oss: system message with langchain and langgraph

[robertotup]

URL

https://python.langchain.com/docs/concepts/messages/

Checklist

• I added a very descriptive title to this issue.
• I included a link to the documentation page I am referring to (if applicable).

Issue with current documentation:

Let's say invoke the llm once with both system message and user message
Next time, do I need to feed the system message again? It's not clear from the documentation.

In particular, I don't understand what happens when using checkpoint memory in langgraph. In that case, can I just feed messages with no system message after the first time I run invoke with a node?

Idea or request for content:

No response

[keenborder786]

• If you are using no checkpointer then you will have the feed all the messages again including AI, User and System.
• However, if you are using a checkpointer (long-term persistence), in that case you do not need to pass any messages again since the entire state is persisted in the checkpointer. But you will have to pass the same thread_id along with your latest message.
• I am assuming you are using MessageState where in your graph you are only passing a list of Messages.

See this:
Memory

[robertotup]

Thank you. I will try using checkpointer then. Will let you know if any issues.

[robertotup]

Mohammad Mohtashim (@keenborder786) just to clarify, if there are multiple LLMs on different nodes, does that mean they all see the entire message history (including all other LLMs responses)?

and one follow up question is this: if one of these nodes is a deterministic functions, will the other LLMs get to see the answer?

Thank you,

[keenborder786]

Okay and yes, other nodes will see it as long as you are updating the state in the given node.