These models are stochastic though, so saving conversations with elaborate context so they're actually useful in the domain you need them to be useful in is a fundamental feature. The alternative to that is saving a document to copy-paste in as a "startup prompt" of things it needs to know contextually, which is kind of silly.