refactor: enhance input extraction for OpenAI responses API to support messages array format

Author: viniciusdsmello

src/openlayer/lib/integrations/openai_tracer.py

@@ -472,8 +472,7 @@ def stream_responses_chunks(
                 raw_output=raw_outputs,
                 id=inference_id,
                 metadata={
-                    "timeToFirstToken": ((first_token_time - start_time) * 1000 if first_token_time else None),
-                    "api_type": "responses",
+                    "timeToFirstToken": ((first_token_time - start_time) * 1000 if first_token_time else None)
                 },
             )
             add_to_trace(
@@ -534,7 +533,6 @@ def handle_responses_non_streaming_create(
             model_parameters=get_responses_model_parameters(kwargs),
             raw_output=response.model_dump() if hasattr(response, "model_dump") else str(response),
             id=inference_id,
-            metadata={"api_type": "responses"},
         )
 
         add_to_trace(
@@ -604,33 +602,46 @@ def extract_responses_chunk_data(chunk: Any) -> Dict[str, Any]:
 def extract_responses_inputs(kwargs: Dict[str, Any]) -> Dict[str, Any]:
     """Extract inputs from Responses API parameters.
 
+    Formats the input as a messages array similar to Chat Completions API format:
+    {"prompt": [{"role": "user", "content": "..."}]}
+
     Args:
         kwargs: The parameters passed to the Responses API
 
     Returns:
-        Dictionary with prompt/input data
+        Dictionary with prompt as a messages array
     """
-    inputs = {}
+    messages = []
 
     # Handle different input formats for Responses API
-    if "input" in kwargs:
-        inputs["prompt"] = kwargs["input"]
-    elif "conversation" in kwargs:
-        inputs["prompt"] = kwargs["conversation"]
-    elif "instructions" in kwargs:
-        inputs["prompt"] = kwargs["instructions"]
-    elif "prompt" in kwargs:
-        inputs["prompt"] = kwargs["prompt"]
+    if "conversation" in kwargs:
+        # Conversation is already in messages format
+        conversation = kwargs["conversation"]
+        if isinstance(conversation, list):
+            messages = conversation
+        else:
+            # Single message, wrap it
+            messages = [{"role": "user", "content": str(conversation)}]
     else:
-        # Fallback: try to construct from available parameters
-        prompt_parts = []
+        # Build messages array from available parameters
         if "instructions" in kwargs:
-            prompt_parts.append(f"Instructions: {kwargs['instructions']}")
+            messages.append({"role": "system", "content": kwargs["instructions"]})
+        
         if "input" in kwargs:
-            prompt_parts.append(f"Input: {kwargs['input']}")
-        inputs["prompt"] = " | ".join(prompt_parts) if prompt_parts else "No input provided"
+            messages.append({"role": "user", "content": kwargs["input"]})
+        elif "prompt" in kwargs:
+            messages.append({"role": "user", "content": kwargs["prompt"]})
+        
+        # If no user message was added, create a fallback
+        if not any(msg.get("role") == "user" for msg in messages):
+            if messages:
+                # Only system message, add empty user message
+                messages.append({"role": "user", "content": ""})
+            else:
+                # No messages at all, add placeholder
+                messages.append({"role": "user", "content": "No input provided"})
 
-    return inputs
+    return {"prompt": messages}
 
 
 def parse_responses_output_data(response: Any) -> Union[str, Dict[str, Any], None]:
@@ -815,7 +826,7 @@ def trace_openai_assistant_thread_run(client: "openai.OpenAI", run: "openai.type
 
     # pylint: disable=broad-except
     except Exception as e:
-        print(f"Failed to monitor run. {e}")  # noqa: T201
+        logger.error("Failed to monitor run. %s", e)
 
 
 def _type_check_run(run: "openai.types.beta.threads.run.Run") -> None: