Should I use Conversation Summary Memory?

[JanPalasek]

I have a langgraph (a runnable langchain component), where I want to record only its input (human message) and output (ai message). I would like to have some kind of memory class, which can record the messages after they passed through the graph and record the answers.

This purpose is fulfilled by ConversationSummaryMemory, which does exactly that. However, I noticed that its missing from the newest documentation. Should I use it? Is it deprecated? I noticed that one of its mixin ancestors is.

I also noticed in the newest documentation how the summary is handled. Is this the new recommended way?

[prcastro]

I believe both langchain.memory and langchain.core.memory should be deprecated. They don't seem to integrate well with LCEL, lack documentation and seem to have a very contrived API. There's no way good, integrated way to implement Summary Memory today.