[akvadrako]

Can't you use fine-tuning for this?

A other option is to ask GPT to compress your tokens into a shorter prompt for itself.