A build from scratch implementation of CodeMusai in python, complete with API.

CodeMusai is a new form of AI with changes to traditional components to better mimic human thought process. Traditional Attention Transformers are written with mathematical terms, this repository will be written through a psychology a neuroscience lens. Terms like Thought Processor, Temporal Focus, Cortex, dendrites as input, axons as output, etc. will be used to write the code.

This is a work in progress, and will be updated regularly. This repository will represent the public portion of the code base.

After implementing all the need-to-have features from my first prototype, I will produce a new version that uses MLX.

~ CodeMusai is a unique AI, it has two GPT components, one being a LanguageCortex, and the other being an EmotionCortex. When generating from either logical or emotional corticies a multimodal system message allows for an emotion component it influence logic generation, and similarly allowing a logical statment to adjust the emotion component being produced; this will be implemented through the use of cross attention.