From 5197f505703c653f846e669097c568aed10c2524 Mon Sep 17 00:00:00 2001
From: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Date: Fri, 17 Jul 2026 11:06:04 +0000
Subject: [PATCH] fix(ollama_chat): preserve per-tool-call index in streaming
 deltas

---
 litellm/llms/ollama/chat/transformation.py    |  6 +-
 .../ollama/test_ollama_chat_transformation.py | 85 +++++++++++++++++++
 2 files changed, 90 insertions(+), 1 deletion(-)

diff --git a/litellm/llms/ollama/chat/transformation.py b/litellm/llms/ollama/chat/transformation.py
index 694f8cdd6c2..f9927ba237c 100644
--- a/litellm/llms/ollama/chat/transformation.py
+++ b/litellm/llms/ollama/chat/transformation.py
@@ -475,7 +475,11 @@ class OllamaChatCompletionResponseIterator(BaseModelResponseIterator):
             tool_calls = chunk["message"].get("tool_calls")
             if tool_calls is not None:
                 for tool_call in tool_calls:
-                    function_args = tool_call.get("function").get("arguments")
+                    function = tool_call.get("function") or {}
+                    function_index = function.get("index")
+                    if function_index is not None:
+                        tool_call["index"] = function_index
+                    function_args = function.get("arguments")
                     if function_args is not None and len(function_args) > 0:
                         is_function_call_complete = self._is_function_call_complete(function_args)
                         if is_function_call_complete:
diff --git a/tests/test_litellm/llms/ollama/test_ollama_chat_transformation.py b/tests/test_litellm/llms/ollama/test_ollama_chat_transformation.py
index 906c51d8064..e22963bca31 100644
--- a/tests/test_litellm/llms/ollama/test_ollama_chat_transformation.py
+++ b/tests/test_litellm/llms/ollama/test_ollama_chat_transformation.py
@@ -618,6 +618,91 @@ class TestOllamaFinishReasonLength:
         ), f"Expected 'stop' for natural finish, got '{result.choices[0].finish_reason}'"
 
 
+class TestOllamaStreamingToolCallIndex:
+    """Streaming: parallel tool calls must preserve Ollama's per-call function.index."""
+
+    def test_parallel_tool_calls_preserve_distinct_index(self):
+        """
+        Ollama streams one tool call per chunk with the correct index nested at
+        function.index (0, then 1). Previously LiteLLM dropped it and every delta
+        was emitted with index=0, so clients that accumulate arguments by index
+        merged parallel calls into one malformed tool call.
+        """
+        iterator = OllamaChatCompletionResponseIterator(
+            streaming_response=iter([]),
+            sync_stream=True,
+        )
+
+        first_chunk = {
+            "model": "qwen3:14b",
+            "message": {
+                "role": "assistant",
+                "content": "",
+                "tool_calls": [
+                    {
+                        "function": {
+                            "index": 0,
+                            "name": "read_file",
+                            "arguments": {"path": "a.rs"},
+                        }
+                    }
+                ],
+            },
+            "done": False,
+        }
+        second_chunk = {
+            "model": "qwen3:14b",
+            "message": {
+                "role": "assistant",
+                "content": "",
+                "tool_calls": [
+                    {
+                        "function": {
+                            "index": 1,
+                            "name": "read_file",
+                            "arguments": {"path": "b.rs"},
+                        }
+                    }
+                ],
+            },
+            "done": False,
+        }
+
+        first_result = iterator.chunk_parser(first_chunk)
+        second_result = iterator.chunk_parser(second_chunk)
+
+        assert first_result.choices[0].delta.tool_calls[0].index == 0
+        assert second_result.choices[0].delta.tool_calls[0].index == 1
+
+    def test_missing_function_index_falls_back_to_positional(self):
+        """When Ollama omits function.index, Delta still assigns a positional index."""
+        iterator = OllamaChatCompletionResponseIterator(
+            streaming_response=iter([]),
+            sync_stream=True,
+        )
+
+        chunk = {
+            "model": "qwen3:14b",
+            "message": {
+                "role": "assistant",
+                "content": "",
+                "tool_calls": [
+                    {
+                        "function": {
+                            "name": "read_file",
+                            "arguments": {"path": "a.rs"},
+                        }
+                    }
+                ],
+            },
+            "done": False,
+        }
+
+        result = iterator.chunk_parser(chunk)
+
+        assert result.choices[0].delta.tool_calls[0].index == 0
+
+
 class TestOllamaReasoningContentStreaming:
     """Test that reasoning_content is properly extracted from all thinking chunks."""