fix(proxy): post-call guardrail response not captured for logging

litellm/llms/a2a/chat/guardrail_translation/handler.py

@@ -111,6 +111,7 @@ async def process_output_response(
         guardrail_to_apply: "CustomGuardrail",
         litellm_logging_obj: Optional["LiteLLMLoggingObj"] = None,
         user_api_key_dict: Optional["UserAPIKeyAuth"] = None,
+        request_data: Optional[dict] = None,
     ) -> Any:
         """
         Process A2A output response by applying guardrails to text content.
@@ -166,13 +167,21 @@ async def process_output_response(
             return response
 
         # Step 2: Apply guardrail to all texts in batch
-        # Create a request_data dict with response info and user API key metadata
-        request_data: dict = {"response": response_dict}
+        # Use the real request_data if provided (proxy path), otherwise
+        # create a standalone dict (SDK / direct-call path).
+        if request_data is None:
+            request_data = {"response": response_dict}
+        else:
+            if "response" not in request_data:
+                request_data["response"] = response_dict
 
         # Add user API key metadata with prefixed keys
-        user_metadata = self.transform_user_api_key_dict_to_metadata(user_api_key_dict)
-        if user_metadata:
-            request_data["litellm_metadata"] = user_metadata
+        if "litellm_metadata" not in request_data:
+            user_metadata = self.transform_user_api_key_dict_to_metadata(
+                user_api_key_dict
+            )
+            if user_metadata:
+                request_data["litellm_metadata"] = user_metadata
 
         inputs = GenericGuardrailAPIInputs(texts=texts_to_check)
 
@@ -213,6 +222,7 @@ async def process_output_streaming_response(
         guardrail_to_apply: "CustomGuardrail",
         litellm_logging_obj: Optional["LiteLLMLoggingObj"] = None,
         user_api_key_dict: Optional["UserAPIKeyAuth"] = None,
+        request_data: Optional[dict] = None,
     ) -> List[Any]:
         """
         Process A2A streaming output by applying guardrails to accumulated text.
@@ -258,10 +268,18 @@ async def process_output_streaming_response(
         if not combined_text:
             return responses_so_far
 
-        request_data: dict = {"responses_so_far": responses_so_far}
-        user_metadata = self.transform_user_api_key_dict_to_metadata(user_api_key_dict)
-        if user_metadata:
-            request_data["litellm_metadata"] = user_metadata
+        if request_data is None:
+            request_data = {"responses_so_far": responses_so_far}
+        else:
+            if "responses_so_far" not in request_data:
+                request_data["responses_so_far"] = responses_so_far
+
+        if "litellm_metadata" not in request_data:
+            user_metadata = self.transform_user_api_key_dict_to_metadata(
+                user_api_key_dict
+            )
+            if user_metadata:
+                request_data["litellm_metadata"] = user_metadata
 
         inputs = GenericGuardrailAPIInputs(texts=[combined_text])
         guardrailed_inputs = await guardrail_to_apply.apply_guardrail(

litellm/llms/anthropic/chat/guardrail_translation/handler.py

@@ -252,6 +252,7 @@ async def process_output_response(
         guardrail_to_apply: "CustomGuardrail",
         litellm_logging_obj: Optional[Any] = None,
         user_api_key_dict: Optional[Any] = None,
+        request_data: Optional[dict] = None,
     ) -> Any:
         """
         Process output response by applying guardrails to text content and tool calls.
@@ -323,15 +324,21 @@ async def process_output_response(
 
         # Step 2: Apply guardrail to all texts in batch
         if texts_to_check or tool_calls_to_check:
-            # Create a request_data dict with response info and user API key metadata
-            request_data: dict = {"response": response}
+            # Use the real request_data if provided (proxy path), otherwise
+            # create a standalone dict (SDK / direct-call path).
+            if request_data is None:
+                request_data = {"response": response}
+            else:
+                if "response" not in request_data:
+                    request_data["response"] = response
 
             # Add user API key metadata with prefixed keys
-            user_metadata = self.transform_user_api_key_dict_to_metadata(
-                user_api_key_dict
-            )
-            if user_metadata:
-                request_data["litellm_metadata"] = user_metadata
+            if "litellm_metadata" not in request_data:
+                user_metadata = self.transform_user_api_key_dict_to_metadata(
+                    user_api_key_dict
+                )
+                if user_metadata:
+                    request_data["litellm_metadata"] = user_metadata
 
             inputs = GenericGuardrailAPIInputs(texts=texts_to_check)
             if images_to_check:
@@ -375,6 +382,7 @@ async def process_output_streaming_response(
         guardrail_to_apply: "CustomGuardrail",
         litellm_logging_obj: Optional[Any] = None,
         user_api_key_dict: Optional[Any] = None,
+        request_data: Optional[dict] = None,
     ) -> List[Any]:
         """
         Process output streaming response by applying guardrails to text content.
@@ -413,7 +421,7 @@ async def process_output_streaming_response(
 
                 _guardrailed_inputs = await guardrail_to_apply.apply_guardrail(  # allow rejecting the response, if invalid
                     inputs=guardrail_inputs,
-                    request_data={},
+                    request_data=request_data if request_data is not None else {},
                     input_type="response",
                     logging_obj=litellm_logging_obj,
                 )
@@ -426,7 +434,7 @@ async def process_output_streaming_response(
         string_so_far = self.get_streaming_string_so_far(responses_so_far)
         _guardrailed_inputs = await guardrail_to_apply.apply_guardrail(  # allow rejecting the response, if invalid
             inputs={"texts": [string_so_far]},
-            request_data={},
+            request_data=request_data if request_data is not None else {},
             input_type="response",
             logging_obj=litellm_logging_obj,
         )

litellm/llms/base_llm/guardrail_translation/base_translation.py

@@ -73,6 +73,7 @@ async def process_output_response(
         guardrail_to_apply: "CustomGuardrail",
         litellm_logging_obj: Optional["LiteLLMLoggingObj"] = None,
         user_api_key_dict: Optional["UserAPIKeyAuth"] = None,
+        request_data: Optional[dict] = None,
     ) -> Any:
         """
         Process output response with guardrails.
@@ -91,6 +92,7 @@ async def process_output_streaming_response(
         guardrail_to_apply: "CustomGuardrail",
         litellm_logging_obj: Optional["LiteLLMLoggingObj"] = None,
         user_api_key_dict: Optional["UserAPIKeyAuth"] = None,
+        request_data: Optional[dict] = None,
     ) -> Any:
         """
         Process output streaming response with guardrails.

litellm/llms/cohere/rerank/guardrail_translation/handler.py

@@ -83,6 +83,7 @@ async def process_output_response(
         guardrail_to_apply: "CustomGuardrail",
         litellm_logging_obj: Optional[Any] = None,
         user_api_key_dict: Optional[Any] = None,
+        request_data: Optional[dict] = None,
     ) -> Any:
         """
         Process output response - not applicable for rerank.

litellm/llms/mistral/ocr/guardrail_translation/handler.py

@@ -91,6 +91,7 @@ async def process_output_response(
         guardrail_to_apply: "CustomGuardrail",
         litellm_logging_obj: Optional[Any] = None,
         user_api_key_dict: Optional[Any] = None,
+        request_data: Optional[dict] = None,
     ) -> Any:
         """
         Process OCR output by applying guardrails to extracted page text.
@@ -127,14 +128,27 @@ async def process_output_response(
         if model:
             inputs["model"] = model
 
+        # Use the real request_data if provided (proxy path), otherwise
+        # create a standalone dict (SDK / direct-call path).
+        if request_data is None:
+            request_data = {}
+
         # Add user metadata if available
         if user_api_key_dict is not None:
-            metadata = self.transform_user_api_key_dict_to_metadata(user_api_key_dict)
-            inputs.update(metadata)  # type: ignore
+            user_metadata = self.transform_user_api_key_dict_to_metadata(
+                user_api_key_dict
+            )
+            if user_metadata:
+                # Preserve original behavior: inject metadata into inputs for
+                # third-party guardrail providers that read it from there
+                inputs.update(user_metadata)  # type: ignore
+                # Also store in request_data for the logging pipeline
+                if "litellm_metadata" not in request_data:
+                    request_data["litellm_metadata"] = user_metadata
 
         guardrailed_inputs = await guardrail_to_apply.apply_guardrail(
             inputs=inputs,
-            request_data={},
+            request_data=request_data,
             input_type="response",
             logging_obj=litellm_logging_obj,
         )

litellm/llms/openai/chat/guardrail_translation/handler.py

@@ -86,9 +86,9 @@ async def process_input_messages(
             if tool_calls_to_check:
                 inputs["tool_calls"] = tool_calls_to_check  # type: ignore
             if messages:
-                inputs[
-                    "structured_messages"
-                ] = messages  # pass the openai /chat/completions messages to the guardrail, as-is
+                inputs["structured_messages"] = (
+                    messages  # pass the openai /chat/completions messages to the guardrail, as-is
+                )
             # Pass tools (function definitions) to the guardrail
             tools = data.get("tools")
             if tools:
@@ -260,6 +260,7 @@ async def process_output_response(
         guardrail_to_apply: "CustomGuardrail",
         litellm_logging_obj: Optional[Any] = None,
         user_api_key_dict: Optional[Any] = None,
+        request_data: Optional[dict] = None,
     ) -> Any:
         """
         Process output response by applying guardrails to text content.
@@ -308,15 +309,21 @@ async def process_output_response(
 
         # Step 2: Apply guardrail to all texts and tool calls in batch
         if texts_to_check or tool_calls_to_check:
-            # Create a request_data dict with response info and user API key metadata
-            request_data: dict = {"response": response}
+            # Use the real request_data if provided (proxy path), otherwise
+            # create a standalone dict (SDK / direct-call path).
+            if request_data is None:
+                request_data = {"response": response}
+            else:
+                if "response" not in request_data:
+                    request_data["response"] = response
 
             # Add user API key metadata with prefixed keys
-            user_metadata = self.transform_user_api_key_dict_to_metadata(
-                user_api_key_dict
-            )
-            if user_metadata:
-                request_data["litellm_metadata"] = user_metadata
+            if "litellm_metadata" not in request_data:
+                user_metadata = self.transform_user_api_key_dict_to_metadata(
+                    user_api_key_dict
+                )
+                if user_metadata:
+                    request_data["litellm_metadata"] = user_metadata
 
             inputs = GenericGuardrailAPIInputs(texts=texts_to_check)
             if images_to_check:
@@ -364,6 +371,7 @@ async def process_output_streaming_response(
         guardrail_to_apply: "CustomGuardrail",
         litellm_logging_obj: Optional[Any] = None,
         user_api_key_dict: Optional[Any] = None,
+        request_data: Optional[dict] = None,
     ) -> List["ModelResponseStream"]:
         """
         Process output streaming responses by applying guardrails to text content.
@@ -402,6 +410,7 @@ async def process_output_streaming_response(
                 guardrail_to_apply=guardrail_to_apply,
                 litellm_logging_obj=litellm_logging_obj,
                 user_api_key_dict=user_api_key_dict,
+                request_data=request_data,
             )
 
             return responses_so_far
@@ -436,15 +445,21 @@ async def process_output_streaming_response(
 
         # Step 3: Apply guardrail to all combined texts in batch
         if texts_to_check:
-            # Create a request_data dict with response info and user API key metadata
-            request_data: dict = {"responses": responses_so_far}
+            # Use the real request_data if provided (proxy path), otherwise
+            # create a standalone dict (SDK / direct-call path).
+            if request_data is None:
+                request_data = {"responses": responses_so_far}
+            else:
+                if "responses" not in request_data:
+                    request_data["responses"] = responses_so_far
 
             # Add user API key metadata with prefixed keys
-            user_metadata = self.transform_user_api_key_dict_to_metadata(
-                user_api_key_dict
-            )
-            if user_metadata:
-                request_data["litellm_metadata"] = user_metadata
+            if "litellm_metadata" not in request_data:
+                user_metadata = self.transform_user_api_key_dict_to_metadata(
+                    user_api_key_dict
+                )
+                if user_metadata:
+                    request_data["litellm_metadata"] = user_metadata
 
             inputs = GenericGuardrailAPIInputs(texts=texts_to_check)
             if images_to_check:

litellm/llms/openai/completion/guardrail_translation/handler.py

@@ -125,6 +125,7 @@ async def process_output_response(
         guardrail_to_apply: "CustomGuardrail",
         litellm_logging_obj: Optional[Any] = None,
         user_api_key_dict: Optional[Any] = None,
+        request_data: Optional[dict] = None,
     ) -> Any:
         """
         Process output response by applying guardrails to completion text.
@@ -155,15 +156,21 @@ async def process_output_response(
 
         # Apply guardrails in batch
         if texts_to_check:
-            # Create a request_data dict with response info and user API key metadata
-            request_data: dict = {"response": response}
+            # Use the real request_data if provided (proxy path), otherwise
+            # create a standalone dict (SDK / direct-call path).
+            if request_data is None:
+                request_data = {"response": response}
+            else:
+                if "response" not in request_data:
+                    request_data["response"] = response
 
             # Add user API key metadata with prefixed keys
-            user_metadata = self.transform_user_api_key_dict_to_metadata(
-                user_api_key_dict
-            )
-            if user_metadata:
-                request_data["litellm_metadata"] = user_metadata
+            if "litellm_metadata" not in request_data:
+                user_metadata = self.transform_user_api_key_dict_to_metadata(
+                    user_api_key_dict
+                )
+                if user_metadata:
+                    request_data["litellm_metadata"] = user_metadata
 
             inputs = GenericGuardrailAPIInputs(texts=texts_to_check)
             # Include model information from the response if available

litellm/llms/openai/embeddings/guardrail_translation/handler.py

@@ -155,6 +155,7 @@ async def process_output_response(
         guardrail_to_apply: "CustomGuardrail",
         litellm_logging_obj: Optional[Any] = None,
         user_api_key_dict: Optional[Any] = None,
+        request_data: Optional[dict] = None,
     ) -> Any:
         """
         Process output response - embeddings responses contain vectors, not text.

litellm/llms/openai/image_generation/guardrail_translation/handler.py

@@ -87,6 +87,7 @@ async def process_output_response(
         guardrail_to_apply: "CustomGuardrail",
         litellm_logging_obj: Optional[Any] = None,
         user_api_key_dict: Optional[Any] = None,
+        request_data: Optional[dict] = None,
     ) -> Any:
         """
         Process output response - typically not needed for image generation.