The core philosophy of DeepSeek Harness (DSH) is “everything is a plugin.” When developing AI Agents, contextual memory is often lost after a session ends or DSH restarts. The Mnemosyne plugin provides DSH with cross-session long-term memory capabilities through vector and LLM reflection mechanisms. The project is maintained by Witchwarren2344, licensed under MIT, and is completely free.

Core Features

This plugin provides the following capabilities:

  • Permanent memory: Stores memory persistently, supporting no loss across sessions or restarts.
  • Vector semantic search: Performs semantic understanding and retrieval based on vectors.
  • LLM reflection: Automatically extracts decisions, insights, and conventions from sessions.
  • Automatic knowledge page generation: Automatically generates architecture diagrams, convention checklists, and project summaries.
  • Code mapping: Automatically identifies and indexes 30+ configuration file patterns.
  • Cross-session retrospection: Imports historical sessions to inherit existing knowledge.
  • Multi-Workspace support: Supports project-isolated memory, as well as team sharing.
  • Delta refresh: Updates only changed pages, improving synchronization efficiency.

Installation and Activation

The installation command is:

./scripts/install.sh

After installation, it is recommended to run the health check:

dsh doctor

Usage Examples

  1. Pull the embedding model (local deployment mode):
    ollama pull nomic-embed-text

  1. Store memory:
    Use the mnemo_store command in DSH.

  2. Recall memory:
    Use the mnemo_recall command in DSH.

Deployment and Cost

This plugin is completely free and licensed under MIT.

  • Local deployment: Uses Ollama, fully offline and zero-cost.
  • Cloud deployment: Uses the Gemini or DeepSeek API as an optional upgrade. The Gemini API provides 1500 free embedding requests per month.

Notes

  • The plugin runs with DSH process privileges; ensure that you trust the installed source code.
  • Before use, choose between a local embedding model or a cloud API configuration based on your actual needs.

Summary

For DSH Agent developers who require long-term memory capabilities, Mnemosyne provides a complete solution from vector retrieval to knowledge page generation, serving as a foundational component for building persistent agents.

GitHub | Directory