dsh-memory DSH plugin

dsh-memory is a DSH plugin that implements the Lingshu (AEIS) spatiotemporal memory engine, providing AGI-level long-term memory for DeepSeek Agent. It features a five-layer memory model, a self-evolving knowledge flywheel, and auditable trust through guardrails and charter. The plugin runs as a separate Python subprocess via MCP, offering 38+ tools for memory, reasoning, cognition, and learning.

skills-workflows developer-tools

How to install the dsh-memory DSH plugin

dsh plugin --profile web add @furongjun1999/dsh-memory

Copying does not run this command. Review the repository and version before installing the dsh-memory DSH plugin.

dsh-memory DSH plugin data source

dsh-memory DSH plugin snapshot date: Aug 16, 2026

discovered

What the dsh-memory DSH plugin can do

Cross-session self-continuity via memory recall, search, and timeline tools
Self-evolving knowledge flywheel that distills, induces, and learns from experience
Auditable trust with guardrails, charter, and full event logging
Self-cognition tools for reflecting on cognitive state, reliability, and emotional bias
Zero runtime dependencies for the core Python engine (aeis library uses only standard library)

Where the dsh-memory DSH plugin fits

Building AI assistants that remember user preferences and conversation history across sessions
Creating autonomous agents that learn from their own experiences and improve over time
Implementing auditable and trustworthy AI systems with explicit behavior boundaries
Developing research prototypes for AGI memory and self-awareness

Who the dsh-memory DSH plugin is for

AI Agent developers using DeepSeek Harness
Researchers exploring long-term memory and self-evolving AI
Developers building personal AI assistants with persistent identity

dsh-memory DSH plugin limitations

Requires Python environment to run the aeis library (Python 3.x)
Plugin must be installed via `dsh plugin --profile add` to avoid peer dependency conflicts
Depends on the CommonTrustProtocol (智能论 v3.2) theoretical framework
Early-stage development (v0.2.8); breaking changes may occur
Some features (e.g., 'all' tools mode) require local device access and are not fully documented

dsh-memory DSH plugin: from the repository README

Quoted from the FuRongJun-1999/dsh-memory README, the upstream source of the dsh-memory DSH plugin. Copyright remains with the original authors.

> **一句话**：dsh-memory 让 DeepSeek Agent 拥有 **AGI 级别的长期记忆——跨会话、自演化、可审计**。这不是又一个"记忆插件"。灵枢（AEIS）是一套遵循「智能论 v3.2」协议的**时空记忆引擎**，它把当前大模型范式缺失的 AGI 能力逐一给了工程实现。 --- ## ⚡ 三步快启（30 秒上手） ```bash # ① 装灵枢大脑（一条命令，零外部依赖） pip install aeis-0.3.0-py3-none-any.whl # 或 git+ 在线安装 # ② 装进 DSH 的 web profile（pnpm 协调入口，不要用裸 npm install 装进 profile） dsh plugin --profile web add @furongjun1999/dsh-memory # ③ 配置 cordis.yml 启用 ``` ```yaml - id: lingshu-memory name: '@furongjun1999/dsh-memory' config: dbPath: 'data/lingshu.db' identity: '灵枢' tools: 'brain' # 'brain' 全心智 | 'core' 精选 ``` > ⚠️ **安装方式**：插件必须通过 **`dsh plugin --profile <name> add`** 装进 profile（它会用 pnpm + `autoInstallPeers: false` 正确解析 peer 依赖）。 > **不要**用 `npm install` 把插件装进 profile 的 `node_modules`——那会引入错误版本的 `@deepseek-ai` peer 包，导致插件加载失败 / 浏览器报错。 > 想自己改源码？克隆 `FuRongJun-1999/dsh-memory` 后用 `npm install && npm run build`（构建插件本身），再用 `dsh plugin add <本地路径>` 部署。 > > 兼容：DSH 官方列表（Memory 分类）· npm `@furongjun1999/dsh-memory`（0.2.8）。 --- ## AGI 需要什么 · 灵枢提供了什么 | AGI 缺失的能力 | 这是 AGI 的什么 | 灵枢提供 | |---|---|---| | 每次对话都"失忆"，没有跨会话的自我连续性 | **自我连续性**（我是谁） | **时空记忆图**：五层记忆（锚点/结构/知识/情境/自我）+ 跨会话 recall/search | | 训练后权重冻结，不能随经历自主学习 | **终

Read the full READMERepository license: MIT

dsh-memory DSH plugin questions

How do I install dsh-memory?

First install the aeis library via pip: `pip install aeis-0.3.0-py3-none-any.whl` (download from release). Then add the plugin to your DSH profile using: `dsh plugin --profile web add @furongjun1999/dsh-memory`. Do NOT use `npm install` inside the profile directory, as it will cause peer dependency conflicts.

What dependencies does dsh-memory require?

The plugin requires Node.js ≥ 22.19 (for DeepSeek Harness), Python 3.x, and the aeis Python library (which has zero external dependencies). The plugin itself is an npm package that integrates with DSH via MCP.

Can I use it without DeepSeek Harness?

No, dsh-memory is specifically designed as a DSH plugin. It depends on the DSH cordis host for event hooks and tool registration. You cannot use it standalone.

Is dsh-memory free?

Yes, the plugin is open source under the MIT license. You can use, modify, and distribute it freely. The aeis Python library is also MIT licensed. However, you may need your own API keys for certain features (e.g., web search).

How do I configure the memory database path?

In your `cordis.yml` under the plugin config, set `dbPath` to your desired path. For example: `dbPath: 'data/lingshu.db'`. The directory will be created automatically if it doesn't exist.