We have a “memory issue” in our work with a poet. The computational poet is called Fred The Heretic (FTH) and it is a custom GPT. FTH is primed with previous poetry and a guide for quality poetry gleaned from previous conversations with the poet. The poet laments that FTH doesn’t remember anything. I would say we have 2 roles, ideally: a user and a trainer. The user uses FTH but would benefit from memory of previous chats. The trainer (poet) has the same need but also FTH should learn from the trainer, which should change its behavior for all users. We are now exploring MemGPT and Letta and wonder if anyone else has experimented with this sort of memory enhancement, or other tools?