LangGraph: Avoiding Unnecessary State Persistence

[VolanDeVovan]

I have my application which built on langchain + langgraph and store their state in postgres database using langgraph-checkpoint-postgres

My app have a few runnables which operates State schema

My state splited in 3 object:

class InputState(TypedDict):
    user_input: str
class OutputState(TypedDict):
    messages: Annotated[list, add_messages]
class OverallState(InputState, OutputState):
    character_rag_info: str  

This state used in graph.py

graph_builder = StateGraph(OverallState, input=InputState, output=OutputState)

// ...

I have 3 runnables

1. First runnable accept user_input
2. Second runnable load character_rag_info (you can use mock)
3. Generate llm response using loaded character_rag_info from prev step

character_rag_info passed through State mechanism from 2 step to 3

i faced with problem that my character_rag_info writes in postgres database and takes up a lot of space
How to solve this problem.
I need either a way to pass data around states so that it is not written to the database
Or to pass through the states so that the individual values are not written to the

Please advise how this problem is usually solved when using this stack.