Cortexgraph

CortexGraph is an MCP server that provides a human-like memory system for AI assistants. It realizes the natural decay and reinforcement use of memories by simulating the Ebbinghaus forgetting curve, including short-term memory storage, long-term memory archiving, and knowledge graph functions.

Knowledge management and memory Artificial intelligence chatbots #Memory System #Cognitive Architecture #AI Assistant #Knowledge Graph .Python

rating : 2.5 points

downloads : 7.4K

update time : 2025-12-12

Open Site

What is CortexGraph?

CortexGraph is a human-like memory system designed for AI assistants (such as Claude). It solves the problem of AI forgetting information during conversations, allowing AI to naturally remember your preferences, decisions, and important facts while letting less frequently used information fade away over time.

How to use CortexGraph?

Using CortexGraph is very simple: 1) Install the CortexGraph MCP server; 2) Configure it in Claude Desktop; 3) Have conversations with Claude as usual. The system will automatically detect the content to be remembered and recall relevant information at appropriate times. You don't need to use special commands, and everything is done in natural conversations.

Use Cases

CortexGraph is particularly suitable for the following scenarios: Personal AI assistants (remembering user preferences), development environments (remembering technical decisions), research assistants (retaining important findings in the long term), and customer service robots (remembering customer history). It makes AI conversations more personalized and coherent.

Main Features

Human-like Memory Decay

Based on the Ebbinghaus forgetting curve, memories naturally decay over time unless reinforced through use. Less frequently used information will gradually fade away, while frequently used information will be retained for a longer time.

Automatic Memory Reinforcement

When an AI references a memory during a conversation, that memory is automatically reinforced, and the decay rate slows down. This simulates the 'use reinforcement' mechanism of human memory.

Two-layer Storage Architecture

Short-term memory (STM) stores recent information, and long-term memory (LTM) permanently stores important information. Frequently used memories are automatically promoted from STM to LTM.

Natural Language Activation

No special commands are required. The system will automatically detect the content to be saved or recalled from natural conversations. Say 'Remember: I like the dark mode' or 'What did I say before?' to trigger the corresponding functions.

Multi-agent Integration

Five dedicated agents automatically maintain memories: analyzing decay risks, clustering similar memories, merging duplicate content, promoting important memories, and discovering cross-domain associations.

Local Privacy Protection

All data is stored on the local device, supporting JSONL and SQLite formats. Long-term memories can be exported as Markdown files, which are compatible with note-taking applications such as Obsidian.

Visualized Knowledge Graph

It provides an interactive network graph visualization to show the relationships between memories, supporting zooming, dragging, and viewing detailed information.

Advantages

🤖 Makes AI conversations more coherent and personalized

🧠 A memory model based on cognitive science, more in line with human intuition

🔒 Completely local storage to protect privacy and security

⚡ Automatically runs without manual memory management

📊 Adjustable memory parameters to adapt to different use scenarios

🔄 Well integrated with existing tools (Obsidian, Git)

Limitations

⚠️ A research prototype, not recommended for production environments

📚 Requires certain technical knowledge for initial configuration

🖥️ Mainly developed for macOS, Windows/Linux support needs testing

🧪 The memory algorithm is still being optimized, and there may be instability

🔧 Requires a client that supports MCP, such as Claude Desktop

How to Use

Install CortexGraph

Install using the UV tool (recommended) or via pip. UV installation provides better isolation.

Create a Configuration File

Create a.env file in the configuration directory and set the storage path, decay parameters, etc. You can copy the example file and modify it.

Configure Claude Desktop

Edit the configuration file of Claude Desktop and add the CortexGraph MCP server. You need to use the absolute path to ensure that the GUI application can find the command.

Restart and Start Using

Restart Claude Desktop and start having normal conversations. The system will automatically detect the content to be remembered and recall it at appropriate times.

Maintenance and Monitoring

Use the CLI tool to view the memory status and perform maintenance operations. Regularly check the memory health.