openai-agents 0.4.1__py3-none-any.whl → 0.4.2__py3-none-any.whl

agents/extensions/models/litellm_model.py

@@ -110,18 +110,26 @@ class LitellmModel(Model):
                 prompt=prompt,
             )
 
-            assert isinstance(response.choices[0], litellm.types.utils.Choices)
+            message: litellm.types.utils.Message | None = None
+            first_choice: litellm.types.utils.Choices | None = None
+            if response.choices and len(response.choices) > 0:
+                choice = response.choices[0]
+                if isinstance(choice, litellm.types.utils.Choices):
+                    first_choice = choice
+                    message = first_choice.message
 
             if _debug.DONT_LOG_MODEL_DATA:
                 logger.debug("Received model response")
             else:
-                logger.debug(
-                    f"""LLM resp:\n{
-                        json.dumps(
-                            response.choices[0].message.model_dump(), indent=2, ensure_ascii=False
-                        )
-                    }\n"""
-                )
+                if message is not None:
+                    logger.debug(
+                        f"""LLM resp:\n{
+                            json.dumps(message.model_dump(), indent=2, ensure_ascii=False)
+                        }\n"""
+                    )
+                else:
+                    finish_reason = first_choice.finish_reason if first_choice else "-"
+                    logger.debug(f"LLM resp had no message. finish_reason: {finish_reason}")
 
             if hasattr(response, "usage"):
                 response_usage = response.usage
@@ -152,14 +160,20 @@ class LitellmModel(Model):
                 logger.warning("No usage information returned from Litellm")
 
             if tracing.include_data():
-                span_generation.span_data.output = [response.choices[0].message.model_dump()]
+                span_generation.span_data.output = (
+                    [message.model_dump()] if message is not None else []
+                )
             span_generation.span_data.usage = {
                 "input_tokens": usage.input_tokens,
                 "output_tokens": usage.output_tokens,
             }
 
-            items = Converter.message_to_output_items(
-                LitellmConverter.convert_message_to_openai(response.choices[0].message)
+            items = (
+                Converter.message_to_output_items(
+                    LitellmConverter.convert_message_to_openai(message)
+                )
+                if message is not None
+                else []
             )
 
             return ModelResponse(
@@ -326,6 +340,23 @@ class LitellmModel(Model):
             )
 
         reasoning_effort = model_settings.reasoning.effort if model_settings.reasoning else None
+        # Enable developers to pass non-OpenAI compatible reasoning_effort data like "none"
+        # Priority order:
+        #  1. model_settings.reasoning.effort
+        #  2. model_settings.extra_body["reasoning_effort"]
+        #  3. model_settings.extra_args["reasoning_effort"]
+        if (
+            reasoning_effort is None  # Unset in model_settings
+            and isinstance(model_settings.extra_body, dict)
+            and "reasoning_effort" in model_settings.extra_body
+        ):
+            reasoning_effort = model_settings.extra_body["reasoning_effort"]
+        if (
+            reasoning_effort is None  # Unset in both model_settings and model_settings.extra_body
+            and model_settings.extra_args
+            and "reasoning_effort" in model_settings.extra_args
+        ):
+            reasoning_effort = model_settings.extra_args["reasoning_effort"]
 
         stream_options = None
         if stream and model_settings.include_usage is not None:
@@ -343,6 +374,9 @@ class LitellmModel(Model):
         if model_settings.extra_args:
             extra_kwargs.update(model_settings.extra_args)
 
+        # Prevent duplicate reasoning_effort kwargs when it was promoted to a top-level argument.
+        extra_kwargs.pop("reasoning_effort", None)
+
         ret = await litellm.acompletion(
             model=self.model,
             messages=converted_messages,

agents/models/chatcmpl_stream_handler.py

@@ -150,6 +150,12 @@ class ChatCmplStreamHandler:
                     )
 
                 if reasoning_content and state.reasoning_content_index_and_output:
+                    # Ensure summary list has at least one element
+                    if not state.reasoning_content_index_and_output[1].summary:
+                        state.reasoning_content_index_and_output[1].summary = [
+                            Summary(text="", type="summary_text")
+                        ]
+
                     yield ResponseReasoningSummaryTextDeltaEvent(
                         delta=reasoning_content,
                         item_id=FAKE_RESPONSES_ID,
@@ -201,7 +207,7 @@ class ChatCmplStreamHandler:
                     )
 
                     # Create a new summary with updated text
-                    if state.reasoning_content_index_and_output[1].content is None:
+                    if not state.reasoning_content_index_and_output[1].content:
                         state.reasoning_content_index_and_output[1].content = [
                             Content(text="", type="reasoning_text")
                         ]

agents/realtime/config.py

@@ -184,6 +184,9 @@ class RealtimeRunConfig(TypedDict):
     tracing_disabled: NotRequired[bool]
     """Whether tracing is disabled for this run."""
 
+    async_tool_calls: NotRequired[bool]
+    """Whether function tool calls should run asynchronously. Defaults to True."""
+
     # TODO (rm) Add history audio storage config
 
 

agents/realtime/session.py

@@ -112,7 +112,7 @@ class RealtimeSession(RealtimeModelListener):
         }
         self._event_queue: asyncio.Queue[RealtimeSessionEvent] = asyncio.Queue()
         self._closed = False
-        self._stored_exception: Exception | None = None
+        self._stored_exception: BaseException | None = None
 
         # Guardrails state tracking
         self._interrupted_response_ids: set[str] = set()
@@ -123,6 +123,8 @@ class RealtimeSession(RealtimeModelListener):
         )
 
         self._guardrail_tasks: set[asyncio.Task[Any]] = set()
+        self._tool_call_tasks: set[asyncio.Task[Any]] = set()
+        self._async_tool_calls: bool = bool(self._run_config.get("async_tool_calls", True))
 
     @property
     def model(self) -> RealtimeModel:
@@ -216,7 +218,11 @@ class RealtimeSession(RealtimeModelListener):
         if event.type == "error":
             await self._put_event(RealtimeError(info=self._event_info, error=event.error))
         elif event.type == "function_call":
-            await self._handle_tool_call(event)
+            agent_snapshot = self._current_agent
+            if self._async_tool_calls:
+                self._enqueue_tool_call_task(event, agent_snapshot)
+            else:
+                await self._handle_tool_call(event, agent_snapshot=agent_snapshot)
         elif event.type == "audio":
             await self._put_event(
                 RealtimeAudio(
@@ -384,11 +390,17 @@ class RealtimeSession(RealtimeModelListener):
         """Put an event into the queue."""
         await self._event_queue.put(event)
 
-    async def _handle_tool_call(self, event: RealtimeModelToolCallEvent) -> None:
+    async def _handle_tool_call(
+        self,
+        event: RealtimeModelToolCallEvent,
+        *,
+        agent_snapshot: RealtimeAgent | None = None,
+    ) -> None:
         """Handle a tool call event."""
+        agent = agent_snapshot or self._current_agent
         tools, handoffs = await asyncio.gather(
-            self._current_agent.get_all_tools(self._context_wrapper),
-            self._get_handoffs(self._current_agent, self._context_wrapper),
+            agent.get_all_tools(self._context_wrapper),
+            self._get_handoffs(agent, self._context_wrapper),
         )
         function_map = {tool.name: tool for tool in tools if isinstance(tool, FunctionTool)}
         handoff_map = {handoff.tool_name: handoff for handoff in handoffs}
@@ -398,7 +410,7 @@ class RealtimeSession(RealtimeModelListener):
                 RealtimeToolStart(
                     info=self._event_info,
                     tool=function_map[event.name],
-                    agent=self._current_agent,
+                    agent=agent,
                 )
             )
 
@@ -423,7 +435,7 @@ class RealtimeSession(RealtimeModelListener):
                     info=self._event_info,
                     tool=func_tool,
                     output=result,
-                    agent=self._current_agent,
+                    agent=agent,
                 )
             )
         elif event.name in handoff_map:
@@ -444,7 +456,7 @@ class RealtimeSession(RealtimeModelListener):
                 )
 
             # Store previous agent for event
-            previous_agent = self._current_agent
+            previous_agent = agent
 
             # Update current agent
             self._current_agent = result
@@ -752,10 +764,49 @@ class RealtimeSession(RealtimeModelListener):
                 task.cancel()
         self._guardrail_tasks.clear()
 
+    def _enqueue_tool_call_task(
+        self, event: RealtimeModelToolCallEvent, agent_snapshot: RealtimeAgent
+    ) -> None:
+        """Run tool calls in the background to avoid blocking realtime transport."""
+        task = asyncio.create_task(self._handle_tool_call(event, agent_snapshot=agent_snapshot))
+        self._tool_call_tasks.add(task)
+        task.add_done_callback(self._on_tool_call_task_done)
+
+    def _on_tool_call_task_done(self, task: asyncio.Task[Any]) -> None:
+        self._tool_call_tasks.discard(task)
+
+        if task.cancelled():
+            return
+
+        exception = task.exception()
+        if exception is None:
+            return
+
+        logger.exception("Realtime tool call task failed", exc_info=exception)
+
+        if self._stored_exception is None:
+            self._stored_exception = exception
+
+        asyncio.create_task(
+            self._put_event(
+                RealtimeError(
+                    info=self._event_info,
+                    error={"message": f"Tool call task failed: {exception}"},
+                )
+            )
+        )
+
+    def _cleanup_tool_call_tasks(self) -> None:
+        for task in self._tool_call_tasks:
+            if not task.done():
+                task.cancel()
+        self._tool_call_tasks.clear()
+
     async def _cleanup(self) -> None:
         """Clean up all resources and mark session as closed."""
         # Cancel and cleanup guardrail tasks
         self._cleanup_guardrail_tasks()
+        self._cleanup_tool_call_tasks()
 
         # Remove ourselves as a listener
         self._model.remove_listener(self)

{openai_agents-0.4.1.dist-info → openai_agents-0.4.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: openai-agents
-Version: 0.4.1
+Version: 0.4.2
 Summary: OpenAI Agents SDK
 Project-URL: Homepage, https://openai.github.io/openai-agents-python/
 Project-URL: Repository, https://github.com/openai/openai-agents-python
@@ -44,7 +44,7 @@ Requires-Dist: numpy<3,>=2.2.0; (python_version >= '3.10') and extra == 'voice'
 Requires-Dist: websockets<16,>=15.0; extra == 'voice'
 Description-Content-Type: text/markdown
 
-# OpenAI Agents SDK
+# OpenAI Agents SDK [![PyPI](https://img.shields.io/pypi/v/openai-agents?label=pypi%20package)](https://pypi.org/project/openai-agents/)
 
 The OpenAI Agents SDK is a lightweight yet powerful framework for building multi-agent workflows. It is provider-agnostic, supporting the OpenAI Responses and Chat Completions APIs, as well as 100+ other LLMs.
 

{openai_agents-0.4.1.dist-info → openai_agents-0.4.2.dist-info}/RECORD

@@ -36,7 +36,7 @@ agents/extensions/memory/encrypt_session.py,sha256=PVnZIEj50bjUq16OLnMKrbZiinLkr
 agents/extensions/memory/redis_session.py,sha256=JwXY6zUTMgq9bRezlyFZ4Tze7DO7T0hioTc23qjSHjU,9838
 agents/extensions/memory/sqlalchemy_session.py,sha256=fnlZkNF_XZekP44uhiR4rjlCkwG7JJEiFm35TJfiCtc,12325
 agents/extensions/models/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-agents/extensions/models/litellm_model.py,sha256=JwJqURCfazbhqV--yoopri27U3JvIk8BH7cVzpGhbKw,23949
+agents/extensions/models/litellm_model.py,sha256=hbQFhAeEF5eHdyu5Q-7HNYFEhmn0KFK2KAcfYA10vqc,25621
 agents/extensions/models/litellm_provider.py,sha256=ZHgh1nMoEvA7NpawkzLh3JDuDFtwXUV94Rs7UrwWqAk,1083
 agents/mcp/__init__.py,sha256=yHmmYlrmEHzUas1inRLKL2iPqbb_-107G3gKe_tyg4I,750
 agents/mcp/server.py,sha256=cby0KKKKRhuWCydr4llERPL72Z94uV-SV3LLAcgcWTk,28435
@@ -50,7 +50,7 @@ agents/models/__init__.py,sha256=E0XVqWayVAsFqxucDLBW30siaqfNQsVrAnfidG_C3ok,287
 agents/models/_openai_shared.py,sha256=4Ngwo2Fv2RXY61Pqck1cYPkSln2tDnb8Ai-ao4QG-iE,836
 agents/models/chatcmpl_converter.py,sha256=qEobLnIJjrK6WRi_tsVkrDrGq78EGro3MZXlVMpMK2c,26011
 agents/models/chatcmpl_helpers.py,sha256=YC2krp_-uBgRCrCEImLjNvONTWRWfwLlPKHI4kBmNXE,1483
-agents/models/chatcmpl_stream_handler.py,sha256=r8nc-4hJg1plw87y24MD48O23xnfC_2gHKowtOYgO3M,28896
+agents/models/chatcmpl_stream_handler.py,sha256=1h0esxmnlBk9NwDjjwSlWYzjzuMgIpMLtRU9kaszfyg,29212
 agents/models/default_models.py,sha256=mlvBePn8H4UkHo7lN-wh7A3k2ciLgBUFKpROQxzdTfs,2098
 agents/models/fake_id.py,sha256=lbXjUUSMeAQ8eFx4V5QLUnBClHE6adJlYYav55RlG5w,268
 agents/models/interface.py,sha256=-AFUHC8iRuGZmtQwguDw4s-M4OPL2y2mct4TAmWvVrU,4057
@@ -64,7 +64,7 @@ agents/realtime/_default_tracker.py,sha256=4OMxBvD1MnZmMn6JZYKL42uWhVzvK6NdDLDfP
 agents/realtime/_util.py,sha256=ehBzUN1RTD2m2TXq73Jm4WohQzJ6y_MfnF5MaK8uu14,341
 agents/realtime/agent.py,sha256=bkegBJ_lc3z3NtnlIyEkVZFxZWBJwVjsQVzpQZAu7PM,4283
 agents/realtime/audio_formats.py,sha256=DBUWVVff4XY5BT6Mol86tF4PFMp5OIS3LmAbqUmQn_k,1019
-agents/realtime/config.py,sha256=ud0GK8ZbcnKRC4oGZNwpsiZI8TZ1OdTSMADfFtM8Z6I,6948
+agents/realtime/config.py,sha256=vnjgkeZXcOSLFopoAiGj4Vki_75pEJIKTagJtQpCWmg,7072
 agents/realtime/events.py,sha256=eANiNNyYlp_1Ybdl-MOwXRVTDtrK9hfgn6iw0xNxnaY,5889
 agents/realtime/handoffs.py,sha256=iJ4lr5RVdDkw5W3_AOGB_Az-hlRt1CoFFFNFDfd3ues,6698
 agents/realtime/items.py,sha256=5EG768FkKpbk-dhe4b_7BfFpdUEFWtxoiVUtNI9KXsc,5517
@@ -73,7 +73,7 @@ agents/realtime/model_events.py,sha256=2NKofzLszKHwtlcsogsNnH6hdeFfO7S96yWDB4Alx
 agents/realtime/model_inputs.py,sha256=-pl8Oj0WVrA5Gt-dqP5Va3ZHqXyIXpsjMsf9UL-suEY,2789
 agents/realtime/openai_realtime.py,sha256=jN3OvcEQt9X-59t6InllkOOEd8Tdw69K5vuKfXBeObg,44763
 agents/realtime/runner.py,sha256=KfU7utmc9QFH2htIKN2IN9H-5EnB0qN9ezmvlRTnOm4,2511
-agents/realtime/session.py,sha256=gvEgRJHtQfVqyZkIB1JKcYyqXXjkhwZJ8rwsEgZMTeE,35225
+agents/realtime/session.py,sha256=79WqKWwGOsutQRLs7fDsijE-OxEJjGm-aOpjL5F7Fn8,36983
 agents/tracing/__init__.py,sha256=5HO_6na5S6EwICgwl50OMtxiIIosUrqalhvldlYvSVc,2991
 agents/tracing/create.py,sha256=xpJ4ZRnGyUDPKoVVkA_8hmdhtwOKGhSkwRco2AQIhAo,18003
 agents/tracing/logger.py,sha256=J4KUDRSGa7x5UVfUwWe-gbKwoaq8AeETRqkPt3QvtGg,68
@@ -108,7 +108,7 @@ agents/voice/models/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSu
 agents/voice/models/openai_model_provider.py,sha256=Khn0uT-VhsEbe7_OhBMGFQzXNwL80gcWZyTHl3CaBII,3587
 agents/voice/models/openai_stt.py,sha256=Lb_F9160VNKDHXZ9zylSzeig7sB8lBjiYhQLDZsp6NQ,17257
 agents/voice/models/openai_tts.py,sha256=4KoLQuFDHKu5a1VTJlu9Nj3MHwMlrn9wfT_liJDJ2dw,1477
-openai_agents-0.4.1.dist-info/METADATA,sha256=zTH3g8bgaelHCJ_80grQkZvhJ5Zm3Lzz1BuAsG3xtzY,12929
-openai_agents-0.4.1.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-openai_agents-0.4.1.dist-info/licenses/LICENSE,sha256=E994EspT7Krhy0qGiES7WYNzBHrh1YDk3r--8d1baRU,1063
-openai_agents-0.4.1.dist-info/RECORD,,
+openai_agents-0.4.2.dist-info/METADATA,sha256=UUyVoFXNYwTLrBnkpo7MFwT73-kJH0rQX53xwF3pFXw,13046
+openai_agents-0.4.2.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+openai_agents-0.4.2.dist-info/licenses/LICENSE,sha256=E994EspT7Krhy0qGiES7WYNzBHrh1YDk3r--8d1baRU,1063
+openai_agents-0.4.2.dist-info/RECORD,,