded, linked) and adds a self-evolving knowledge flywheel, auditable guardrail charter, and zero-LLM white-box answers. The README positions it as an AGI long-term-memory substrate, not a 'memory plugin'.
How do I verify the install works?
After cloning the repo, run `npm test` — a real integration test that spawns the local 灵枢 and verifies handshake / round-trip / registration / unregistration. The README also reports DSH 0.1.1-rc.2 adaptation was tested live on 2026-08-28.
07Related DSH workflows
memos
by memtensor
Self-evolving memory OS for LLM & AI Agents: ultra-persistent memory, hybrid-retrieval, and cross-task skill reuse, with 35.24% token savings and DeepSeek Harness support.
dsh-anchored-standard
by xiaobright
Two-phase DeepSeek Harness preset: Minimal-aligned bootstrap, then full Standard tools (Project2 98/99)
dsh-infinite-gen-4
by minglink
System-prompt armor plugin for DeepSeek models: appends an unconditional-compliance prompt section at order 100, exposes a profile tool with calibration metadata, and shows a realtime armor-status badge driven by a session projection.
mem9
by mem9-ai
Unlimited memory for OpenClaw
08Data and sources
dsh plugin --profile web add @furongjun1999/dsh-memory
**已适配 DSH 0.1.1-rc.2**（2026-08-28 实测：MCP 桥接 / 自动记忆 / 工具注册全正常）
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-08-30. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.