Is there a way to pass a limit to the checkpointer so it only fetches the last N messages for state["messages"]?

[aidin1324]

Is there a way to pass a limit to the checkpointer so it only fetches the last N messages for state["messages"]?

I see that the checkpointer’s list(...) method already supports limit, e.g.:

This adds LIMIT 5 to the SQL query, ensuring that only the last 5 checkpoints are retrieved.

checkpointer.list(config, limit=5)
The problem is also last 5 checkpoint is not last 5 user-ai messages

However, in my current implementation—when I create the agent via:

react_agent = create_react_agent(
    model=model,
    tools=tools,
    checkpointer=checkpointer,
    ...
)

—it seems that the entire conversation history is fetched by default, and then I trim down to the last N messages afterward, which isn’t ideal. Currently, I handle this trimming manually using:

trim_messages(
    messages,
    max_tokens=30,
    token_counter=len, 
    strategy="last", 
    allow_partial=True,
    ...
)

Question:
How can I configure or pass a limit (or any similar parameter) to the checkpointer/agent so that only the last N messages are retrieved from the database in the first place, instead of loading the entire conversation and trimming in memory?

Does langgraph support this directly, or is it something I need to implement manually?

Any detailed guidance or examples would be very helpful. Thank you!

[vbarda]

@aidin1324 this should not be done at the checkpointer level. you can change the reducer in your agent to always keep at most last N messages if you want the messages to be updated in the graph state (e.g. update add_messages in your state schema implementation https://github.com/langchain-ai/langgraph/blob/main/libs/langgraph/langgraph/graph/message.py#L266).

you would just need to be careful to make sure the resulting message history is valid

see more on reducers here https://langchain-ai.github.io/langgraph/concepts/low_level/#reducers

[tchiayan]

@vbarda , I checked on postgres saver checkpoint and the list function have this limit parameter (https://github.com/langchain-ai/langgraph/blob/main/libs/checkpoint-postgres/langgraph/checkpoint/postgres/__init__.py#L104) . May I know what is this purpose for? I found no resource that mentioning about this and how is this will be triggered to limit the checkpoint limit.

[tingyiy]

@aidin1324 please see https://python.langchain.com/api_reference/core/messages/langchain_core.messages.utils.trim_messages.html you can trim messages safely using this method and a pre_model_hook