Memgpt Agent

🚀 MemGPT Sample Project

This project is a proof-of-concept implementation that simplifies and abbreviates the concepts in the MemGPT research paper. It is implemented in the form of an agent and demonstrates how to integrate it as a simple REST service or as an MCP server.

🚀 Quick Start

This project implements the MemGPT concept using Spring Boot and Spring AI, and exposes its functionality through simple REST interfaces and an MCP server. Additionally, it includes a command-line-based ChatBot application that can integrate with the REST or MCP interfaces. The project consists of the following sub - projects:

• core - agent: Defines and implements the MemGPT interface, including the configuration API.
• restful - agent - app: A wrapper service built around the agent, exposing its functionality through a simple REST interface.
• memgpt - mcp - server: A wrapper service built around the agent, exposing its functionality as an MCP server.
• simple - chat - bot - app: A command - line - based chat application that can communicate with the agent via the REST or MCP interface.

✨ Features

Prerequisites

• Java 17 or higher
• Gradle
• OpenAI API key

Core Agent

The core agent is the component that contains the MemGPT interface and implementation. It is responsible for maintaining the agent's configuration and state, including:

• System prompts
• Tools
• Core memory
• Archive memory
• Message persistence (including messages outside the context window)

It is also responsible for core agent functions, including but not limited to:

• Message compilation
• LLM communication
• Tool invocation
• Memory management (including self - editing of the context window)
• Message retrieval and archiving
• Message summarization

Externally, the agent exposes its functionality through the following three main interfaces:

• AgentManager: Configures and maintains agent instances.
• AgentLoader: Creates agent instances.
• Agent: Defines the core agent chat functionality.

📦 Installation

Running the Application

The simple chatbot application communicates with either the restful - agent - app or the memgpt - mcp - server. Choose the server application you want the chatbot application to use, configure and run it. Then configure the ChatBot application to use the selected server application and run it.

Server Applications

Both the MCP server and the REST server are responsible for handling the context window and agent logic. Here are the two main implementations:

1. MCP Server (REST - based)

   • Implements the ModelContextProtocol interface.
   • Provides the /api/v1/chat/completions endpoint.
   • Supports compatibility with OpenAI's API.

2. REST Server

   • Implements ChatMessageStore and ChatContext.
   • Provides a REST - based interface for managing chat sessions and context.

Running Steps

1. MCP Server (REST - based)

   ./gradlew :memgpt - mcp - server:run
   

   • After starting, it runs on http://localhost:8080 by default.
   • You can modify the port and other settings through the configuration file.

2. REST Server

   ./gradlew :restful - agent - app:run
   

   • After starting, it runs on http://localhost:9090 by default.
   • Provides a REST interface for agent management.

Running the ChatBot Application

1. Build the Application

   ./gradlew :simple - chat - bot - app:build
   

2. Run the ChatBot Application

   java -jar ./simple - chat - bot - app/build/libs/simple - chat - bot - app - 0.0.1 - SNAPSHOT.jar
   

   • After starting, it will prompt you to enter a user ID.
   • The user ID is used to identify the chat session and can be any string.

By following the above steps, you can easily run the MemGPT sample project and experience its functionality and integration methods.