Wire through parallel_tool_calls to Responses API

Signed-off-by: Anastas Stoyanovsky <astoyano@redhat.com>

src/llama_stack/providers/inline/agents/meta_reference/agents.py

@@ -92,6 +92,7 @@ class MetaReferenceAgentsImpl(Agents):
         model: str,
         prompt: OpenAIResponsePrompt | None = None,
         instructions: str | None = None,
+        parallel_tool_calls: bool | None = True,
         previous_response_id: str | None = None,
         conversation: str | None = None,
         store: bool | None = True,
@@ -120,6 +121,7 @@ class MetaReferenceAgentsImpl(Agents):
             include,
             max_infer_iters,
             guardrails,
+            parallel_tool_calls,
             max_tool_calls,
         )
         return result  # type: ignore[no-any-return]

src/llama_stack/providers/inline/agents/meta_reference/responses/openai_responses.py

@@ -252,6 +252,7 @@ class OpenAIResponsesImpl:
         include: list[str] | None = None,
         max_infer_iters: int | None = 10,
         guardrails: list[str | ResponseGuardrailSpec] | None = None,
+        parallel_tool_calls: bool | None = None,
         max_tool_calls: int | None = None,
     ):
         stream = bool(stream)
@@ -296,6 +297,7 @@ class OpenAIResponsesImpl:
             tools=tools,
             max_infer_iters=max_infer_iters,
             guardrail_ids=guardrail_ids,
+            parallel_tool_calls=parallel_tool_calls,
             max_tool_calls=max_tool_calls,
         )
 
@@ -346,6 +348,7 @@ class OpenAIResponsesImpl:
         tools: list[OpenAIResponseInputTool] | None = None,
         max_infer_iters: int | None = 10,
         guardrail_ids: list[str] | None = None,
+        parallel_tool_calls: bool | None = True,
         max_tool_calls: int | None = None,
     ) -> AsyncIterator[OpenAIResponseObjectStream]:
         # These should never be None when called from create_openai_response (which sets defaults)
@@ -385,6 +388,7 @@ class OpenAIResponsesImpl:
             created_at=created_at,
             text=text,
             max_infer_iters=max_infer_iters,
+            parallel_tool_calls=parallel_tool_calls,
             tool_executor=self.tool_executor,
             safety_api=self.safety_api,
             guardrail_ids=guardrail_ids,

src/llama_stack/providers/inline/agents/meta_reference/responses/streaming.py

@@ -114,6 +114,7 @@ class StreamingResponseOrchestrator:
         safety_api,
         guardrail_ids: list[str] | None = None,
         prompt: OpenAIResponsePrompt | None = None,
+        parallel_tool_calls: bool | None = None,
         max_tool_calls: int | None = None,
     ):
         self.inference_api = inference_api
@@ -128,6 +129,8 @@ class StreamingResponseOrchestrator:
         self.prompt = prompt
         # System message that is inserted into the model's context
         self.instructions = instructions
+        # Whether to allow more than one function tool call generated per turn.
+        self.parallel_tool_calls = parallel_tool_calls
         # Max number of total calls to built-in tools that can be processed in a response
         self.max_tool_calls = max_tool_calls
         self.sequence_number = 0
@@ -190,6 +193,7 @@ class StreamingResponseOrchestrator:
             usage=self.accumulated_usage,
             instructions=self.instructions,
             prompt=self.prompt,
+            parallel_tool_calls=self.parallel_tool_calls,
             max_tool_calls=self.max_tool_calls,
         )
 
@@ -301,6 +305,7 @@ class StreamingResponseOrchestrator:
                     completion_result_data,
                     output_messages,
                     next_turn_messages,
+                    not self.parallel_tool_calls,
                 ):
                     yield stream_event
 
@@ -897,6 +902,7 @@ class StreamingResponseOrchestrator:
         completion_result_data: ChatCompletionResult,
         output_messages: list[OpenAIResponseOutput],
         next_turn_messages: list,
+        incremental_function_calling: bool,
     ) -> AsyncIterator[OpenAIResponseObjectStream]:
         """Coordinate execution of both function and non-function tool calls."""
         # Execute non-function tool calls
@@ -1020,6 +1026,10 @@ class StreamingResponseOrchestrator:
                 sequence_number=self.sequence_number,
             )
 
+            # TODO: Make sure that multi-turn incremental execution works
+            if incremental_function_calling:
+                break
+
     async def _process_new_tools(
         self, tools: list[OpenAIResponseInputTool], output_messages: list[OpenAIResponseOutput]
     ) -> AsyncIterator[OpenAIResponseObjectStream]:

src/llama_stack_api/agents.py

@@ -72,6 +72,7 @@ class Agents(Protocol):
         model: str,
         prompt: OpenAIResponsePrompt | None = None,
         instructions: str | None = None,
+        parallel_tool_calls: bool | None = True,
         previous_response_id: str | None = None,
         conversation: str | None = None,
         store: bool | None = True,

src/llama_stack_api/openai_responses.py

@@ -585,7 +585,7 @@ class OpenAIResponseObject(BaseModel):
     :param model: Model identifier used for generation
     :param object: Object type identifier, always "response"
     :param output: List of generated output items (messages, tool calls, etc.)
-    :param parallel_tool_calls: Whether tool calls can be executed in parallel
+    :param parallel_tool_calls: (Optional) Whether to allow more than one function tool call generated per turn.
     :param previous_response_id: (Optional) ID of the previous response in a conversation
     :param prompt: (Optional) Reference to a prompt template and its variables.
     :param status: Current status of the response generation
@@ -605,7 +605,7 @@ class OpenAIResponseObject(BaseModel):
     model: str
     object: Literal["response"] = "response"
     output: Sequence[OpenAIResponseOutput]
-    parallel_tool_calls: bool = False
+    parallel_tool_calls: bool | None = True
     previous_response_id: str | None = None
     prompt: OpenAIResponsePrompt | None = None
     status: str