This feels like a good point. I am curious how the "feeding" would occur here. If there's an API to do this sort of thing right now I'm not aware of it. But I definitely believe that it's coming!
It supports a context length of 32k tokens (approximately 25,000 words or about half a novel).