what is expected behavior of SummarizationMiddleware when summarization is disabled?

[ccurme]

Privileged issue

• I am a LangChain maintainer, or was asked directly by a LangChain maintainer to create an issue here.

Issue Content

SummarizationMiddleware has a max_tokens_before_summary parameter that says "If None, summarization is disabled":

langchain/libs/langchain_v1/langchain/agents/middleware/summarization.py

Lines 83 to 84 in 3286a98

	max_tokens_before_summary: Token threshold to trigger summarization.
	If `None`, summarization is disabled.

there is a test case against this here:

langchain/libs/langchain_v1/tests/unit_tests/agents/test_middleware_agent.py

Lines 1059 to 1063 in 3286a98

	# Test when summarization is disabled
	middleware_disabled = SummarizationMiddleware(model=model, max_tokens_before_summary=None)
	state = {"messages": [HumanMessage(content="Hello"), AIMessage(content="Hi")]}
	result = middleware_disabled.before_model(state, None)
	assert result is None

but the number of messages (two) in the test is below the default messages_to_keep of 20. if you pass in more messages, summarization happens— I've pushed up a failing test case at 0c2def8.

What is the expected behavior when we set max_tokens_before_summary=None, but the number of input messages exceeds messages_to_keep?

[sydney-runkle]

Ah yeah good question, like do we ever summarize if the token count isn't exceeded + specified.

I think we might want to be able to trigger summarization either on token count of message count. Should refine API surface accordingly.

I think current behavior is buggy.

[keenborder786]

@sydney-runkle Why don't we just do the following:

def before_model(self, state: AgentState, runtime: Runtime) -> dict[str, Any] | None:  # noqa: ARG002
    """Process messages before model invocation, potentially triggering summarization."""
    # If max_tokens_before_summary is None, don't summarize
    if self.max_tokens_before_summary is None:
        return None