Agent 2 – Chatbot with Persistent Memory (LangGraph + LangChain) A conversational AI agent that maintains full conversation history using structured message types (HumanMessage, AIMessage) and demonstrates how to persist memory beyond runtime using files or databases. This project is a foundational step toward stateful, memory-enabled AI agents.

🚀 Objectives

✅ Use different message types:

HumanMessage

AIMessage

✅ Maintain full conversation history

✅ Build the agent using LangGraph

✅ Use LLM via LangChain

ChatGoogleGenerativeAI (Gemini)

(Can be swapped with ChatOpenAI – GPT-4o)

✅ Create a memory mechanism for the agent

🧩 Tech Stack

Python

LangChain

LangGraph

Google Gemini Pro

dotenv

TypedDict for state management

📂 Project Structure ├── agent.py ├── .env ├── logging.txt ├── README.md

🔐 Environment Setup Create a .env file in the root directory: GEMINI_API_KEY=your_google_api_key_here

🧠 How Memory Works The agent stores conversation as: List[Union[HumanMessage, AIMessage]]

Each turn:

User input → HumanMessage

LLM response → AIMessage

Messages appended to state

Full history passed back to the model

This allows the agent to remember past context during the session.

⚙️ Core Agent Logic Agent State Definition class AgentState(TypedDict): messages: List[Union[HumanMessage, AIMessage]]

Processing Node (Memory Update) def process(state: AgentState) -> AgentState: response = llm.invoke(state["messages"]) state["messages"].append(AIMessage(content=response.content)) return state

This node:

Receives full conversation history

Generates a response

Appends it back to memory

LangGraph Flow START → process → END

💬 Running the Chatbot python agent.py

Example: Enter: Hi AI: Hello! How can I help you today? Enter: What did I just say? AI: You said "Hi".

Type exit to end the conversation.

📝 Conversation Logging At the end of the session, the full conversation is saved to: logging.txt

Format: You : ... AI : ...

❗ Current Limitation ❌ Memory gets wiped after program exits Because memory is stored in-memory (RAM), it resets when the script stops.

✅ Solutions for Persistent Memory To prevent memory loss, you can store conversation data in: 1️⃣ Text Files / Markdown Append messages after every turn Reload on startup 2️⃣ Database (SQL / NoSQL)

SQLite PostgreSQL MongoDB

3️⃣ Vector Database (Best for Long-Term Memory)

ChromaDB FAISS Pinecone Weaviate This allows: Semantic recall Long-term memory

RAG-based agents