Skip to content

The agent engineering platform.

refactor(langchain): engage summarization based on reported `usage_metadata` (#34632)

ccurmed383f00489b1
diff --git a/libs/langchain_v1/langchain/agents/middleware/summarization.py b/libs/langchain_v1/langchain/agents/middleware/summarization.py
index 8f6fb2789c..b3533c2eb6 100644
--- a/libs/langchain_v1/langchain/agents/middleware/summarization.py
+++ b/libs/langchain_v1/langchain/agents/middleware/summarization.py
@@ -324,6 +324,25 @@ class SummarizationMiddleware(AgentMiddleware):
]
}
+ def _should_summarize_based_on_reported_tokens(
+ self, messages: list[AnyMessage], threshold: float
+ ) -> bool:
+ """Check if reported token usage from last AIMessage exceeds threshold."""
+ last_ai_message = next(
+ (msg for msg in reversed(messages) if isinstance(msg, AIMessage)),
+ None,
+ )
+ if ( # noqa: SIM103
+ isinstance(last_ai_message, AIMessage)
+ and last_ai_message.usage_metadata is not None
+ and (reported_tokens := last_ai_message.usage_metadata.get("total_tokens", -1))
+ and reported_tokens >= threshold
+ and (message_provider := last_ai_message.response_metadata.get("model_provider"))
+ and message_provider == self.model._get_ls_params().get("ls_provider") # noqa: SLF001
+ ):
+ return True
+ return False
+
def _should_summarize(self, messages: list[AnyMessage], total_tokens: int) -> bool:
"""Determine whether summarization should run for the current token usage."""
if not self._trigger_conditions:
@@ -334,6 +353,10 @@ class SummarizationMiddleware(AgentMiddleware):
return True
if kind == "tokens" and total_tokens >= value:
return True
+ if kind == "tokens" and self._should_summarize_based_on_reported_tokens(
+ messages, value
+ ):
+ return True
if kind == "fraction":
max_input_tokens = self._get_profile_limits()
if max_input_tokens is None:
@@ -343,6 +366,9 @@ class SummarizationMiddleware(AgentMiddleware):
threshold = 1
if total_tokens >= threshold:
return True
+
+ if self._should_summarize_based_on_reported_tokens(messages, threshold):
+ return True
return False
def _determine_cutoff_index(self, messages: list[AnyMessage]) -> int:
diff --git a/libs/langchain_v1/tests/unit_tests/agents/middleware/implementations/test_summarization.py b/libs/langchain_v1/tests/unit_tests/agents/middleware/implementations/test_summarization.py
index c0ac6c82f5..728c6c97df 100644
--- a/libs/langchain_v1/tests/unit_tests/agents/middleware/implementations/test_summarization.py
+++ b/libs/langchain_v1/tests/unit_tests/agents/middleware/implementations/test_summarization.py
@@ -9,6 +9,7 @@ from langchain_core.outputs import ChatGeneration, ChatResult
from langgraph.graph.message import REMOVE_ALL_MESSAGES
from langchain.agents.middleware.summarization import SummarizationMiddleware
+from langchain.chat_models import init_chat_model
from tests.unit_tests.agents.model import FakeToolCallingModel
@@ -1014,3 +1015,80 @@ def test_create_summary_uses_get_buffer_string_format() -> None:
f"str(messages) should produce significantly more tokens. "
f"Got ratio {str_ratio:.2f}x (expected > 1.5)"
)
+
+
[email protected]("langchain_anthropic")
+def test_usage_metadata_trigger() -> None:
+ model = init_chat_model("anthropic:claude-sonnet-4-5")
+ middleware = SummarizationMiddleware(
+ model=model, trigger=("tokens", 10_000), keep=("messages", 4)
+ )
+ messages: list[AnyMessage] = [
+ HumanMessage(content="msg1"),
+ AIMessage(
+ content="msg2",
+ tool_calls=[{"name": "tool", "args": {}, "id": "call1"}],
+ response_metadata={"model_provider": "anthropic"},
+ usage_metadata={
+ "input_tokens": 5000,
+ "output_tokens": 1000,
+ "total_tokens": 6000,
+ },
+ ),
+ ToolMessage(content="result", tool_call_id="call1"),
+ AIMessage(
+ content="msg3",
+ response_metadata={"model_provider": "anthropic"},
+ usage_metadata={
+ "input_tokens": 6100,
+ "output_tokens": 900,
+ "total_tokens": 7000,
+ },
+ ),
+ HumanMessage(content="msg4"),
+ AIMessage(
+ content="msg5",
+ response_metadata={"model_provider": "anthropic"},
+ usage_metadata={
+ "input_tokens": 7500,
+ "output_tokens": 2501,
+ "total_tokens": 10_001,
+ },
+ ),
+ ]
+ # reported token count should override count of zero
+ assert middleware._should_summarize(messages, 0)
+
+ # don't engage unless model provider matches
+ messages.extend(
+ [
+ HumanMessage(content="msg6"),
+ AIMessage(
+ content="msg7",
+ response_metadata={"model_provider": "not-anthropic"},
+ usage_metadata={
+ "input_tokens": 7500,
+ "output_tokens": 2501,
+ "total_tokens": 10_001,
+ },
+ ),
+ ]
+ )
+ assert not middleware._should_summarize(messages, 0)
+
+ # don't engage if subsequent message stays under threshold (e.g., after summarization)
+ messages.extend(
+ [
+ HumanMessage(content="msg8"),
+ AIMessage(
+ content="msg9",
+ response_metadata={"model_provider": "anthropic"},
+ usage_metadata={
+ "input_tokens": 7500,
+ "output_tokens": 2499,
+ "total_tokens": 9999,
+ },
+ ),
+ ]
+ )
+ assert not middleware._should_summarize(messages, 0)