sadly non of that right now. I think down the line I'll hook it up to an optional context/memory layer, so that it has an idea of what you've read so far. (lots of local open-source options iirc)

As you read, pages would be broken down and fed into this layer.