Decouple Dialogue History from Graph Schema When Refactoring

[Khaled-Abdelhamid]

I’ve been using LangGraph with the MongoDB checkpointer for about a year. It reliably stores full state, including message history under the messages channel. However, if I significantly refactor or rename nodes in my graph, I can no longer access the prior conversation history—even though it still exists in MongoDB.
My goal is: I only care about preserving conversation messages (user and assistant), not the entire internal agent state. I’d like to refactor my graph later (e.g. add features, rename nodes), and still be able to continue previous sessions under the same thread_id.
What is the best practice in the LangGraph ecosystem for this scenario?
• Should I use a separate message-only store (independent of LangGraph checkpoint state)?
• Are there built-in strategies or recommended reducers/hooks (e.g. trimming, custom state channels) to decouple conversation logs from schema changes?
• Has anyone implemented a robust method to persist and reload only messages across refactored graphs?