diff --git a/litellm/proxy/management_endpoints/usage_endpoints/ai_usage_chat.py b/litellm/proxy/management_endpoints/usage_endpoints/ai_usage_chat.py index e2bcfd9380a..04ea99e1baf 100644 --- a/litellm/proxy/management_endpoints/usage_endpoints/ai_usage_chat.py +++ b/litellm/proxy/management_endpoints/usage_endpoints/ai_usage_chat.py @@ -474,14 +474,19 @@ def _resolve_model(model: str | None) -> str: _RouterRoute: TypeAlias = Literal["model_group", "deployment"] -def _router_route(router: "Router", model: str, caller_is_proxy_admin: bool) -> _RouterRoute | None: +def _router_route( + router: "Router", model: str, caller_is_proxy_admin: bool, request_team_id: str | None +) -> _RouterRoute | None: if router.is_recognized_model(model) or (caller_is_proxy_admin and model in router.team_public_model_names): return "model_group" if router.router_general_settings.pass_through_all_models: return None if router.default_deployment is not None or router.pattern_router.route(model) is not None: # pyright: ignore[reportUnknownMemberType] # PatternMatchDeployments.route returns untyped dicts return "model_group" - serves_model: Final = any(deployment["litellm_params"]["model"] == model for deployment in router.model_list) # pyright: ignore[reportUnknownVariableType] # Router.model_list is an untyped list + deployments: Final = router._get_deployment_by_litellm_model(model=model) # pyright: ignore[reportPrivateUsage] # the lookup the router's specific_deployment path runs + serves_model: Final = any( + router.deployment_usable_by_team(deployment, request_team_id) for deployment in deployments + ) return "deployment" if serves_model else None @@ -605,7 +610,7 @@ async def _acompletion( router_settings: Mapping[str, object] | None = None, ) -> ModelResponse | CustomStreamWrapper: from litellm.proxy.proxy_server import llm_router - from litellm.router_utils.common_utils import is_proxy_admin_request + from litellm.router_utils.common_utils import get_request_team_id, is_proxy_admin_request from litellm.types.llms.openai import AllMessageValues completion_messages: Final = cast( # cast-ok: acompletion accepts the same Mapping messages it already receives @@ -614,8 +619,9 @@ async def _acompletion( caller_metadata: Final[Mapping[str, object]] = ( _caller_metadata(user_api_key_dict) if user_api_key_dict is not None else MappingProxyType({}) ) + router_request: Final = MappingProxyType({"metadata": caller_metadata}) route: Final = ( - _router_route(llm_router, model, is_proxy_admin_request({"metadata": caller_metadata})) + _router_route(llm_router, model, is_proxy_admin_request(router_request), get_request_team_id(router_request)) if llm_router is not None else None ) diff --git a/tests/unit/proxy/management_endpoints/usage_endpoints/test_ai_usage_chat.py b/tests/unit/proxy/management_endpoints/usage_endpoints/test_ai_usage_chat.py index 0af9f403415..f6a53357206 100644 --- a/tests/unit/proxy/management_endpoints/usage_endpoints/test_ai_usage_chat.py +++ b/tests/unit/proxy/management_endpoints/usage_endpoints/test_ai_usage_chat.py @@ -1005,6 +1005,27 @@ class TestUsageAiChatProviderRequest: assert respx_mock.calls.last.request.headers["Authorization"] == "Bearer sk-env" + @pytest.mark.asyncio + async def test_team_owned_deployment_name_is_called_with_the_environment_key( + self, respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch + ) -> None: + from litellm.router import Router + + self._mock_openai(respx_mock) + router: Final = Router( + model_list=[ + { + "model_name": "team-gpt-5", + "litellm_params": {"model": "openai/gpt-5.6", "api_key": "sk-team-deployment"}, + "model_info": {"team_id": "team-1"}, + } + ] + ) + + await self._ask("openai/gpt-5.6", monkeypatch, router) + + assert respx_mock.calls.last.request.headers["Authorization"] == "Bearer sk-env" + class TestUsageAiChatKeepalive: async def _collect_endpoint_body(self, monkeypatch, interval, delay=0.3) -> tuple[list[bytes], dict]: