fix(litellm): generate fallback ID for null tool_call IDs

src/strands/models/litellm.py

@@ -5,6 +5,7 @@
 
 import json
 import logging
+import uuid
 from collections.abc import AsyncGenerator
 from typing import Any, TypedDict, TypeVar, cast
 
@@ -445,7 +446,13 @@ async def _process_tool_calls(self, tool_calls: dict[int, list[Any]]) -> AsyncGe
             Formatted tool call chunks.
         """
         for tool_deltas in tool_calls.values():
-            yield self.format_chunk({"chunk_type": "content_start", "data_type": "tool", "data": tool_deltas[0]})
+            # Some LiteLLM proxy backends return null tool_call IDs.
+            # Ensure the first delta (used for content_start) has a valid ID.
+            first = tool_deltas[0]
+            if getattr(first, "id", None) is None:
+                first.id = f"tooluse_{uuid.uuid4().hex[:24]}"
+
+            yield self.format_chunk({"chunk_type": "content_start", "data_type": "tool", "data": first})
 
             for tool_delta in tool_deltas:
                 yield self.format_chunk({"chunk_type": "content_delta", "data_type": "tool", "data": tool_delta})

tests/strands/models/test_litellm.py

@@ -848,3 +848,64 @@ def test_format_request_messages_with_tool_calls_no_content():
         },
     ]
     assert tru_result == exp_result
+
+
+@pytest.mark.asyncio
+async def test_stream_non_streaming_null_tool_call_id(litellm_acompletion, api_key, model_id, alist):
+    """Verify that null tool_call IDs from proxy backends get a generated fallback.
+
+    Some LiteLLM proxy backends (e.g. running behind a custom proxy server)
+    return tool_calls with id=None, causing downstream failures when
+    the event loop tries to match tool results back to tool_call IDs.
+
+    See: https://github.com/strands-agents/sdk-python/issues/1259
+    """
+    mock_function = unittest.mock.Mock()
+    mock_function.name = "search"
+    mock_function.arguments = '{"query": "test"}'
+
+    mock_tool_call = unittest.mock.Mock(index=0, function=mock_function, id=None)
+
+    mock_message = unittest.mock.Mock()
+    mock_message.content = None
+    mock_message.reasoning_content = None
+    mock_message.tool_calls = [mock_tool_call]
+
+    mock_choice = unittest.mock.Mock()
+    mock_choice.message = mock_message
+    mock_choice.finish_reason = "tool_calls"
+
+    mock_response = unittest.mock.Mock()
+    mock_response.choices = [mock_choice]
+
+    mock_usage = unittest.mock.Mock()
+    mock_usage.prompt_tokens = 5
+    mock_usage.completion_tokens = 10
+    mock_usage.total_tokens = 15
+    mock_usage.prompt_tokens_details = None
+    mock_usage.cache_creation_input_tokens = None
+    mock_response.usage = mock_usage
+
+    litellm_acompletion.side_effect = unittest.mock.AsyncMock(return_value=mock_response)
+
+    model = LiteLLMModel(
+        client_args={"api_key": api_key},
+        model_id=model_id,
+        params={"stream": False},
+    )
+
+    messages = [{"role": "user", "content": [{"type": "text", "text": "search test"}]}]
+    response = model.stream(messages)
+    events = await alist(response)
+
+    # Find the content_start event for the tool
+    tool_start = [
+        e for e in events if "contentBlockStart" in e and "toolUse" in e["contentBlockStart"].get("start", {})
+    ]
+    assert len(tool_start) == 1
+
+    tool_use_id = tool_start[0]["contentBlockStart"]["start"]["toolUse"]["toolUseId"]
+    assert tool_use_id is not None
+    assert isinstance(tool_use_id, str)
+    assert len(tool_use_id) > 0
+    assert tool_use_id.startswith("tooluse_")