markdown

🧠 CONTINUUM AI

Persistent Memory RAG Chatbot That Never Forgets

---

✨ Overview

Continuum AI is a conversational AI assistant with persistent memory that never forgets. Unlike traditional chatbots that lose context after each session, Continuum remembers facts about you across conversations, automatically decays old memories (just like humans!), and retrieves relevant information using RAG (Retrieval-Augmented Generation).

🎯 Why Continuum AI?

Feature	Traditional Chatbots	Continuum AI
Memory Across Sessions	❌ No	✅ Yes
Automatic Fact Extraction	❌ No	✅ Yes
Memory Decay (Forgetting)	❌ No	✅ Yes
Reinforcement Learning	❌ No	✅ Yes
Local LLM (No API Keys)	❌ No	✅ Yes
Persistent Storage	❌ No	✅ Yes

---

🚀 Live Demo

🔗 Try it now: continuum-rag-chatbot.streamlit.app

Note: First load may take 3–5 minutes as the model downloads (~2.3GB). Subsequent loads are instant!

---

🌟 Key Features

🧠 Persistent Memory

• Stores memories in ChromaDB vector database
• Remembers facts across multiple conversations
• Never forgets what you tell it

📉 Ebbinghaus Memory Decay

• Memories naturally fade over time (just like humans!)
• Configurable decay rate (λ = 0.01 to 0.5)
• Frequently accessed memories get reinforced

🔍 RAG (Retrieval-Augmented Generation)

• Retrieves relevant memories before responding
• Uses cosine similarity for semantic search
• Top-K configurable (1–10 memories)

💪 Reinforcement Learning

• Strengthens memories when accessed
• strength = min(1.0, old_strength + 0.3)
• Frequently discussed topics stay strong

🔧 100% Free & Local

• No API keys required
• Runs Microsoft Phi-3-mini locally
• All processing on your machine

🎨 Beautiful UI

• Dark theme with neon accents
• Real-time memory statistics
• Streaming responses
• Mobile responsive

---

🛠️ Technology Stack

Component	Technology	Purpose
Frontend	Streamlit	Web UI framework
LLM	Microsoft Phi-3-mini (3.8B params)	Text generation
Embeddings	sentence-transformers (all-MiniLM-L6-v2)	Semantic search
Vector DB	ChromaDB	Persistent memory storage
Framework	PyTorch / Transformers	Model inference
Language	Python 3.10+	Core logic

---

📦 Installation

Prerequisites

• Python 3.10 or higher
• 8GB+ RAM (16GB recommended)
• 5GB free disk space

Local Setup

# 1. Clone the repository
git clone https://github.com/Junaid-Ahmed-Rupok/continuum-rag-chatbot.git
cd continuum-rag-chatbot

# 2. Create virtual environment
python -m venv venv

# 3. Activate virtual environment
# On Windows:
venv\Scripts\activate
# On Mac/Linux:
source venv/bin/activate

# 4. Install dependencies
pip install -r requirements.txt

# 5. Run the app
streamlit run app.py

---

🎮 Usage Guide

Basic Commands

Action	How To
Send message	Type in chat box → Press Enter
Reset all memory	Click "🗑️ Reset Memory"
Export memories	Click "💾 Export Memory"
Adjust Top-K	Settings panel → Slider
Change decay rate	Settings panel → Slider

Example Conversations

User: "Hi, my name is Alex"
Continuum: "Nice to meet you, Alex! How can I help you today?"
[Memory saved: "User's name is Alex"]

--- Next Session ---

User: "What's my name?"
Continuum: "Your name is Alex! You told me that in our previous conversation."
[Memory retrieved and reinforced]

Memory Management Tips

1. Reinforce important memories by discussing them frequently
2. Use the export feature to back up your memories
3. Adjust decay rate:
   • Low (0.05): Very slow forgetting
   • Medium (0.1): Default, natural forgetting
   • High (0.5): Fast forgetting, short-term focus

---

🏗️ Architecture

┌─────────────────────────────────────────────────────────────┐
│                       USER INTERFACE                        │
│                      (Streamlit UI)                         │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│                      QUERY PROCESSING                       │
│  ┌─────────────┐  ┌─────────────┐  ┌─────────────────────┐ │
│  │  Decay All  │→ │  Retrieve   │→ │    Reinforce        │ │
│  │  Memories   │  │  Memories   │  │  Retrieved Ones     │ │
│  └─────────────┘  └─────────────┘  └─────────────────────┘ │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│                      MEMORY SYSTEM                          │
│  ┌─────────────┐  ┌─────────────┐  ┌─────────────────────┐ │
│  │  ChromaDB   │←→│  Embeddings │←→│  Sentence           │ │
│  │  Vector DB  │  │  (384-dim)  │  │  Transformer        │ │
│  └─────────────┘  └─────────────┘  └─────────────────────┘ │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│                         LLM LAYER                           │
│  ┌─────────────────────────────────────────────────────┐   │
│  │           Phi-3-mini (3.8B parameters)              │   │
│  │         Context: System + Memories + History        │   │
│  └─────────────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│                      POST PROCESSING                        │
│  ┌─────────────┐  ┌─────────────────────────────────────┐  │
│  │  Save to    │  │  Extract Facts & Store to Memory    │  │
│  │  Buffer     │  │                                     │  │
│  └─────────────┘  └─────────────────────────────────────┘  │
└─────────────────────────────────────────────────────────────┘

---

📊 Memory Decay Formula

Continuum implements the Ebbinghaus Forgetting Curve:

new_strength = old_strength × e^(-λ × days_since_reinforcement)

Where:
- λ = decay rate (default: 0.1)
- days_since = time since last access
- min_strength = pruning threshold (default: 0.05)

Decay Visualization

Days	λ=0.05	λ=0.1	λ=0.2	λ=0.5
0	1.00	1.00	1.00	1.00
7	0.70	0.50	0.25	0.03
14	0.50	0.25	0.06	0.001
30	0.22	0.05	0.002	0.000

---

🔧 Configuration

Environment Variables

Variable	Description	Default
TOP_K	Number of memories to retrieve	5
TEMPERATURE	LLM creativity (0.0–1.0)	0.7
DECAY_LAMBDA	Memory decay rate	0.1
CONTINUUM_DATA_DIR	Data storage location	./continuum_data

Streamlit Configuration (.streamlit/config.toml)

[theme]
base = "dark"
primaryColor = "#00e5ff"
backgroundColor = "#0a0e1a"
secondaryBackgroundColor = "#151c2c"
textColor = "#f0f3ff"

---

🧪 Testing

python -m pytest tests/

Test coverage includes:

• ✅ Directory structure validation
• ✅ ChromaDB read/write operations
• ✅ Embedding dimension verification
• ✅ Decay formula accuracy
• ✅ Top-K retrieval respect
• ✅ Reinforcement functionality
• ✅ Fact extraction accuracy
• ✅ JSON export validity

---

📁 Project Structure

continuum-rag-chatbot/
│
├── app.py                      # Main Streamlit application
├── requirements.txt            # Python dependencies
├── README.md                   # This file
├── .gitignore                  # Git ignore rules
│
├── .streamlit/
│   └── config.toml             # Streamlit configuration
│
├── utils/
│   ├── __init__.py             # Package exports
│   ├── config.py               # Configuration management
│   ├── memory.py               # RAGMemory core class
│   ├── conversation.py         # Conversation buffer
│   └── helpers.py              # Fact extraction utilities
│
└── continuum_data/             # Auto-created on first run
    ├── db/                     # ChromaDB storage
    ├── exports/                # JSON exports
    └── config/                 # Session statistics

---

🐛 Troubleshooting

Issue	Solution
Model won't load	Check internet connection, restart Streamlit
Slow responses	Streamlit Cloud has limited CPU — run locally for better speed
Memory not persisting	Check continuum_data/ folder permissions
Import errors	Run pip install -r requirements.txt
Port already in use	streamlit run app.py --server.port 8502

Getting Help

• 📧 Open an Issue
• ⭐ Star the repo for updates
• 🔔 Watch for new releases

---

🗺️ Roadmap

Version 1.0 ✅ (Current)

• Persistent memory with ChromaDB
• Ebbinghaus decay curves
• Fact extraction
• Streamlit UI
• Local Phi-3-mini integration

Version 1.1 (Planned)

• Multi-user support
• Memory categories/tags
• Export to CSV/PDF
• Voice input support

Version 2.0 (Exploring)

• Fine-tuning on user conversations
• Multi-modal support (images)
• Web search integration
• Mobile app

---

🤝 Contributing

1. Fork the repository
2. Create a feature branch: git checkout -b feature/amazing-feature
3. Commit changes: git commit -m 'Add amazing feature'
4. Push to branch: git push origin feature/amazing-feature
5. Open a Pull Request

Development Setup

git clone https://github.com/YOUR_USERNAME/continuum-rag-chatbot.git
cd continuum-rag-chatbot
pip install -r requirements.txt
pip install black pytest
black .
pytest

---

📄 License

Distributed under the MIT License. See LICENSE for details.

---

🙏 Acknowledgments

• Microsoft for Phi-3-mini
• Hugging Face for the Transformers library
• ChromaDB for vector database
• Streamlit for the UI framework
• Sentence Transformers for embeddings

---

📞 Contact

Junaid Ahmed Rupok

• GitHub: @Junaid-Ahmed-Rupok
• Project: continuum-rag-chatbot
• Live Demo: continuum-rag-chatbot.streamlit.app

---

Built with 🧠 by Junaid Ahmed Rupok

Never forget. Always remember.

⭐ If you found this helpful, give it a star on GitHub!

```