Add option to run models as local activities

temporalio/contrib/openai_agents/_model_parameters.py

@@ -70,3 +70,6 @@ class ModelActivityParameters:
 
     priority: Priority = Priority.default
     """Priority for the activity execution."""
+
+    use_local_activity: bool = False
+    """Whether to use a local activity. If changed during a workflow execution, that would break determinism."""

temporalio/contrib/openai_agents/_temporal_model_stub.py

@@ -154,20 +154,32 @@ def make_tool_info(tool: Tool) -> ToolInput:
         else:
             summary = None
 
-        return await workflow.execute_activity_method(
-            ModelActivity.invoke_model_activity,
-            activity_input,
-            summary=summary,
-            task_queue=self.model_params.task_queue,
-            schedule_to_close_timeout=self.model_params.schedule_to_close_timeout,
-            schedule_to_start_timeout=self.model_params.schedule_to_start_timeout,
-            start_to_close_timeout=self.model_params.start_to_close_timeout,
-            heartbeat_timeout=self.model_params.heartbeat_timeout,
-            retry_policy=self.model_params.retry_policy,
-            cancellation_type=self.model_params.cancellation_type,
-            versioning_intent=self.model_params.versioning_intent,
-            priority=self.model_params.priority,
-        )
+        if self.model_params.use_local_activity:
+            return await workflow.execute_local_activity_method(
+                ModelActivity.invoke_model_activity,
+                activity_input,
+                summary=summary,
+                schedule_to_close_timeout=self.model_params.schedule_to_close_timeout,
+                schedule_to_start_timeout=self.model_params.schedule_to_start_timeout,
+                start_to_close_timeout=self.model_params.start_to_close_timeout,
+                retry_policy=self.model_params.retry_policy,
+                cancellation_type=self.model_params.cancellation_type,
+            )
+        else:
+            return await workflow.execute_activity_method(
+                ModelActivity.invoke_model_activity,
+                activity_input,
+                summary=summary,
+                task_queue=self.model_params.task_queue,
+                schedule_to_close_timeout=self.model_params.schedule_to_close_timeout,
+                schedule_to_start_timeout=self.model_params.schedule_to_start_timeout,
+                start_to_close_timeout=self.model_params.start_to_close_timeout,
+                heartbeat_timeout=self.model_params.heartbeat_timeout,
+                retry_policy=self.model_params.retry_policy,
+                cancellation_type=self.model_params.cancellation_type,
+                versioning_intent=self.model_params.versioning_intent,
+                priority=self.model_params.priority,
+            )
 
     def stream_response(
         self,

tests/contrib/openai_agents/test_openai.py

@@ -2670,3 +2670,34 @@ async def test_model_conversion_loops():
     triage_agent = seat_booking_agent.handoffs[0]
     assert isinstance(triage_agent, Agent)
     assert isinstance(triage_agent.model, _TemporalModelStub)
+
+
+async def test_local_hello_world_agent(client: Client):
+    new_config = client.config()
+    new_config["plugins"] = [
+        openai_agents.OpenAIAgentsPlugin(
+            model_params=ModelActivityParameters(
+                start_to_close_timeout=timedelta(seconds=30),
+                use_local_activity=True,
+            ),
+            model_provider=TestModelProvider(TestHelloModel()),
+        )
+    ]
+    client = Client(**new_config)
+
+    async with new_worker(client, HelloWorldAgent) as worker:
+        handle = await client.start_workflow(
+            HelloWorldAgent.run,
+            "Tell me about recursion in programming.",
+            id=f"hello-workflow-{uuid.uuid4()}",
+            task_queue=worker.task_queue,
+            execution_timeout=timedelta(seconds=5),
+        )
+        result = await handle.result()
+        assert result == "test"
+
+        local_activity_found = False
+        async for e in handle.fetch_history_events():
+            if e.HasField("marker_recorded_event_attributes"):
+                local_activity_found = True
+        assert local_activity_found