Fix Azure realtime beta session shape

cursoragent · cursoragent · commit 2cf367f7cdc7 · 2026-05-05T02:10:34.000Z
diff --git a/litellm/litellm_core_utils/realtime_streaming.py b/litellm/litellm_core_utils/realtime_streaming.py
@@ -46,6 +46,7 @@ def __init__(
         model: str = "",
         user_api_key_dict: Optional[Any] = None,
         request_data: Optional[Dict] = None,
+        backend_uses_beta_protocol: Optional[bool] = None,
     ):
         self.websocket = websocket
         self.backend_ws = backend_ws
@@ -58,6 +59,11 @@ def __init__(
 
         # Detect whether the client is explicitly opting into the beta protocol.
         self._client_wants_beta = self._detect_beta_header(websocket)
+        self._backend_uses_beta_protocol = (
+            self._client_wants_beta
+            if backend_uses_beta_protocol is None
+            else backend_uses_beta_protocol
+        )
 
         _logged_real_time_event_types = litellm.logged_real_time_event_types
 
@@ -261,7 +267,7 @@ async def _send_to_backend(self, message: str) -> None:
 
     def _make_disable_auto_response_message(self) -> str:
         """Return a session.update that disables VAD auto-response."""
-        if self._client_wants_beta:
+        if self._backend_uses_beta_protocol:
             session: Dict[str, Any] = {
                 "turn_detection": {"create_response": False},
             }
@@ -821,7 +827,10 @@ async def client_ack_messages(self):
                     # GA compatibility: remap beta-style session fields only when
                     # the upstream is in GA mode. Beta upstreams expect the flat
                     # session shape unchanged.
-                    if msg_type == "session.update" and not self._client_wants_beta:
+                    if (
+                        msg_type == "session.update"
+                        and not self._backend_uses_beta_protocol
+                    ):
                         session = msg_obj.get("session", {})
                         if isinstance(session, dict):
                             session = self._remap_beta_session_to_ga(session)
diff --git a/litellm/llms/azure/realtime/handler.py b/litellm/llms/azure/realtime/handler.py
@@ -89,9 +89,10 @@ async def async_realtime(
 
         if api_base is None:
             raise ValueError("api_base is required for Azure OpenAI calls")
-        if api_version is None and (
+        backend_uses_beta_protocol = (
             realtime_protocol is None or realtime_protocol.upper() not in ("GA", "V1")
-        ):
+        )
+        if api_version is None and backend_uses_beta_protocol:
             raise ValueError("api_version is required for Azure OpenAI calls")
 
         url = self._construct_url(
@@ -114,6 +115,7 @@ async def async_realtime(
                     logging_obj,
                     user_api_key_dict=user_api_key_dict,
                     request_data={"litellm_metadata": litellm_metadata or {}},
+                    backend_uses_beta_protocol=backend_uses_beta_protocol,
                 )
                 await realtime_streaming.bidirectional_forward()
 
diff --git a/tests/test_litellm/litellm_core_utils/test_realtime_streaming.py b/tests/test_litellm/litellm_core_utils/test_realtime_streaming.py
@@ -189,6 +189,45 @@ async def test_client_ack_messages_keeps_beta_session_shape_for_beta_clients():
     assert "audio" not in session
 
 
+@pytest.mark.asyncio
+async def test_client_ack_messages_keeps_beta_session_shape_for_beta_backend():
+    client_ws = MagicMock()
+    session_update = json.dumps(
+        {
+            "type": "session.update",
+            "session": {
+                "modalities": ["audio", "text"],
+                "voice": "alloy",
+                "turn_detection": {"create_response": False},
+            },
+        }
+    )
+    client_ws.receive_text = AsyncMock(
+        side_effect=[
+            session_update,
+            Exception("connection closed"),
+        ]
+    )
+    backend_ws = MagicMock()
+    backend_ws.send = AsyncMock()
+    logging_obj = MagicMock()
+    logging_obj.pre_call = MagicMock()
+    streaming = RealTimeStreaming(
+        client_ws, backend_ws, logging_obj, backend_uses_beta_protocol=True
+    )
+
+    await streaming.client_ack_messages()
+
+    sent_to_backend = json.loads(backend_ws.send.call_args_list[0].args[0])
+    session = sent_to_backend["session"]
+    assert session["modalities"] == ["audio", "text"]
+    assert session["voice"] == "alloy"
+    assert session["turn_detection"] == {"create_response": False}
+    assert "type" not in session
+    assert "output_modalities" not in session
+    assert "audio" not in session
+
+
 def test_translate_event_to_beta_renames_delta_types():
     ev = RealTimeStreaming._translate_event_to_beta(
         {"type": "response.output_audio.delta", "delta": "abc", "event_id": "e1"}
diff --git a/tests/test_litellm/llms/azure/realtime/test_azure_realtime_handler.py b/tests/test_litellm/llms/azure/realtime/test_azure_realtime_handler.py
@@ -247,6 +247,10 @@ async def __aexit__(self, exc_type, exc, tb):
         assert "model=gpt-4o-realtime-preview" in called_url
         assert "api-version" not in called_url
         assert "deployment" not in called_url
+        assert (
+            mock_realtime_streaming.call_args.kwargs["backend_uses_beta_protocol"]
+            is False
+        )
 
 
 @pytest.mark.asyncio
@@ -419,3 +423,7 @@ async def __aexit__(self, exc_type, exc, tb):
         called_url = mock_ws_connect.call_args[0][0]
         assert "/openai/realtime?" in called_url
         assert "/openai/v1/realtime" not in called_url
+        assert (
+            mock_realtime_streaming.call_args.kwargs["backend_uses_beta_protocol"]
+            is True
+        )
