holocron-ai 0.1.0

A Star Wars character AI chat interface using PyQt and LLMs

HolocronAI

A Star Wars character AI chat interface using PyQt and LLMs. This tool allows users to interact with AI-powered versions of characters from Knights of the Old Republic, leveraging game dialog data to maintain authentic character personalities.

Features

• Modern PyQt-based UI with dark theme

• Real-time character-by-character response generation

• Dialog embedding generation for character personality modeling

• Customizable LLM backend through LiteLLM

• Game installation integration through PyKotor

• Extracts all dialogs from KOTOR installations

• Uses DITTO self-alignment framework for dynamic response generation

• Implements Beyond Dialog concepts for contextual understanding

• No hardcoded patterns or pre-training required

• Maintains character consistency through embedding-based alignment

Installation

1. Create a virtual environment:

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

2. Install in development mode:

pip install -e .

Usage

Run the application:

holocron-ai

1. Select your KOTOR game installation directory
2. Wait for dialog processing and embedding generation
3. Choose a character to interact with
4. Start chatting!

The system will:

1. Extract all dialogs from the installation
2. Process them through the self-alignment system
3. Start an interactive session where you can chat with the character

How It Works

HolocronAI uses three main components:

1. Dialog Processor: Extracts and processes dialogs from KOTOR installations, converting them into embeddings that capture semantic meaning.

2. Self-Alignment System: Implements the DITTO framework to maintain character consistency without relying on hardcoded patterns. This system:

   • Dynamically learns from dialog embeddings
   • Adjusts responses based on conversation context
   • Maintains character consistency through embedding alignment

3. Character Agent: Combines the dialog processor and self-alignment system to generate responses that are:

   • Contextually appropriate
   • Character-consistent
   • Based on actual game dialog

Example Interaction

> Hello there!

Character: I see you're not one for subtlety. Very well - I'll be direct as well.

> What do you think about the Jedi?

Character: The Jedi... Their Code is their weakness. The galaxy has far more 
shades of gray than their black and white philosophies would suggest.

Technical Details

• Uses sentence-transformers for generating embeddings
• Implements dynamic context weighting based on semantic similarity
• Maintains conversation history with sliding context window
• Automatically saves processed data for faster subsequent runs

Requirements

• Python 3.8+
• KOTOR installation
• Dependencies listed in setup.py

Contributing

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

License

This project is licensed under the MIT License - see the LICENSE file for details.