LLM Chat with MongoDB Memory

A chatbot web application with long-term memory capabilities using MongoDB. Persistently stores conversations with optimized retrieval using indexes on agent_id, timestamp, and tags. Removes duplicate sentences and combines recency with importance for smart memory retrieval.

Screenshots

Web Application Interface

MongoDB Compass - Memories Collection

MongoDB Document Details

---

How to Run

Requirement	Details
Python	Flask
Database	MongoDB (mongodb://localhost:27017)
AI Libraries	Hugging Face transformers + PyTorch

# Start Flask app
python app.py

# Access chat UI at http://localhost:5000

# (Optional) Inspect memories
python inspect_memories.py

---

Workflow

Step	Action
1	User sends message → Stored in MongoDB
2	System retrieves relevant memories
3	LLM generates response using prompt + context
4	AI response stored in memory
5	Response displayed in chat UI

---

Core Components

Flask Web Application (app.py)

• Routes: / (chat interface), /chat (POST - processes messages)
• Handles HTTP requests/responses

AI Model (Microsoft Phi-2)

• Uses Hugging Face transformers
• Parameters: temperature=0.7, top_k=50, repetition_penalty=1.2

Long-Term Memory System

• Storage: MongoDB for persistent conversations
• Features: Memory indexing, deduplication, relevance scoring
• Tracking: last_accessed, access_count, updated_at

Memory Inspection (inspect_memories.py)

• Counts total memories, user inputs, AI responses
• Displays latest 3 memories with timestamps and importance

---

Key Features

• Memory-Augmented Chat: Stores user messages and AI responses
• Smart Retrieval: Text search + regex fallback with recency/importance weighting
• Web Interface: Chat bubbles, context usage indicators, real-time messaging