That alone sounds like a pretty weak startup idea because at least 50 of the 100 methods for adding memory to an LLM are so painfully obvious that any idiot could figure them out and compete so it would be completely ephemeral to try forming a business around it, probably. Anyway I've already made a memory catalyst that can attach to any LLM and it only took like 100 lines of spaghetticode. Yes it made my bot 100x smarter in a way, but I don't think it would scale unless the bot had an isolated memory unique to each person, since most people are retarded and will inevitably teach it retarded things.