mfcs-memory 0.1.1

A smart conversation memory management system

MFCS Memory

English | 中文

MFCS Memory is an intelligent conversation memory management system that helps AI assistants remember conversation history with users and dynamically adjust response strategies based on conversation content.

Key Features

• Intelligent Conversation Memory: Automatically analyzes and summarizes user characteristics and preferences
• Vector Storage: Uses Qdrant for efficient similar conversation retrieval
• Session Management: Supports multi-user, multi-session management
• Automatic Chunking: Automatically creates chunks when conversation history exceeds threshold
• Async Support: All operations support asynchronous execution
• Extensibility: Modular design, easy to extend and customize

Core Features

MemoryManager Core Methods

1. get(user_id: str, query: Optional[str] = None, top_k: int = 2) -> str

   • Get current session information for specified user
   • Includes conversation summary and user memory summary
   • Supports query-based relevant historical conversation retrieval
   • Returns formatted memory information

2. update(user_id: str, user_input: str, assistant_response: str) -> bool

   • Automatically gets or creates current session for user
   • Updates conversation history
   • Automatically updates user memory summary every 3 rounds
   • Automatically updates session summary every 5 rounds
   • Automatically handles conversation chunking

3. delete(user_id: str) -> bool

   • Deletes all data for specified user
   • Includes session data and vector storage data
   • Returns whether operation was successful

4. reset() -> bool

   • Resets all user records
   • Clears all session data and vector storage
   • Returns whether operation was successful

Installation

1. Install the package:

pip install mfcs-memory

2. Install SentenceTransformer for text embedding:

pip install sentence-transformers

Note: The default embedding model is BAAI/bge-large-zh-v1.5. You can change it in the configuration.

Quick Start

1. First, create a .env file and configure necessary environment variables:

# MongoDB Configuration
MONGO_USER=your_username
MONGO_PASSWD=your_password
MONGO_HOST=localhost:27017

# Qdrant Configuration
QDRANT_HOST=localhost
QDRANT_PORT=6333

# Model Configuration
EMBEDDING_MODEL_PATH=./model/BAAI/bge-large-zh-v1.5
EMBEDDING_DIM=768
LLM_MODEL=qwen-plus-latest  # Default value

# OpenAI Configuration
OPENAI_API_KEY=your_api_key

# Other Configuration
MONGO_REPLSET=''  # Optional, if using replica set
MAX_RECENT_HISTORY=20  # Default value
CHUNK_SIZE=100  # Default value
MAX_CONCURRENT_ANALYSIS=3  # Default value

2. Usage Example:

import asyncio
from mfcs_memory.utils.config import Config
from mfcs_memory.core.memory_manager import MemoryManager

async def main():
    # Load configuration
    config = Config.from_env()
    
    # Initialize memory manager
    memory_manager = MemoryManager(config)
    
    # Update conversation
    await memory_manager.update(
        "user123",
        "Hello, I want to learn about Python programming",
        "Python is a simple yet powerful programming language..."
    )
    
    # Get memory information
    memory_info = await memory_manager.get(
        "user123",
        query="How to start Python programming?",
        top_k=2
    )
    
    # Delete user data
    await memory_manager.delete("user123")
    
    # Reset all data
    await memory_manager.reset()

if __name__ == "__main__":
    asyncio.run(main())

Project Structure

src/
├── mfcs_memory/
│   ├── core/
│   │   ├── base.py              # Base
│   │   ├── memory_manager.py    # Memory Manager
│   │   ├── session_manager.py   # Session Manager
│   │   ├── vector_store.py      # Vector Store
│   │   └── conversation_analyzer.py  # Conversation Analyzer
│   ├── utils/
│   │   └── config.py           # Configuration Management
│   └── __init__.py
├── example/                    # Example Code
├── model/                      # model Directory
├── setup.py                    # Installation Configuration
└── README.md                   # Project Documentation

Configuration Guide

Required Configuration

• MONGO_USER: MongoDB username
• MONGO_PASSWD: MongoDB password
• MONGO_HOST: MongoDB host address
• QDRANT_HOST: Qdrant host address
• EMBEDDING_MODEL_PATH: Model path for generating text vectors
• EMBEDDING_DIM: Vector dimension
• OPENAI_API_KEY: OpenAI API key
• OPENAI_API_BASE: OpenAI API Base

Optional Configuration

• MONGO_REPLSET: MongoDB replica set name (if using replica set)
• QDRANT_PORT: Qdrant port number (default: 6333)
• LLM_MODEL: LLM model name (default: qwen-plus-latest)
• MAX_RECENT_HISTORY: Number of recent conversations kept in main table (default: 20)
• CHUNK_SIZE: Number of conversations stored in each chunk (default: 100)
• MAX_CONCURRENT_ANALYSIS: Maximum number of concurrent analysis tasks (default: 3)

Contributing

Issues and Pull Requests are welcome!

License

MIT License