Cache LLM model instances in registry

src/models/registry.py

@@ -9,6 +9,7 @@
 """
 
 from __future__ import annotations
+from functools import lru_cache
 
 import importlib
 import logging
@@ -42,6 +43,8 @@ def _build_from_module(module_name: str, func_name: str, **kwargs) -> BaseChatMo
 }
 
 
+@lru_cache(maxsize=16)
+# Cache model instances to avoid recreating them on every call
 def get_model(
     provider: Optional[Provider] = None,
     model_name: Optional[str] = None,

tests/unit/models/test_registry.py

@@ -12,6 +12,10 @@ def mock_modules():
             "src.models.gemini": MagicMock(),
             "src.models.claude": MagicMock(),
             "src.models.openai": MagicMock(),
+            "langchain_core": MagicMock(),
+            "langchain_core.language_models": MagicMock(),
+            "pydantic": MagicMock(),
+            "pydantic_settings": MagicMock(),
         },
     ):
         yield