fix(responses): correlate streamed tool call events on normalized item ids

litellm/litellm_core_utils/prompt_templates/factory.py

@@ -1696,9 +1696,9 @@ def convert_function_to_anthropic_tool_invoke(
 
 def _find_server_tool_result(
     tool_id: str,
-    web_search_results: Sequence[Any] | None,
-    tool_results: Sequence[Any] | None,
-) -> dict[str, Any] | None:
+    web_search_results: Sequence[object] | None,
+    tool_results: Sequence[object] | None,
+) -> dict[str, object] | None:
     candidates: Final = (*(web_search_results or ()), *(tool_results or ()))
     return next(
         (result for result in candidates if isinstance(result, dict) and result.get("tool_use_id") == tool_id),

litellm/responses/litellm_completion_transformation/streaming_iterator.py

@@ -114,6 +114,7 @@ class LiteLLMCompletionStreamingIterator(ResponsesAPIStreamingIterator):
         self._pending_tool_events: list[BaseLiteLLMOpenAIResponseObject] = []
         self._tool_output_index_by_call_id: dict[str, int] = {}
         self._tool_args_by_call_id: dict[str, str] = {}
+        self._tool_item_id_by_call_id: dict[str, str] = {}  # mutable-ok: filled per call id as tool call events stream
         self._tool_call_id_by_index: dict[int, str] = {}
         self._ambiguous_tool_call_indexes: set[int] = set()
         self._next_tool_output_index: int = 1  # output_index=0 reserved for the message item
@@ -227,6 +228,7 @@ class LiteLLMCompletionStreamingIterator(ResponsesAPIStreamingIterator):
                 self._sequence_number += 1
                 names = self._custom_tool_names
                 item_kwargs = build_tool_call_item_kwargs(call_id, tool_name, "", "in_progress", names)
+                self._tool_item_id_by_call_id[call_id] = item_kwargs["id"]
                 if tool_namespace:
                     item_kwargs["namespace"] = tool_namespace
                 event = OutputItemAddedEvent(
@@ -248,7 +250,7 @@ class LiteLLMCompletionStreamingIterator(ResponsesAPIStreamingIterator):
                     self._sequence_number += 1
                     delta_event: BaseLiteLLMOpenAIResponseObject = FunctionCallArgumentsDeltaEvent(
                         type=ResponsesAPIStreamEvents.FUNCTION_CALL_ARGUMENTS_DELTA,
-                        item_id=call_id,
+                        item_id=self._tool_item_id_by_call_id.get(call_id, call_id),
                         output_index=output_index,
                         delta=delta_chunk,
                     )
@@ -300,6 +302,7 @@ class LiteLLMCompletionStreamingIterator(ResponsesAPIStreamingIterator):
                 self._sequence_number += 1
                 names = self._custom_tool_names
                 item_kwargs = build_tool_call_item_kwargs(call_id, tool_name, "", "in_progress", names)
+                self._tool_item_id_by_call_id[call_id] = item_kwargs["id"]
                 if tool_namespace:
                     item_kwargs["namespace"] = tool_namespace
                 event = OutputItemAddedEvent(
@@ -325,7 +328,7 @@ class LiteLLMCompletionStreamingIterator(ResponsesAPIStreamingIterator):
                     self._sequence_number += 1
                     delta_event = FunctionCallArgumentsDeltaEvent(
                         type=ResponsesAPIStreamEvents.FUNCTION_CALL_ARGUMENTS_DELTA,
-                        item_id=call_id,
+                        item_id=self._tool_item_id_by_call_id.get(call_id, call_id),
                         output_index=output_index,
                         delta=delta_chunk,
                     )
@@ -335,7 +338,7 @@ class LiteLLMCompletionStreamingIterator(ResponsesAPIStreamingIterator):
             self._sequence_number += 1
             done_event = FunctionCallArgumentsDoneEvent(
                 type=ResponsesAPIStreamEvents.FUNCTION_CALL_ARGUMENTS_DONE,
-                item_id=call_id,
+                item_id=self._tool_item_id_by_call_id.get(call_id, call_id),
                 output_index=output_index,
                 arguments=final_args,
             )
@@ -345,6 +348,7 @@ class LiteLLMCompletionStreamingIterator(ResponsesAPIStreamingIterator):
             self._sequence_number += 1
             names = self._custom_tool_names
             item_kwargs = build_tool_call_item_kwargs(call_id, tool_name, final_args, "completed", names)
+            item_kwargs["id"] = self._tool_item_id_by_call_id.setdefault(call_id, item_kwargs["id"])
             if tool_namespace:
                 item_kwargs["namespace"] = tool_namespace
             item_done_event = OutputItemDoneEvent(

tests/test_litellm/responses/litellm_completion_transformation/test_litellm_completion_responses.py

@@ -3409,6 +3409,7 @@ class TestEnsureOutputItemContentPartAdded:
         iterator._pending_tool_events = []
         iterator._tool_output_index_by_call_id = {}
         iterator._tool_args_by_call_id = {}
+        iterator._tool_item_id_by_call_id = {}
         iterator._tool_call_id_by_index = {}
         iterator._ambiguous_tool_call_indexes = set()
         iterator._next_tool_output_index = 1

tests/test_litellm/responses/litellm_completion_transformation/test_streaming_iterator_transformation.py

@@ -132,7 +132,7 @@ def test_tool_call_delta_is_emitted_as_responses_events():
     evt2 = iterator._transform_chat_completion_chunk_to_response_api_chunk(chunk)
     assert evt2 is not None
     assert evt2.type == ResponsesAPIStreamEvents.FUNCTION_CALL_ARGUMENTS_DELTA
-    assert evt2.item_id == "call_1"
+    assert evt2.item_id == "fc_call_1"
     assert evt2.output_index == 1
     # The delta will be a chunk of the arguments, not the full arguments
     assert len(evt2.delta) <= 10  # Chunks are max 10 characters
@@ -197,7 +197,7 @@ def test_tool_calls_present_only_in_final_response_are_emitted_before_completed(
 
     # The last event should be FUNCTION_CALL_ARGUMENTS_DONE
     assert evt.type == ResponsesAPIStreamEvents.FUNCTION_CALL_ARGUMENTS_DONE
-    assert evt.item_id == "call_2"
+    assert evt.item_id == "fc_call_2"
     assert evt.output_index == 1
     assert evt.arguments == '{"y":2}'
 
@@ -291,7 +291,7 @@ def test_tool_call_arguments_are_chunked_to_match_openai_behavior():
     # Verify each delta is at most 10 characters
     for evt in delta_events:
         assert len(evt.delta) <= 10
-        assert evt.item_id == "call_test"
+        assert evt.item_id == "fc_call_test"
         assert evt.output_index == 1
         assert hasattr(evt, "__dict__") and "sequence_number" in evt.__dict__
 
@@ -405,8 +405,8 @@ def test_parallel_tool_calls_without_ids_use_index_mapping():
         arguments_by_call_id.setdefault(evt.item_id, "")
         arguments_by_call_id[evt.item_id] += evt.delta
 
-    assert arguments_by_call_id["call_a"] == '{"x":1}'
-    assert arguments_by_call_id["call_b"] == '{"y":2}'
+    assert arguments_by_call_id["fc_call_a"] == '{"x":1}'
+    assert arguments_by_call_id["fc_call_b"] == '{"y":2}'
 
 
 def test_reused_index_with_new_call_id_marks_fallback_ambiguous():
@@ -462,10 +462,10 @@ def test_reused_index_with_new_call_id_marks_fallback_ambiguous():
         arguments_by_call_id.setdefault(evt.item_id, "")
         arguments_by_call_id[evt.item_id] += evt.delta
 
-    assert arguments_by_call_id["call_a"] == '{"a":'
-    assert arguments_by_call_id["call_b"] == '{"b":'
-    assert arguments_by_call_id["call_a"] != '{"a":1}'
-    assert arguments_by_call_id["call_b"] != '{"b":1}'
+    assert arguments_by_call_id["fc_call_a"] == '{"a":'
+    assert arguments_by_call_id["fc_call_b"] == '{"b":'
+    assert arguments_by_call_id["fc_call_a"] != '{"a":1}'
+    assert arguments_by_call_id["fc_call_b"] != '{"b":1}'
 
 
 @pytest.mark.asyncio
@@ -558,3 +558,58 @@ def test_object_tool_call_arguments_stream_as_valid_json():
     )
 
     assert json.loads(streamed_arguments) == {"command": "ls", "flags": ["-l"]}
+
+
+def test_streamed_anthropic_tool_call_events_correlate_on_normalized_item_id():
+    iterator = LiteLLMCompletionStreamingIterator(
+        model="test-model",
+        litellm_custom_stream_wrapper=AsyncMock(),
+        request_input="Test input",
+        responses_api_request={},
+    )
+
+    response = ModelResponse(
+        id="resp-anthropic",
+        created=123,
+        model="test-model",
+        object="chat.completion",
+        choices=[
+            {
+                "index": 0,
+                "finish_reason": "tool_calls",
+                "message": {
+                    "role": "assistant",
+                    "content": None,
+                    "tool_calls": [
+                        {
+                            "id": "toolu_01AbCdEf",
+                            "type": "function",
+                            "function": {"name": "get_weather", "arguments": '{"city":"Paris"}'},
+                            "index": 0,
+                        }
+                    ],
+                },
+            }
+        ],
+    )
+    iterator.litellm_model_response = response
+
+    events = []
+    while True:
+        evt = iterator.common_done_event_logic(sync_mode=True)
+        events.append(evt)
+        if evt.type == ResponsesAPIStreamEvents.OUTPUT_ITEM_DONE:
+            break
+
+    added = [e for e in events if e.type == ResponsesAPIStreamEvents.OUTPUT_ITEM_ADDED]
+    deltas = [e for e in events if e.type == ResponsesAPIStreamEvents.FUNCTION_CALL_ARGUMENTS_DELTA]
+    dones = [e for e in events if e.type == ResponsesAPIStreamEvents.FUNCTION_CALL_ARGUMENTS_DONE]
+    item_dones = [e for e in events if e.type == ResponsesAPIStreamEvents.OUTPUT_ITEM_DONE]
+
+    assert len(added) == 1 and len(dones) == 1 and len(item_dones) == 1 and deltas
+    assert added[0].item.id == "fc_toolu_01AbCdEf"
+    assert added[0].item.call_id == "toolu_01AbCdEf"
+    assert item_dones[0].item.id == "fc_toolu_01AbCdEf"
+    assert item_dones[0].item.call_id == "toolu_01AbCdEf"
+    for evt in deltas + dones:
+        assert evt.item_id == added[0].item.id