Need Feedback on Custom Reducer to Summarize Conversations

[droideronline]

I’m newbie to Langchain and experimenting with LangGraph to build an SQL analysis workflow. I’ve come up with a pattern for maintaining conversation context and would love feedback:

• At the end of each query, I summarize the Q&A using a node.
• A custom state reducer takes the previous summary, combines it with the new one, updates the state, and stores it as previous_convo.

The goal is to keep a condensed version of the entire conversation history accessible.

Does this seem efficient? Any better approaches I should consider?

[mfaizanse]

I am trying a similar approach but I have two concerns:

1. RunnableConfig in a reducer? #2969
2. Error handling, what to do if LLM call fails.