Merge pull request #350 from jrobertboos/lcore-385-response-structure

LCORE-385: Restructure tool_call responses

Author: tisnik

src/app/endpoints/streaming_query.py

@@ -213,7 +213,9 @@ def _handle_turn_complete_event(chunk: Any, chunk_id: int) -> Iterator[str]:
             "event": "turn_complete",
             "data": {
                 "id": chunk_id,
-                "token": chunk.event.payload.turn.output_message.content,
+                "token": interleaved_content_as_str(
+                    chunk.event.payload.turn.output_message.content
+                ),
             },
         }
     )
@@ -335,7 +337,10 @@ def _handle_tool_execution_event(
                     "data": {
                         "id": chunk_id,
                         "role": chunk.event.payload.step_type,
-                        "token": f"Tool:{t.tool_name} arguments:{t.arguments}",
+                        "token": {
+                            "tool_name": t.tool_name,
+                            "arguments": t.arguments,
+                        },
                     },
                 }
             )
@@ -349,7 +354,10 @@ def _handle_tool_execution_event(
                         "data": {
                             "id": chunk_id,
                             "role": chunk.event.payload.step_type,
-                            "token": f"Fetched {len(inserted_context)} bytes from memory",
+                            "token": {
+                                "tool_name": r.tool_name,
+                                "response": f"Fetched {len(inserted_context)} bytes from memory",
+                            },
                         },
                     }
                 )
@@ -380,7 +388,10 @@ def _handle_tool_execution_event(
                         "data": {
                             "id": chunk_id,
                             "role": chunk.event.payload.step_type,
-                            "token": f"Tool:{r.tool_name} summary:{summary}",
+                            "token": {
+                                "tool_name": r.tool_name,
+                                "summary": summary,
+                            },
                         },
                     }
                 )
@@ -392,7 +403,10 @@ def _handle_tool_execution_event(
                         "data": {
                             "id": chunk_id,
                             "role": chunk.event.payload.step_type,
-                            "token": f"Tool:{r.tool_name} response:{r.content}",
+                            "token": {
+                                "tool_name": r.tool_name,
+                                "response": interleaved_content_as_str(r.content),
+                            },
                         },
                     }
                 )
@@ -446,16 +460,20 @@ async def streaming_query_endpoint_handler(
         async def response_generator(turn_response: Any) -> AsyncIterator[str]:
             """Generate SSE formatted streaming response."""
             chunk_id = 0
-            complete_response = ""
+            complete_response = "No response from the model"
 
             # Send start event
             yield stream_start_event(conversation_id)
 
             async for chunk in turn_response:
                 for event in stream_build_event(chunk, chunk_id, metadata_map):
-                    complete_response += json.loads(event.replace("data: ", ""))[
-                        "data"
-                    ]["token"]
+                    if (
+                        json.loads(event.replace("data: ", ""))["event"]
+                        == "turn_complete"
+                    ):
+                        complete_response = json.loads(event.replace("data: ", ""))[
+                            "data"
+                        ]["token"]
                     chunk_id += 1
                     yield event
 

tests/unit/app/endpoints/test_streaming_query.py

@@ -181,11 +181,21 @@ async def _test_streaming_query_endpoint_handler(mocker, store_transcript=False)
                 payload=AgentTurnResponseStepProgressPayload(
                     event_type="step_progress",
                     step_type="inference",
-                    delta=TextDelta(text="LLM answer", type="text"),
+                    delta=TextDelta(text="LLM ", type="text"),
                     step_id="s1",
                 )
             )
         ),
+        AgentTurnResponseStreamChunk(
+            event=TurnResponseEvent(
+                payload=AgentTurnResponseStepProgressPayload(
+                    event_type="step_progress",
+                    step_type="inference",
+                    delta=TextDelta(text="answer", type="text"),
+                    step_id="s2",
+                )
+            )
+        ),
         AgentTurnResponseStreamChunk(
             event=TurnResponseEvent(
                 payload=AgentTurnResponseStepCompletePayload(
@@ -194,7 +204,7 @@ async def _test_streaming_query_endpoint_handler(mocker, store_transcript=False)
                     step_type="tool_execution",
                     step_details=ToolExecutionStep(
                         turn_id="t1",
-                        step_id="s2",
+                        step_id="s3",
                         step_type="tool_execution",
                         tool_responses=[
                             ToolResponse(
@@ -215,6 +225,27 @@ async def _test_streaming_query_endpoint_handler(mocker, store_transcript=False)
                 )
             )
         ),
+        AgentTurnResponseStreamChunk(
+            event=TurnResponseEvent(
+                payload=AgentTurnResponseTurnCompletePayload(
+                    event_type="turn_complete",
+                    turn=Turn(
+                        turn_id="t1",
+                        input_messages=[],
+                        output_message=CompletionMessage(
+                            role="assistant",
+                            content=[TextContentItem(text="LLM answer", type="text")],
+                            stop_reason="end_of_turn",
+                        ),
+                        session_id="test_session_id",
+                        started_at=datetime.now(),
+                        steps=[],
+                        completed_at=datetime.now(),
+                        output_attachments=[],
+                    ),
+                )
+            )
+        ),
     ]
 
     query = "What is OpenStack?"
@@ -263,8 +294,8 @@ async def _test_streaming_query_endpoint_handler(mocker, store_transcript=False)
     assert "LLM answer" in full_content
 
     # Assert referenced documents
-    assert len(streaming_content) == 5
-    d = json.loads(streaming_content[4][5:])
+    assert len(streaming_content) == 7
+    d = json.loads(streaming_content[6][5:])
     referenced_documents = d["data"]["referenced_documents"]
     assert len(referenced_documents) == 2
     assert referenced_documents[1]["doc_title"] == "Doc2"
@@ -277,8 +308,7 @@ async def _test_streaming_query_endpoint_handler(mocker, store_transcript=False)
             query_is_valid=True,
             query=query,
             query_request=query_request,
-            response="LLM answerTool:knowledge_search arguments:{}Tool:knowledge_search "
-            "summary:knowledge_search tool found 2 chunks:",
+            response="LLM answer",
             attachments=[],
             rag_chunks=[],
             truncated=False,
@@ -940,12 +970,12 @@ def test_stream_build_event_step_complete():
     assert result is not None
     assert "data: " in result
     assert '"event": "tool_call"' in result
-    assert '"token": "Tool:knowledge_search arguments:' in result
+    assert '"token": {"tool_name": "knowledge_search", "arguments": {}}' in result
 
     result = next(itr)
     assert (
-        '"token": "Tool:knowledge_search summary:knowledge_search tool found 2 chunks:"'
-        in result
+        '"token": {"tool_name": "knowledge_search", '
+        '"summary": "knowledge_search tool found 2 chunks:"}' in result
     )
     assert '"role": "tool_execution"' in result
     assert '"id": 0' in result