>We are still exploring. Fair point. Worth of looking into, is to create/train/tune small model (2b/7b) based on previous cached answers in case your knowledge index/domain is without changes in time.

Exciting times