Fantastic article and impressive work by this company. They're basically wrapping LLM's with a working memory, and tying it to user input. And thus we step a little closer to AGI/ASI.