I built an open-source memory layer for LLMs — here's how it works

LLMs are stateless by design. You send a message, you get a reply, and the model instantly forgets...