it is pretty simple at the moment – just keeps a buffer of the past N conversations in memory. a pretty simple improvement would be dynamically adjusting N based on the length of the conversations

→ View original post on X — @langchain