Merge pull request #49 from chaizhenhua/fix/streaming-tool-call-index

fix: add missing index field to streaming tool_call chunks

Author: CaddyGlow

ccproxy/llms/formatters/anthropic_to_openai/_helpers.py

@@ -42,3 +42,33 @@ def build_openai_tool_call(
             arguments=str(args_str),
         ),
     )
+
+
+def build_openai_tool_call_chunk(
+    *,
+    index: int,
+    tool_id: str | None,
+    tool_name: str | None,
+    tool_input: Any,
+    arguments: Any = None,
+    fallback_index: int = 0,
+) -> openai_models.ToolCallChunk:
+    args_str = (
+        arguments
+        if isinstance(arguments, str) and arguments
+        else serialize_tool_arguments(tool_input)
+    )
+    call_id = (
+        tool_id if isinstance(tool_id, str) and tool_id else f"call_{fallback_index}"
+    )
+    name = tool_name if isinstance(tool_name, str) and tool_name else "function"
+
+    return openai_models.ToolCallChunk(
+        index=index,
+        id=str(call_id),
+        type="function",
+        function=openai_models.FunctionCall(
+            name=str(name),
+            arguments=str(args_str),
+        ),
+    )

ccproxy/llms/formatters/anthropic_to_openai/streams.py

@@ -27,7 +27,7 @@
 from ccproxy.llms.models import openai as openai_models
 from ccproxy.llms.streaming.accumulators import ClaudeAccumulator
 
-from ._helpers import build_openai_tool_call
+from ._helpers import build_openai_tool_call_chunk
 from .requests import _build_responses_payload_from_anthropic_request
 from .responses import convert__anthropic_usage_to_openai_responses__usage
 
@@ -88,10 +88,10 @@ def _anthropic_delta_to_text(
     return None
 
 
-def _build_openai_tool_call(
+def _build_openai_tool_call_chunk(
     accumulator: ClaudeAccumulator,
     block_index: int,
-) -> openai_models.ToolCall | None:
+) -> openai_models.ToolCallChunk | None:
     for tool_call in accumulator.get_complete_tool_calls():
         if tool_call.get("index") != block_index:
             continue
@@ -102,7 +102,8 @@ def _build_openai_tool_call(
         tool_name = function_payload.get("name") or tool_call.get("name")
         arguments = function_payload.get("arguments")
 
-        return build_openai_tool_call(
+        return build_openai_tool_call_chunk(
+            index=tool_call.get("index", block_index),
             tool_id=tool_call.get("id"),
             tool_name=tool_name,
             tool_input=tool_call.get("input", {}),
@@ -1413,7 +1414,7 @@ async def generator() -> AsyncGenerator[
                         continue
                     if block_index in emitted_tool_indices:
                         continue
-                    tool_call = _build_openai_tool_call(accumulator, block_index)
+                    tool_call = _build_openai_tool_call_chunk(accumulator, block_index)
                     if tool_call is None:
                         continue
                     emitted_tool_indices.add(block_index)

ccproxy/llms/formatters/openai_to_openai/streams.py

@@ -389,7 +389,8 @@ def create_text_chunk(
 
                     # Emit initial tool call chunk to surface id/name information
                     if not state.initial_emitted:
-                        tool_call = openai_models.ToolCall(
+                        tool_call = openai_models.ToolCallChunk(
+                            index=state.index,
                             id=state.id,
                             type="function",
                             function=openai_models.FunctionCall(
@@ -442,7 +443,8 @@ def create_text_chunk(
                             state.name = guessed
 
                     if state.initial_emitted:
-                        tool_call = openai_models.ToolCall(
+                        tool_call = openai_models.ToolCallChunk(
+                            index=state.index,
                             id=state.id,
                             type="function",
                             function=openai_models.FunctionCall(
@@ -494,7 +496,8 @@ def create_text_chunk(
                             if guessed:
                                 state.name = guessed
 
-                        tool_call = openai_models.ToolCall(
+                        tool_call = openai_models.ToolCallChunk(
+                            index=state.index,
                             id=state.id,
                             type="function",
                             function=openai_models.FunctionCall(
@@ -586,7 +589,8 @@ def create_text_chunk(
                             if guessed:
                                 state.name = guessed
                         if not state.arguments_emitted:
-                            tool_call = openai_models.ToolCall(
+                            tool_call = openai_models.ToolCallChunk(
+                                index=state.index,
                                 id=state.id,
                                 type="function",
                                 function=openai_models.FunctionCall(
@@ -616,7 +620,8 @@ def create_text_chunk(
 
                     # Emit a patch chunk if the name was never surfaced earlier
                     if state.name and not state.name_emitted:
-                        tool_call = openai_models.ToolCall(
+                        tool_call = openai_models.ToolCallChunk(
+                            index=state.index,
                             id=state.id,
                             type="function",
                             function=openai_models.FunctionCall(

ccproxy/llms/models/openai.py

@@ -185,11 +185,22 @@ class FunctionCall(LlmBaseModel):
 
 
 class ToolCall(LlmBaseModel):
+    """Non-streaming tool call (ChatCompletionMessageToolCall)."""
+
     id: str
     type: Literal["function"] = Field(default="function")
     function: FunctionCall
 
 
+class ToolCallChunk(LlmBaseModel):
+    """Streaming tool call delta (ChoiceDeltaToolCall)."""
+
+    index: int
+    id: str | None = None
+    type: Literal["function"] | None = None
+    function: FunctionCall | None = None
+
+
 class ChatMessage(LlmBaseModel):
     """
     A message within a chat conversation.
@@ -309,7 +320,7 @@ class ChatCompletionResponse(LlmBaseModel):
 class DeltaMessage(LlmBaseModel):
     role: Literal["assistant"] | None = None
     content: str | list[Any] | None = None
-    tool_calls: list[ToolCall] | None = None
+    tool_calls: list[ToolCallChunk] | None = None
     audio: dict[str, Any] | None = None
     reasoning: ResponseMessageReasoning | None = None