Knowledge Graph Memory Server logo

Knowledge Graph Memory Server

by T1nker-1220

A basic implementation of persistent memory using a local knowledge graph. This lets Claude remember information about the user across chats and learn from past errors through a lesson system.

View on GitHub

Last updated: N/A

What is Knowledge Graph Memory Server?

This server is a persistent memory system built on a local knowledge graph. It allows language models like Claude to remember user information across chats and learn from past errors through a lesson system.

How to use Knowledge Graph Memory Server?

The server provides an API with tools to create entities, relations, add/delete observations, manage lessons, and search the knowledge graph. It can be integrated with Cursor MCP client or used with Claude Desktop by configuring the claude_desktop_config.json file. The system prompt can be customized to define how the model utilizes the memory.

Key features of Knowledge Graph Memory Server

  • Persistent memory using a local knowledge graph

  • Entity and relation management

  • Observation storage and retrieval

  • Lesson system for learning from errors

  • API for interacting with the knowledge graph

  • File management for memory and lesson data

  • Integration with Cursor MCP client and Claude Desktop

Use cases of Knowledge Graph Memory Server

  • Personalized chatbot experiences

  • Context-aware AI assistants

  • Error tracking and resolution

  • Adaptive learning systems

FAQ from Knowledge Graph Memory Server

What are entities?

Entities are the primary nodes in the knowledge graph, representing objects, people, or concepts. Each entity has a unique name, entity type, and a list of observations.

What are relations?

Relations define directed connections between entities, describing how they interact or relate to each other. They are always stored in active voice.

What are observations?

Observations are discrete pieces of information about an entity, stored as strings and attached to specific entities. They should be atomic, containing only one fact per observation.

What are lessons?

Lessons are special entities that capture knowledge about errors and their solutions. They include error pattern information, solution steps, success rate tracking, and environmental context.

How do I integrate this with Claude Desktop?

Add the appropriate configuration (Docker or NPX) to your claude_desktop_config.json file, specifying the command and arguments to run the server.