fix(usage-ai-chat): resolve proxy model aliases before acompletion

litellm/proxy/management_endpoints/usage_endpoints/ai_usage_chat.py

@@ -426,6 +426,29 @@ def _sse(event: SSEEvent) -> str:
     return f"data: {json.dumps(event)}\n\n"
 
 
+def _resolve_usage_chat_model_alias(model: str) -> str:
+    """Resolve model aliases configured on the proxy before calling acompletion."""
+    resolved_model = model.strip()
+    if not resolved_model:
+        return resolved_model
+
+    if litellm.model_alias_map and resolved_model in litellm.model_alias_map:
+        return litellm.model_alias_map[resolved_model]
+
+    try:
+        from litellm.proxy.proxy_server import llm_router
+
+        if (
+            llm_router is not None
+            and resolved_model in getattr(llm_router, "model_group_alias", {})
+        ):
+            return llm_router._get_model_from_alias(resolved_model)
+    except Exception:
+        pass
+
+    return resolved_model
+
+
 def _resolve_fetch_kwargs(
     fn_name: str,
     fn_args: Dict[str, str],
@@ -534,7 +557,8 @@ async def stream_usage_ai_chat(
     is_admin: bool = False,
 ) -> AsyncIterator[str]:
     """Stream SSE events: status → tool_call → chunk → done."""
-    resolved_model = (model or "").strip() or DEFAULT_COMPETITOR_DISCOVERY_MODEL
+    selected_model = (model or "").strip() or DEFAULT_COMPETITOR_DISCOVERY_MODEL
+    resolved_model = _resolve_usage_chat_model_alias(selected_model)
     truncated = (
         messages[-MAX_CHAT_MESSAGES:] if len(messages) > MAX_CHAT_MESSAGES else messages
     )

tests/test_litellm/proxy/management_endpoints/usage_endpoints/test_ai_usage_chat.py

@@ -177,6 +177,64 @@ class TestSummariseEntityData:
 
 
 class TestStreamUsageAiChat:
+    @pytest.mark.asyncio
+    async def test_stream_resolves_global_model_alias_before_acompletion(self):
+        mock_first_response = MagicMock()
+        mock_first_response.choices = [MagicMock()]
+        mock_first_response.choices[0].message.tool_calls = []
+        mock_first_response.choices[0].message.content = "ok"
+
+        with patch(
+            "litellm.proxy.management_endpoints.usage_endpoints.ai_usage_chat.litellm"
+        ) as mock_litellm:
+            mock_litellm.model_alias_map = {"usage-alias": "gpt-4o-mini"}
+            mock_litellm.acompletion = AsyncMock(return_value=mock_first_response)
+
+            events = []
+            async for event in stream_usage_ai_chat(
+                messages=[{"role": "user", "content": "hi"}],
+                model="usage-alias",
+                is_admin=True,
+            ):
+                events.append(event)
+
+            first_call_kwargs = mock_litellm.acompletion.call_args_list[0].kwargs
+            assert first_call_kwargs["model"] == "gpt-4o-mini"
+
+    @pytest.mark.asyncio
+    async def test_stream_resolves_router_model_group_alias_before_acompletion(self):
+        mock_first_response = MagicMock()
+        mock_first_response.choices = [MagicMock()]
+        mock_first_response.choices[0].message.tool_calls = []
+        mock_first_response.choices[0].message.content = "ok"
+
+        mock_router = MagicMock()
+        mock_router.model_group_alias = {"friendly-alias": "hidden-route"}
+        mock_router._get_model_from_alias.return_value = "anthropic/claude-3-5-sonnet"
+
+        with (
+            patch(
+                "litellm.proxy.management_endpoints.usage_endpoints.ai_usage_chat.litellm"
+            ) as mock_litellm,
+            patch.dict(
+                "sys.modules",
+                {"litellm.proxy.proxy_server": MagicMock(llm_router=mock_router)},
+            ),
+        ):
+            mock_litellm.model_alias_map = {}
+            mock_litellm.acompletion = AsyncMock(return_value=mock_first_response)
+
+            events = []
+            async for event in stream_usage_ai_chat(
+                messages=[{"role": "user", "content": "hi"}],
+                model="friendly-alias",
+                is_admin=True,
+            ):
+                events.append(event)
+
+            first_call_kwargs = mock_litellm.acompletion.call_args_list[0].kwargs
+            assert first_call_kwargs["model"] == "anthropic/claude-3-5-sonnet"
+
     @pytest.mark.asyncio
     async def test_stream_emits_status_events(self):
         mock_tool_call = MagicMock()