checkpoint>channel_values>messages records the entire chat history. Is this scalable?

[humbroll]

As we implement a chatting agent with langgraph and Firestore database, we're trying to implement a Firestore checkpointer to record historical data in the Firestore db. We've just realized that each checkpoint data contains the entire chat message history in checkpoint>channel_values>messages.

This means if I have 5 chat messages in the chat history, each checkpoint>channel_values>messages would look like this:
[msg1, msg2, msg3, msg4, msg5]
[msg1, msg2, msg3, msg4]
[msg1, msg2, msg3]
[msg1, msg2]
[msg1]

Is this a scalable approach? As more messages accumulate, the last checkpoint data size would become huge. Yes, it'd be convenient to have all historical data, but we'd definitely need to summarize, trim, or filter messages when calling the LLM.

Am I missing something? I'm trying to understand the intention behind this design.

Thanks!

[manu156]

Why is this still unanswered 🥲