OpenAI conversation: Standardizing API response (#318)

* passing additional params in failure case as well

* adding support to pass metadata in fail responses

Author: AkhileshNegi

backend/app/api/routes/responses.py

@@ -102,12 +102,27 @@ def get_file_search_results(response):
 
 def get_additional_data(request: dict) -> dict:
     """Extract additional data from request, excluding specific keys."""
-    return {
-        k: v
-        for k, v in request.items()
-        if k not in {"assistant_id", "callback_url", "response_id", "question"}
+    # Keys to exclude for async request (ResponsesAPIRequest)
+    async_exclude_keys = {"assistant_id", "callback_url", "response_id", "question"}
+    # Keys to exclude for sync request (ResponsesSyncAPIRequest)
+    sync_exclude_keys = {
+        "model",
+        "instructions",
+        "vector_store_ids",
+        "max_num_results",
+        "temperature",
+        "response_id",
+        "question",
     }
 
+    # Determine which keys to exclude based on the request structure
+    if "assistant_id" in request:
+        exclude_keys = async_exclude_keys
+    else:
+        exclude_keys = sync_exclude_keys
+
+    return {k: v for k, v in request.items() if k not in exclude_keys}
+
 
 def process_response(
     request: ResponsesAPIRequest,
@@ -249,7 +264,11 @@ def process_response(
             exc_info=True,
         )
         tracer.log_error(error_message, response_id=request.response_id)
-        callback_response = ResponsesAPIResponse.failure_response(error=error_message)
+
+        request_dict = request.model_dump()
+        callback_response = ResponsesAPIResponse.failure_response(
+            error=error_message, metadata=get_additional_data(request_dict)
+        )
 
     tracer.flush()
 
@@ -360,11 +379,13 @@ async def responses_sync(
         project_id=project_id,
     )
     if not credentials or "api_key" not in credentials:
+        request_dict = request.model_dump()
         logger.error(
             f"[response_sync] OpenAI API key not configured for org_id={organization_id}, project_id={project_id}"
         )
         return APIResponse.failure_response(
-            error="OpenAI API key not configured for this organization."
+            error="OpenAI API key not configured for this organization.",
+            metadata=get_additional_data(request_dict),
         )
 
     client = OpenAI(api_key=credentials["api_key"])
@@ -457,4 +478,8 @@ async def responses_sync(
         )
         tracer.log_error(error_message, response_id=request.response_id)
         tracer.flush()
-        return ResponsesAPIResponse.failure_response(error=error_message)
+
+        request_dict = request.model_dump()
+        return ResponsesAPIResponse.failure_response(
+            error=error_message, metadata=get_additional_data(request_dict)
+        )

backend/app/utils.py

@@ -37,13 +37,15 @@ def success_response(
         return cls(success=True, data=data, error=None, metadata=metadata)
 
     @classmethod
-    def failure_response(cls, error: str | list) -> "APIResponse[None]":
+    def failure_response(
+        cls, error: str | list, metadata: Optional[Dict[str, Any]] = None
+    ) -> "APIResponse[None]":
         if isinstance(error, list):  # to handle cases when error is a list of errors
             error_message = "\n".join([f"{err['loc']}: {err['msg']}" for err in error])
         else:
             error_message = error
 
-        return cls(success=False, data=None, error=error_message)
+        return cls(success=False, data=None, error=error_message, metadata=metadata)
 
 
 @dataclass