diff --git a/litellm/realtime_api/main.py b/litellm/realtime_api/main.py index f94ba4c4a02..d67e4555a29 100644 --- a/litellm/realtime_api/main.py +++ b/litellm/realtime_api/main.py @@ -55,6 +55,7 @@ xai_realtime: Final = XAIRealtime() vertex_llm_base: Final = VertexBase() base_llm_http_handler = BaseLLMHTTPHandler() _EMPTY_MODEL_PARAMS: Final[Mapping[str, Any]] = MappingProxyType({}) +_EMPTY_AUTH_HEADERS: Final[Mapping[str, str]] = MappingProxyType({}) def _model_params_with_stored_credentials(model_params: Mapping[str, Any]) -> Mapping[str, Any]: @@ -602,13 +603,15 @@ def _azure_realtime_health_protocol( def _realtime_health_check_auth_headers( custom_llm_provider: str, api_key: str | None, model_params: Mapping[str, Any] -) -> Mapping[str, str | None]: - if custom_llm_provider != "azure": - return MappingProxyType({"api-key": api_key}) - return azure_realtime.get_auth_headers( - api_key=api_key, - azure_ad_token=(None if api_key else get_azure_ad_token(GenericLiteLLMParams(**model_params))), - ) +) -> Mapping[str, str]: + if custom_llm_provider == "azure": + return azure_realtime.get_auth_headers( + api_key=api_key, + azure_ad_token=(None if api_key else get_azure_ad_token(GenericLiteLLMParams(**model_params))), + ) + if api_key is None: + return _EMPTY_AUTH_HEADERS + return MappingProxyType({"Authorization": f"Bearer {api_key}"}) async def _realtime_health_check( diff --git a/tests/test_litellm/realtime_api/test_main.py b/tests/test_litellm/realtime_api/test_main.py index dcd4c01730d..643e65af47c 100644 --- a/tests/test_litellm/realtime_api/test_main.py +++ b/tests/test_litellm/realtime_api/test_main.py @@ -276,6 +276,40 @@ async def test_azure_health_check_resolves_stored_credentials(monkeypatch): assert "api-version=2025-04-01-preview" in connect.url +@pytest.mark.asyncio +@pytest.mark.parametrize( + ("custom_llm_provider", "model", "expected_url"), + [ + ("xai", "grok-voice-latest", "wss://api.x.ai/v1/realtime?model=grok-voice-latest"), + ("openai", "gpt-realtime", "wss://api.openai.com/v1/realtime?model=gpt-realtime"), + ], +) +async def test_bearer_health_check_sends_stored_credential_as_bearer_token( + monkeypatch, custom_llm_provider: str, model: str, expected_url: str +): + monkeypatch.setattr( + litellm, + "credential_list", + [ + CredentialItem( + credential_name="voice-key", + credential_values={"api_key": "sk-from-credential"}, + credential_info={}, + ) + ], + ) + connect = _CapturingConnect() + with patch("websockets.connect", connect): + assert await realtime_main._realtime_health_check( + model=model, + custom_llm_provider=custom_llm_provider, + api_key=None, + model_params={"model": f"{custom_llm_provider}/{model}", "litellm_credential_name": "voice-key"}, + ) + assert connect.kwargs["additional_headers"] == {"Authorization": "Bearer sk-from-credential"} + assert connect.url == expected_url + + @pytest.mark.asyncio async def test_azure_health_check_probes_ga_transcription_url_for_transcription_model(local_model_cost_map): """Regression for LIT-6240: transcription-only models (mode audio_transcription