This is truly dope.

I've been playing with a closely related idea of treating the context as a graph. Inspired by the KGoT paper - https://arxiv.org/abs/2504.02670

I call this "live context" because it's the living brain of my agents