mem-llm 1.0.3

Memory-enabled AI assistant with local LLM support

🧠 Mem-Agent: Memory-Enabled Mini Assistant

A local AI assistant that remembers user interactions and responds with context awareness using a lightweight 4-billion parameter LLM.

Quick Start • Features • Documentation • Examples

---

🎯 Why Mem-Agent?

Most Large Language Models (LLMs) treat every conversation as "new" and don't remember past interactions. Mem-Agent uses a small locally-running model to:

• ✅ Remember user history - Separate memory for each customer/user
• ✅ Context awareness - Responds based on previous conversations
• ✅ Fully local - No internet connection required
• ✅ Lightweight & fast - Only 2.5 GB model size
• ✅ Easy integration - Get started with 3 lines of code

🚀 Quick Start

1. Install Ollama

# Windows/Mac/Linux: https://ollama.ai/download
curl https://ollama.ai/install.sh | sh

# Start the service
ollama serve

2. Download Model

ollama pull granite4:tiny-h

3. Use Mem-Agent

from mem_llm import MemAgent

# Create agent
agent = MemAgent(model="granite4:tiny-h")

# System check
status = agent.check_setup()
if status['status'] == 'ready':
    print("✅ System ready!")
else:
    print("❌ Error:", status)

# Set user
agent.set_user("user123")

# First conversation
response = agent.chat("Hello, my name is Ali")
print(response)

# Second conversation - It remembers me!
response = agent.chat("Do you remember my name?")
print(response)

📚 Example Scripts

1. Simple Test

python examples/example_simple.py

2. Customer Service Simulation

python examples/example_customer_service.py

🏗️ Project Structure

Memory LLM/
├── memory_llm/              # Main package
│   ├── __init__.py          # Package initialization
│   ├── mem_agent.py         # Main assistant class
│   ├── memory_manager.py    # Memory management
│   ├── memory_db.py         # SQL database support
│   ├── llm_client.py        # Ollama integration
│   ├── memory_tools.py      # User tools
│   ├── knowledge_loader.py  # Knowledge base loader
│   ├── prompt_templates.py  # Prompt templates
│   └── config_manager.py    # Configuration manager
├── examples/                # Example scripts
├── tests/                   # Test files
├── setup.py                 # Installation script
├── requirements.txt         # Dependencies
└── README.md               # This file

🔧 API Usage

MemAgent Class

from mem_llm import MemAgent

agent = MemAgent(
    model="granite4:tiny-h",           # Ollama model name
    memory_dir="memories",             # Memory directory
    ollama_url="http://localhost:11434" # Ollama API URL
)

Basic Methods

# Set user
agent.set_user("user_id")

# Chat
response = agent.chat(
    message="Hello",
    user_id="optional_user_id",  # If set_user not used
    metadata={"key": "value"}     # Additional information
)

# Get memory summary
summary = agent.memory_manager.get_summary("user_id")

# Search in history
results = agent.search_user_history("keyword", "user_id")

# Update profile
agent.update_user_info({
    "name": "Ali",
    "preferences": {"language": "en"}
})

# Get statistics
stats = agent.get_statistics()

# Export memory
json_data = agent.export_memory("user_id")

# Clear memory (WARNING!)
agent.clear_user_memory("user_id", confirm=True)

MemoryManager Class

from mem_llm import MemoryManager

memory = MemoryManager(memory_dir="memories")

# Load memory
data = memory.load_memory("user_id")

# Add interaction
memory.add_interaction(
    user_id="user_id",
    user_message="Hello",
    bot_response="Hello! How can I help you?",
    metadata={"timestamp": "2025-01-13"}
)

# Get recent conversations
recent = memory.get_recent_conversations("user_id", limit=5)

# Search
results = memory.search_memory("user_id", "order")

OllamaClient Class

from mem_llm import OllamaClient

client = OllamaClient(model="granite4:tiny-h")

# Simple generation
response = client.generate("Hello world!")

# Chat format
response = client.chat([
    {"role": "system", "content": "You are a helpful assistant"},
    {"role": "user", "content": "Hello"}
])

# Connection check
is_ready = client.check_connection()

# Model list
models = client.list_models()

💡 Usage Scenarios

1. Customer Service Bot

• Remembers customer history
• Knows previous issues
• Makes personalized recommendations

2. Personal Assistant

• Tracks daily activities
• Learns preferences
• Makes reminders

3. Education Assistant

• Tracks student progress
• Adjusts difficulty level
• Remembers past mistakes

4. Support Ticket System

• Stores ticket history
• Finds related old tickets
• Provides solution suggestions

📊 Memory Format

Memories are stored in JSON format:

{
  "conversations": [
    {
      "timestamp": "2025-01-13T10:30:00",
      "user_message": "Hello",
      "bot_response": "Hello! How can I help you?",
      "metadata": {
        "topic": "greeting"
      }
    }
  ],
  "profile": {
    "user_id": "user123",
    "first_seen": "2025-01-13T10:30:00",
    "preferences": {},
    "summary": {}
  },
  "last_updated": "2025-01-13T10:35:00"
}

🔒 Privacy and Security

• ✅ Works completely locally (no internet connection required)
• ✅ Data stored on your computer
• ✅ No data sent to third-party services
• ✅ Memories in JSON format, easily deletable

🛠️ Development

Test Mode

# Simple chat without memory (for testing)
response = agent.simple_chat("Test message")

Using Your Own Model

# Different Ollama model
agent = MemAgent(model="llama2:7b")

# Or another LLM API
# Customize llm_client.py file

🐛 Troubleshooting

Ollama Connection Error

# Start Ollama service
ollama serve

# Port check
netstat -an | findstr "11434"

Model Not Found

# Check model list
ollama list

# Download model
ollama pull granite4:tiny-h

Memory Issues

# Check memory directory
import os
os.path.exists("memories")

# List memory files
os.listdir("memories")

📈 Performance

• Model Size: ~2.5 GB
• Response Time: ~1-3 seconds (depends on CPU)
• Memory Usage: ~4-6 GB RAM
• Disk Usage: ~10-50 KB per user

🤝 Contributing

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

📝 License

MIT License - See LICENSE file for details.

🙏 Acknowledgments

• Ollama - Local LLM server
• Granite - IBM Granite models

📞 Contact

You can open an issue for your questions.

---

Note: This project is for educational and research purposes. Please perform comprehensive testing before using in production environment.