Have you considered modifying your custom memory class to not only be a buffer window but add langchains entity memory for a combined memory class? It would allow for the bot to retain key details about each user it interacts with and pull that users details into context when replying to said user. This way it not only has the buffer lf the last n chats but always keeps key details. There are more and more models coming out with larger context windows that could support this and it would be useful in very active servers where the last n lines of chat may not be enough to keep a thread going.
Have you considered modifying your custom memory class to not only be a buffer window but add langchains entity memory for a combined memory class? It would allow for the bot to retain key details about each user it interacts with and pull that users details into context when replying to said user. This way it not only has the buffer lf the last n chats but always keeps key details. There are more and more models coming out with larger context windows that could support this and it would be useful in very active servers where the last n lines of chat may not be enough to keep a thread going.
Langchain Entity Memory Doc