fix(stream): preserve non-negative tool call indexes

src/api/models/bedrock.py

@@ -421,6 +421,8 @@ async def chat_stream(self, chat_request: ChatRequest) -> AsyncIterable[bytes]:
             message_id = self.generate_message_id()
             stream = response.get("stream")
             self.think_emitted = False
+            self._stream_tool_index_by_content_block = {}
+            self._next_stream_tool_index = 0
             reasoning_tokens = 0
             async for chunk in self._async_iterate(stream):
                 # Accumulate reasoning tokens from delta chunks before processing
@@ -457,6 +459,8 @@ async def chat_stream(self, chat_request: ChatRequest) -> AsyncIterable[bytes]:
             # return an [DONE] message at the end.
             yield self.stream_response_to_bytes()
             self.think_emitted = False  # Cleanup
+            self._stream_tool_index_by_content_block = {}
+            self._next_stream_tool_index = 0
         except Exception as e:
             logger.error("Stream error for model %s: %s", chat_request.model, str(e))
             error_event = Error(error=ErrorMessage(message=str(e)))
@@ -1026,8 +1030,13 @@ def _create_response_stream(
             # tool call start
             delta = chunk["contentBlockStart"]["start"]
             if "toolUse" in delta:
-                # first index is content
-                index = chunk["contentBlockStart"]["contentBlockIndex"] - 1
+                content_block_index = chunk["contentBlockStart"]["contentBlockIndex"]
+                tool_index_map = getattr(self, "_stream_tool_index_by_content_block", {})
+                if content_block_index not in tool_index_map:
+                    tool_index_map[content_block_index] = getattr(self, "_next_stream_tool_index", 0)
+                    self._next_stream_tool_index = tool_index_map[content_block_index] + 1
+                    self._stream_tool_index_by_content_block = tool_index_map
+                index = tool_index_map[content_block_index]
                 message = ChatResponseMessage(
                     tool_calls=[
                         ToolCall(
@@ -1069,7 +1078,13 @@ def _create_response_stream(
                         return None  # Ignore signature if no <think> started
             else:
                 # tool use
-                index = chunk["contentBlockDelta"]["contentBlockIndex"] - 1
+                content_block_index = chunk["contentBlockDelta"]["contentBlockIndex"]
+                tool_index_map = getattr(self, "_stream_tool_index_by_content_block", {})
+                if content_block_index not in tool_index_map:
+                    tool_index_map[content_block_index] = getattr(self, "_next_stream_tool_index", 0)
+                    self._next_stream_tool_index = tool_index_map[content_block_index] + 1
+                    self._stream_tool_index_by_content_block = tool_index_map
+                index = tool_index_map[content_block_index]
                 message = ChatResponseMessage(
                     tool_calls=[
                         ToolCall(

tests/test_bedrock_stream_tool_call_index.py

@@ -0,0 +1,90 @@
+import unittest
+
+from api.models.bedrock import BedrockModel
+
+
+class BedrockStreamToolCallIndexTest(unittest.TestCase):
+    def setUp(self):
+        self.model = BedrockModel()
+        self.model.think_emitted = False
+        self.model._stream_tool_index_by_content_block = {}
+        self.model._next_stream_tool_index = 0
+
+    def test_first_streamed_tool_call_uses_non_negative_index(self):
+        start_chunk = {
+            "contentBlockStart": {
+                "contentBlockIndex": 0,
+                "start": {
+                    "toolUse": {
+                        "toolUseId": "tooluse_123",
+                        "name": "get_weather",
+                    }
+                },
+            }
+        }
+        delta_chunk = {
+            "contentBlockDelta": {
+                "contentBlockIndex": 0,
+                "delta": {
+                    "toolUse": {
+                        "input": "{\"city\":\"Boston\"}",
+                    }
+                },
+            }
+        }
+
+        start_response = self.model._create_response_stream(
+            model_id="us.anthropic.claude-sonnet-4-6",
+            message_id="chatcmpl-test",
+            chunk=start_chunk,
+        )
+        delta_response = self.model._create_response_stream(
+            model_id="us.anthropic.claude-sonnet-4-6",
+            message_id="chatcmpl-test",
+            chunk=delta_chunk,
+        )
+
+        self.assertEqual(start_response.choices[0].delta.tool_calls[0].index, 0)
+        self.assertEqual(delta_response.choices[0].delta.tool_calls[0].index, 0)
+
+    def test_multiple_streamed_tool_calls_are_indexed_by_tool_ordinal(self):
+        tool_one_start = {
+            "contentBlockStart": {
+                "contentBlockIndex": 0,
+                "start": {
+                    "toolUse": {
+                        "toolUseId": "tooluse_1",
+                        "name": "get_weather",
+                    }
+                },
+            }
+        }
+        tool_two_start = {
+            "contentBlockStart": {
+                "contentBlockIndex": 2,
+                "start": {
+                    "toolUse": {
+                        "toolUseId": "tooluse_2",
+                        "name": "lookup_news",
+                    }
+                },
+            }
+        }
+
+        tool_one_response = self.model._create_response_stream(
+            model_id="us.anthropic.claude-sonnet-4-6",
+            message_id="chatcmpl-test",
+            chunk=tool_one_start,
+        )
+        tool_two_response = self.model._create_response_stream(
+            model_id="us.anthropic.claude-sonnet-4-6",
+            message_id="chatcmpl-test",
+            chunk=tool_two_start,
+        )
+
+        self.assertEqual(tool_one_response.choices[0].delta.tool_calls[0].index, 0)
+        self.assertEqual(tool_two_response.choices[0].delta.tool_calls[0].index, 1)
+
+
+if __name__ == "__main__":
+    unittest.main()