🧠 Agentic Long-Term Memory

This project explores the concept of long-term memory in AI agents and demonstrates how to build agentic memory systems using various approaches and tools.

Feel free to ⭐️ the repo if it helps you understand long-term memory for agents!

📚 What You'll Learn

1. How to design a custom agentic long-term memory.
2. How to implement long-term memory using Vector DB and Graph DB with LangChain and LangGraph.
3. How to structure memory into semantic, episodic, and procedural components (based on the LangChain CEO's course on DeepLearning.ai).
4. A brief overview of letta's long-term memory model.

🔗 A great reference paper for this project:
Memory in LLM Agents
PDF available in the repo: letta/MemGPT_Paper.pdf

▶️ Watch the Demo

📺 YouTube Video: Link

💾 How to Run the Project

⚠️ Note: Please use Python 3.11 as Python 3.12 may cause compatibility issues with some dependencies.

1. Create a virtual environment

   python -m venv venv
   source venv/bin/activate  # or venv\Scripts\activate on Windows

2. Install dependencies

   pip install -r requirements.txt

3. Prepare the databases for the custom chatbot

   python src/prepare_sqldb.py          # Setup SQLite DB  
   python src/check_sqldb.py            # Inspect DB contents  
   python src/prepare_vectordb.py       # Setup Vector DB  
   python src/check_vectordb.py         # Inspect Vector DB 

4. Run the chatbots
   • Run in terminal:

     python src/bot.py

   • Run with Gradio UI (all 3 chatbot versions available):

     python src/bot_ui.py

Project Schemas:

LLM Default Behavior

Concept of Memory

Basic Chatbot Schema

Agentic_Chatbot_v2 Schema

Agentic_Chatbot_v3 Schema

Longterm Memory with Graph DB and Vector DB using LangGraph

Longterm Memory (Semantic, Episodical, and Procedural) with LangGraph

🧩 Technologies Used

• Python
• openai
• LangChain
• LangGraph
• SQLite
• Chroma (Vector DBs)
• Gradio (UI)

📂 Project Structure (High-Level)

src/
├── bot.py                     # Terminal-based chatbot
├── bot_ui.py                  # Gradio UI version
├── prepare_sqldb.py           # Creates SQLite DB
├── prepare_vectordb.py        # Creates Vector DB
├── check_sqldb.py             # Checks SQLite DB contents
├── check_vectordb.py          # Checks Vector DB contents
└── utils/
    ├── chat_history_manager.py
    ├── chatbot_agentic_v1.py
    ├── chatbot_agentic_v2.py
    ├── chatbot.py
    ├── config.py
    ├── prepare_system_prompt.py
    ├── search_manager.py
    ├── sql_manager.py
    ├── user_manager.py
    ├── utils.py
    └── vector_db_manager.py

langgraph/
├── online-course/             # Memory system from DeepLearning.ai course
└── website-tutorial/          # Tutorial-based memory system

letta/
├── chatbot_with_longterm_memory.py
└── MemGPT_Paper.pdf           # Research paper reference

data/
├── vectordb/
└── chatbot.db

images/

├── requirements.txt

src Folder Structure

Credits

This project is based on the excellent work and educational content created by Farzad Roozitalab.

• Original Repository: https://github.com/Farzad-R/Agentic-LongTerm-Memory
• Tutorial Video: https://www.youtube.com/watch?v=jw67V_gBzR0&list=WL&index=2

A special thanks to Farzad Roozitalab for open-sourcing the implementation and creating such valuable educational content on AI memory systems, conversational agents, and long-term memory architectures. This repository was created as a personal learning project while studying and reproducing the concepts demonstrated in the original repository and tutorial.

All core concepts, architecture, and initial implementation credit belong to the original author.