MyGPT

MyGPT aims to be a personal assistant designed to learn from user interactions, building a personalized knowledge graph to enhance future conversations. It also generates valuable data for fine-tuning language models for personalization.

Main Concepts

• Knowledge Graph: The chatbot builds a knowledge graph based on user interactions, allowing it to provide more contextually relevant responses over time.
• Automatic Feedback Loop: The system collects feedback on its outputs directly through the chat, creating valuable data for fine-tuning and improvement.
• Personalized Communication: MyGPT learns the user's preferred tone, structure, and style for specific contexts or types of content.
• Relationship Dynamics: Through continued use, the chatbot learns about the user's dynamics with other individuals or entities mentioned in conversations.

Roadmap for v0.1

For v0.1, we are focusing on establishing core functionality:

• FastAPI backend for handling multiple sessions
• SQLite for interaction data storage
• LiteLLM integration for flexible LLM provider support
• LLM backend to work with Ollama
• Basic user interface for performing and managing multiple interaction sessions
• Knowledge graph construction from user interactions
• Personalization using the knowledge graph

Note: This project is under active development. v0.1 aims to provide a functional foundation with basic implementations of key features. Advanced optimizations and more sophisticated implementations of the main concepts will be addressed in future versions.

Installation and Usage

1. Install the required packages:

   pip install -r requirements.txt
   

2. Set up your environment variables: Create a .env file in the backend directory with the following content:

   APP_DB=app.db
   OLLAMA_MODEL=ollama/llama3.2:3b-instruct-q4_K_M
   OLLAMA_API_BASE=http://localhost:11434
   

   Adjust the OLLAMA_MODEL and OLLAMA_API_BASE values if needed.

3. Run the backend server:

   cd backend
   python -m mygpt.main
   

   The FastAPI server will start, and you can interact with it using API calls.

Project Structure

• backend/mygpt/main.py: Main entry point of the application
• backend/mygpt/api.py: FastAPI routes and API logic
• backend/mygpt/db.py: Database operations for chat history
• backend/mygpt/session.py: Session management for chat interactions
• backend/mygpt/utils.py: Utility functions (e.g., logging setup)
• requirements.txt: List of Python package dependencies

API Endpoints and Usage Examples

Start a new chat session or continue an existing one

Endpoint: POST /session

Usage:

Start a new session

curl -X POST http://localhost:8000/session \
   -H "Content-Type: application/json" \
   -d '{ "message": "Hello, I would like to start a chat."}'

The response will include the AI's reply and the newly created session ID:

{
  "response": "Hello! I'm doing well, thank you for asking. How can I assist you today?",
  "session_id": "newly_created_session_id"
}

Continue an existing session

curl -X POST http://localhost:8000/session \
  -H "Content-Type: application/json" \
  -d '{"message": "What was our last conversation about?", "session_id": "previously_returned_session_id"}'

The response will include the AI's reply and the same session ID:

{
  "response": "Hello! I'm doing well, thank you for asking. How can I assist you today?",
  "session_id": "same_session_id"
}

Retrieve the message history for a specific session

Endpoint: GET /session/{session_id}/history

Usage:

curl http://localhost:8000/session/your_session_id_here/history

The response will include the full conversation history for the specified session:

{
  "session_id": "your_session_id_here",
  "history": [
      ["user","Hello, I would like to start a chat."],
      ["assistant","I'd be happy to have a conversation with you. How's your day going so far?"],
      ["user","It's going well, thank you! I'm just exploring this new chatbot technology."],
      ["assistant","That's great! It's always exciting to see new advancements in AI. How can I assist you today?"]
  ]
}

Logging

Logs are stored in the backend/logs directory, with each session creating a new log file.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.