From 4cb0bb858d88cea45fbc4acc5982691f71abbe2a Mon Sep 17 00:00:00 2001
From: Ishaan Jaffer <ishaanjaffer0324@gmail.com>
Date: Thu, 12 Feb 2026 09:15:53 -0800
Subject: [PATCH] test_responses_api_context_management_server_side_compaction

---
 litellm/types/llms/openai.py                  | 15 +++++++
 .../base_responses_api.py                     | 41 ++++++++++++++++++-
 2 files changed, 54 insertions(+), 2 deletions(-)

diff --git a/litellm/types/llms/openai.py b/litellm/types/llms/openai.py
index 299b47199ed..6925e2327c6 100644
--- a/litellm/types/llms/openai.py
+++ b/litellm/types/llms/openai.py
@@ -1074,6 +1074,19 @@ class PromptObject(TypedDict, total=False):
     """Optional version of the prompt template."""
 
 
+class ContextManagementEntry(TypedDict, total=False):
+    """
+    Context management configuration entry for a request.
+    See https://developers.openai.com/api/docs/guides/compaction.
+    """
+
+    type: str
+    """The context management entry type. Currently only ``'compaction'`` is supported."""
+
+    compact_threshold: int
+    """Token threshold at which compaction is triggered for this entry. Minimum 1000."""
+
+
 class ResponsesAPIOptionalRequestParams(TypedDict, total=False):
     """TypedDict for Optional parameters supported by the responses API."""
 
@@ -1104,6 +1117,8 @@ class ResponsesAPIOptionalRequestParams(TypedDict, total=False):
     partial_images: Optional[
         int
     ]  # Number of partial images to generate (1-3) for streaming image generation
+    context_management: Optional[List[ContextManagementEntry]]
+    """Context management configuration. E.g. [{\"type\": \"compaction\", \"compact_threshold\": 200000}] for server-side compaction (minimum 1000)."""
 
 
 class ResponsesAPIRequestParams(ResponsesAPIOptionalRequestParams, total=False):
diff --git a/tests/llm_responses_api_testing/base_responses_api.py b/tests/llm_responses_api_testing/base_responses_api.py
index 37ed1a9b08c..b6bcc1664a2 100644
--- a/tests/llm_responses_api_testing/base_responses_api.py
+++ b/tests/llm_responses_api_testing/base_responses_api.py
@@ -669,7 +669,7 @@ class BaseResponsesAPITest(ABC):
     async def test_cancel_responses_invalid_response_id(self, sync_mode):
         """Test cancel_responses with invalid response ID should raise appropriate error"""
         base_completion_call_args = self.get_base_completion_call_args()
-        
+
         if sync_mode:
             with pytest.raises(Exception):
                 litellm.cancel_responses(
@@ -679,4 +679,41 @@ class BaseResponsesAPITest(ABC):
             with pytest.raises(Exception):
                 await litellm.acancel_responses(
                     response_id="invalid_response_id_12345", **base_completion_call_args
-                )
\ No newline at end of file
+                )
+
+    @pytest.mark.parametrize("sync_mode", [True, False])
+    @pytest.mark.asyncio
+    async def test_responses_api_context_management_server_side_compaction(self, sync_mode):
+        """
+        E2E test for server-side compaction (context_management) on OpenAI Responses API.
+        Passes context_management with compact_threshold; validates that the request is
+        accepted and returns a valid response. Compaction may not run for short inputs.
+        """
+        base_completion_call_args = self.get_base_completion_call_args()
+        model = base_completion_call_args.get("model") or ""
+        # Only run with context_management for OpenAI (OAI) for now
+        if "openai/" not in str(model) and "azure/" not in str(model):
+            pytest.skip(
+                "context_management server-side compaction e2e is only run for OpenAI/Azure"
+            )
+        context_management = [{"type": "compaction", "compact_threshold": 200000}]
+        try:
+            if sync_mode:
+                response = litellm.responses(
+                    input="Short ping to verify context_management is accepted.",
+                    max_output_tokens=20,
+                    context_management=context_management,
+                    **base_completion_call_args,
+                )
+            else:
+                response = await litellm.aresponses(
+                    input="Short ping to verify context_management is accepted.",
+                    max_output_tokens=20,
+                    context_management=context_management,
+                    **base_completion_call_args,
+                )
+        except litellm.InternalServerError:
+            pytest.skip("Skipping test due to litellm.InternalServerError")
+        validate_responses_api_response(response, final_chunk=True)
+        assert response.get("id") is not None
+        assert response.get("status") is not None
\ No newline at end of file