From dc28f0eb4b9aeee5933368055daeb3a09d5fbab1 Mon Sep 17 00:00:00 2001
From: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Date: Thu, 3 Sep 2026 01:54:09 -0700
Subject: [PATCH] fix(vector-stores): report search failures on the responses
 API surface too

---
 .../vector_store_pre_call_hook.py             | 11 +++-
 .../test_vector_store_pre_call_hook.py        | 55 ++++++++++++++++++-
 2 files changed, 64 insertions(+), 2 deletions(-)

diff --git a/litellm/integrations/vector_store_integrations/vector_store_pre_call_hook.py b/litellm/integrations/vector_store_integrations/vector_store_pre_call_hook.py
index 274cb496b8f..849e87fae40 100644
--- a/litellm/integrations/vector_store_integrations/vector_store_pre_call_hook.py
+++ b/litellm/integrations/vector_store_integrations/vector_store_pre_call_hook.py
@@ -17,7 +17,11 @@ import litellm.vector_stores
 from litellm._logging import verbose_logger
 from litellm.exceptions import VectorStoreSearchError
 from litellm.integrations.custom_logger import CustomLogger
-from litellm.types.llms.openai import AllMessageValues, ChatCompletionUserMessage
+from litellm.types.llms.openai import (
+    AllMessageValues,
+    ChatCompletionUserMessage,
+    ResponsesAPIResponse,
+)
 from litellm.types.prompts.init_prompts import PromptSpec
 from litellm.types.utils import CallTypes, StandardCallbackDynamicParams
 from litellm.types.vector_stores import (
@@ -372,6 +376,11 @@ class VectorStorePreCallHook(CustomLogger):
                 verbose_logger.debug("No search results or search failures found")
                 return None
 
+            if isinstance(response, ResponsesAPIResponse):
+                if search_failures:
+                    setattr(response, SEARCH_FAILURES_FIELD, list(search_failures))
+                return response
+
             # Add search results to response object
             if hasattr(response, "choices") and response.choices:
                 for choice in response.choices:
diff --git a/tests/test_litellm/integrations/vector_store_integrations/test_vector_store_pre_call_hook.py b/tests/test_litellm/integrations/vector_store_integrations/test_vector_store_pre_call_hook.py
index eae53becc14..f1f9f7c3f3f 100644
--- a/tests/test_litellm/integrations/vector_store_integrations/test_vector_store_pre_call_hook.py
+++ b/tests/test_litellm/integrations/vector_store_integrations/test_vector_store_pre_call_hook.py
@@ -11,7 +11,7 @@ from litellm.integrations.vector_store_integrations.vector_store_pre_call_hook i
     ProxyServerRuntime,
     VectorStorePreCallHook,
 )
-from litellm.types.llms.openai import AllMessageValues
+from litellm.types.llms.openai import AllMessageValues, ResponsesAPIResponse
 from litellm.types.utils import (
     CallTypes,
     Choices,
@@ -363,6 +363,59 @@ async def test_a_healthy_vector_store_alone_reports_no_failures(registry_with: R
     assert "vector_store_search_failures" not in (_first_message(response).provider_specific_fields or {})
 
 
+@pytest.mark.asyncio
+async def test_a_failing_vector_store_is_reported_on_the_responses_api_response(
+    registry_with: RegisterStores,
+) -> None:
+    """Regression (LIT-6809): /v1/responses answered 200 with no sign the knowledge base was missing."""
+    registry_with("vs-broken")
+    logging_obj = FakeLoggingObj({})
+
+    await _run_hook(
+        VectorStorePreCallHook(
+            proxy_runtime=FakeProxyRuntime(router=RecordingRouter(failing_vector_store_ids=frozenset({"vs-broken"})))
+        ),
+        ["vs-broken"],
+        logging_obj,
+    )
+
+    response = ResponsesAPIResponse(id="resp-lit6809", created_at=0, output=[])
+    await VectorStorePreCallHook(proxy_runtime=FakeProxyRuntime(router=None)).async_post_call_success_deployment_hook(
+        request_data={"litellm_logging_obj": logging_obj},
+        response=response,
+        call_type=CallTypes.aresponses,
+    )
+
+    assert response.model_dump()["vector_store_search_failures"] == [
+        {
+            "vector_store_id": "vs-broken",
+            "custom_llm_provider": "bedrock",
+            "error": "litellm.BadRequestError: no healthy deployments for vs-broken",
+        }
+    ]
+
+
+@pytest.mark.asyncio
+async def test_a_healthy_vector_store_leaves_the_responses_api_response_alone(registry_with: RegisterStores) -> None:
+    registry_with("vs-healthy")
+    logging_obj = FakeLoggingObj({})
+
+    await _run_hook(
+        VectorStorePreCallHook(proxy_runtime=FakeProxyRuntime(router=RecordingRouter())),
+        ["vs-healthy"],
+        logging_obj,
+    )
+
+    response = ResponsesAPIResponse(id="resp-lit6809", created_at=0, output=[])
+    await VectorStorePreCallHook(proxy_runtime=FakeProxyRuntime(router=None)).async_post_call_success_deployment_hook(
+        request_data={"litellm_logging_obj": logging_obj},
+        response=response,
+        call_type=CallTypes.aresponses,
+    )
+
+    assert "vector_store_search_failures" not in response.model_dump()
+
+
 @pytest.mark.asyncio
 async def test_a_failing_vector_store_is_reported_on_the_streaming_chunk(registry_with: RegisterStores) -> None:
     registry_with("vs-broken")