Augmenting Language Models with Long-Term Memory

This was the obvious missing thing, so let's see how it goes