4 ms·

Couldn't the same be done by breaking the conversation down into chunks and adding the context incrementally?