Introduction

DeepSeek Harness (DSH) adopts a plugin-based architecture, making it easy to extend functionality. In practical AI Agent development, context loss after session interruption is a common issue. Every time a new conversation starts, background information must be repeated, which reduces efficiency. dsh-mnemosyne-memory aims to solve this problem by providing DSH with cross-session long-term memory capabilities.

Plugin Overview

The plugin is named Mnemosyne, maintained by user fjzzwxp, and belongs to the “memory” category. It uses the MIT license and is completely free. Core features include cross-session persistent storage, vector semantic search, and an LLM reflection mechanism, helping agents inherit historical knowledge after restarts.

Core Features

The plugin provides the following capabilities:

  • Permanent memory (cross-session persistence): data is stored persistently and remains available across sessions and restarts.
  • Vector semantic search: supports vector semantic search and can understand natural language queries.
  • LLM Reflection: automatically extracts decisions, insights, and conventions from sessions.
  • Automatic knowledge page generation: generates architecture diagrams, convention lists, and project summaries.
  • Codebase Survey: identifies 30+ configuration file patterns and indexes them automatically.
  • Cross-session backtracking: imports historical sessions to inherit existing knowledge.
  • Multi-workspace isolation: isolates data by project and supports shared team memory.
  • Delta incremental refresh: updates only changed pages for efficient synchronization.

Installation and Enablement

Before installation, ensure the environment meets the following requirements: Node.js >= 18.0.0, DSH (DeepSeek Harness) >= 0.1.0-rc.7, and Git.

# 1. 克隆仓库
git clone https://github.com/fjzzwxp/dsh-mnemosyne-memory.git
cd dsh-mnemosyne-memory

# 2. 安装依赖
npm install

# 3. 执行安装脚本
./scripts/install.sh

# 4. 验证安装
dsh plugin list

Deployment Modes

The plugin supports two modes: local deployment and cloud API.

Local Deployment

In local deployment, data does not leave the device, making it suitable for privacy-sensitive or offline environments. At least 8GB of RAM is recommended.

  1. Install Ollama:
    # macOS
    brew install ollama
  1. Pull the embedding model:
    ollama pull nomic-embed-text
  1. Configure config/mnemosyne.json and set the provider to ollama.

Cloud Deployment

Cloud mode requires an internet connection, and data will be uploaded. Google Gemini is recommended, providing 1,500 free embedding requests per month.

  1. Visit Google AI Studio to create an API Key.
  2. Copy the configuration template and fill in the Key:
    cp config/mnemosyne.json.example config/mnemosyne.json
    nano config/mnemosyne.json

Typical Usage

After installation, you can use the following commands:

  • Run health check: dsh doctor
  • Store memory: mnemo_store
  • Recall memory: mnemo_recall

Notes

  • The plugin runs with the permissions of the current DSH process. It is recommended to inspect the source code before installation.
  • When using cloud mode, pay attention to privacy compliance.
  • The plugin is completely free and open-sourced on GitHub.

Summary

dsh-mnemosyne-memory provides a comprehensive long-term memory solution for the DSH ecosystem. Through vector retrieval and reflection mechanisms, it helps agents build coherent conversation history and knowledge bases. For application scenarios that require complex context management, this plugin is a necessary supplement.