openai · rm-openai · Aug 21, 2025 · Aug 21, 2025
diff --git a/examples/realtime/app/server.py b/examples/realtime/app/server.py
@@ -111,6 +111,8 @@ async def _serialize_event(self, event: RealtimeSessionEvent) -> dict[str, Any]:
             }
         elif event.type == "error":
             base_event["error"] = str(event.error) if hasattr(event, "error") else "Unknown error"
+        elif event.type == "input_audio_timeout_triggered":
+            pass
         else:
             assert_never(event)
 

diff --git a/src/agents/realtime/config.py b/src/agents/realtime/config.py
@@ -78,6 +78,9 @@ class RealtimeTurnDetectionConfig(TypedDict):
     threshold: NotRequired[float]
     """The threshold for voice activity detection."""
 
+    idle_timeout_ms: NotRequired[int]
+    """Threshold for server-vad to trigger a response if the user is idle for this duration."""
+
 
 class RealtimeSessionModelSettings(TypedDict):
     """Model settings for a realtime model session."""

diff --git a/src/agents/realtime/events.py b/src/agents/realtime/events.py
@@ -216,6 +216,16 @@ class RealtimeGuardrailTripped:
     type: Literal["guardrail_tripped"] = "guardrail_tripped"
 
 
+@dataclass
+class RealtimeInputAudioTimeoutTriggered:
+    """Called when the model detects a period of inactivity/silence from the user."""
+
+    info: RealtimeEventInfo
+    """Common info for all events, such as the context."""
+
+    type: Literal["input_audio_timeout_triggered"] = "input_audio_timeout_triggered"
+
+
 RealtimeSessionEvent: TypeAlias = Union[
     RealtimeAgentStartEvent,
     RealtimeAgentEndEvent,
@@ -230,5 +240,6 @@ class RealtimeGuardrailTripped:
     RealtimeHistoryUpdated,
     RealtimeHistoryAdded,
     RealtimeGuardrailTripped,
+    RealtimeInputAudioTimeoutTriggered,
 ]
 """An event emitted by the realtime session."""
diff --git a/src/agents/realtime/model_events.py b/src/agents/realtime/model_events.py
@@ -84,6 +84,15 @@ class RealtimeModelInputAudioTranscriptionCompletedEvent:
 
     type: Literal["input_audio_transcription_completed"] = "input_audio_transcription_completed"
 
+@dataclass
+class RealtimeModelInputAudioTimeoutTriggeredEvent:
+    """Input audio timeout triggered."""
+
+    item_id: str
+    audio_start_ms: int
+    audio_end_ms: int
+
+    type: Literal["input_audio_timeout_triggered"] = "input_audio_timeout_triggered"
 
 @dataclass
 class RealtimeModelTranscriptDeltaEvent:
@@ -174,6 +183,7 @@ class RealtimeModelRawServerEvent:
     RealtimeModelAudioEvent,
     RealtimeModelAudioInterruptedEvent,
     RealtimeModelAudioDoneEvent,
+    RealtimeModelInputAudioTimeoutTriggeredEvent,
     RealtimeModelInputAudioTranscriptionCompletedEvent,
     RealtimeModelTranscriptDeltaEvent,
     RealtimeModelItemUpdatedEvent,

diff --git a/src/agents/realtime/openai_realtime.py b/src/agents/realtime/openai_realtime.py
@@ -6,7 +6,7 @@
 import json
 import os
 from datetime import datetime
-from typing import Any, Callable, Literal
+from typing import Annotated, Any, Callable, Literal, Union
 
 import pydantic
 import websockets
@@ -52,7 +52,7 @@
     SessionTracingTracingConfiguration as OpenAISessionTracingConfiguration,
     SessionUpdateEvent as OpenAISessionUpdateEvent,
 )
-from pydantic import TypeAdapter
+from pydantic import BaseModel, Field, TypeAdapter
 from typing_extensions import assert_never
 from websockets.asyncio.client import ClientConnection
 
@@ -83,6 +83,7 @@
     RealtimeModelErrorEvent,
     RealtimeModelEvent,
     RealtimeModelExceptionEvent,
+    RealtimeModelInputAudioTimeoutTriggeredEvent,
     RealtimeModelInputAudioTranscriptionCompletedEvent,
     RealtimeModelItemDeletedEvent,
     RealtimeModelItemUpdatedEvent,
@@ -128,6 +129,22 @@ async def get_api_key(key: str | Callable[[], MaybeAwaitable[str]] | None) -> st
     return os.getenv("OPENAI_API_KEY")
 
 
+class _InputAudioBufferTimeoutTriggeredEvent(BaseModel):
+    type: Literal["input_audio_buffer.timeout_triggered"]
+    event_id: str
+    audio_start_ms: int
+    audio_end_ms: int
+    item_id: str
+
+AllRealtimeServerEvents = Annotated[
+    Union[
+        OpenAIRealtimeServerEvent,
+        _InputAudioBufferTimeoutTriggeredEvent,
+    ],
+    Field(discriminator="type"),
+]
+
+
 class OpenAIRealtimeWebSocketModel(RealtimeModel):
     """A model that uses OpenAI's WebSocket API."""
 
@@ -462,8 +479,8 @@ async def _handle_ws_event(self, event: dict[str, Any]):
         try:
             if "previous_item_id" in event and event["previous_item_id"] is None:
                 event["previous_item_id"] = ""  # TODO (rm) remove
-            parsed: OpenAIRealtimeServerEvent = TypeAdapter(
-                OpenAIRealtimeServerEvent
+            parsed: AllRealtimeServerEvents = TypeAdapter(
+                AllRealtimeServerEvents
             ).validate_python(event)
         except pydantic.ValidationError as e:
             logger.error(f"Failed to validate server event: {event}", exc_info=True)
@@ -554,6 +571,12 @@ async def _handle_ws_event(self, event: dict[str, Any]):
             or parsed.type == "response.output_item.done"
         ):
             await self._handle_output_item(parsed.item)
+        elif parsed.type == "input_audio_buffer.timeout_triggered":
+            await self._emit_event(RealtimeModelInputAudioTimeoutTriggeredEvent(
+                item_id=parsed.item_id,
+                audio_start_ms=parsed.audio_start_ms,
+                audio_end_ms=parsed.audio_end_ms,
+            ))
 
     def _update_created_session(self, session: OpenAISessionObject) -> None:
         self._created_session = session

diff --git a/src/agents/realtime/session.py b/src/agents/realtime/session.py
@@ -28,6 +28,7 @@
     RealtimeHandoffEvent,
     RealtimeHistoryAdded,
     RealtimeHistoryUpdated,
+    RealtimeInputAudioTimeoutTriggered,
     RealtimeRawModelEvent,
     RealtimeSessionEvent,
     RealtimeToolEnd,
@@ -227,6 +228,12 @@ async def on_event(self, event: RealtimeModelEvent) -> None:
             await self._put_event(
                 RealtimeHistoryUpdated(info=self._event_info, history=self._history)
             )
+        elif event.type == "input_audio_timeout_triggered":
+            await self._put_event(
+                RealtimeInputAudioTimeoutTriggered(
+                    info=self._event_info,
+                )
+            )
         elif event.type == "transcript_delta":
             # Accumulate transcript text for guardrail debouncing per item_id
             item_id = event.item_id