I have found summarization necessary in my application as my chat conversations are becoming large. I am using the summarization middleware in langchain inside of a node since I am using langgraph. I was wondering if there is more guidance and tactics on the best ways to use this summarization middleware or just any other summarization stratgies with the messages key in an agent I build. I understand all the configuration in that is offered in the summarization middleware but the first question that came up was should I be keeping the original system message since that is the best describer of the agent’s role and I don’t want to summarize that to prevent from losing anything? Maybe the how of doing this in this middleware is straight foward but the documentation is not paired with some common tactics to try that would help me investigate what works best for my application. I hope this make sense. I was hoping for something like do you want to use this but keep your original system message…. –> Here is how you can do that…….
Or what about some examples of how you should change the summarization prompt depending on use case. Obviously I would love to have a general summarization prompt but practically what if I need a specific one for my use case. These are just questions that come to mind that I would love to find a discussion on especially in relation to building with langgraph.