feat: Implement memory sharing for EvaluatorOptimizerLLM

src/mcp_agent/workflows/evaluator_optimizer/evaluator_optimizer.py

@@ -75,6 +75,7 @@ def __init__(
         max_refinements: int = 3,
         llm_factory: Callable[[Agent], AugmentedLLM] | None = None,
         context: Optional["Context"] = None,
+        share_memory: bool = False,
     ):
         """
         Initialize the evaluator-optimizer workflow.
@@ -89,6 +90,7 @@ def __init__(
             min_rating: Minimum acceptable quality rating
             max_refinements: Maximum refinement iterations
             llm_factory: Optional factory to create LLMs from agents
+            share_memory: Whether to share the memory between the optimizer and evaluator
         """
         super().__init__(
             name=name,
@@ -147,7 +149,8 @@ def __init__(
             )
         else:
             raise ValueError(f"Unsupported evaluator type: {type(evaluator)}")
-
+        if share_memory:
+            self.evaluator_llm.share_memory_from(self.optimizer_llm)
         self.min_rating = min_rating
         self.max_refinements = max_refinements
 

src/mcp_agent/workflows/llm/augmented_llm.py

@@ -274,6 +274,26 @@ def __init__(
         self.model_selector = self.context.model_selector
         self.type_converter = type_converter
 
+    def share_memory_from(self, other: "AugmentedLLM"):
+        """
+        Share memory from another AugmentedLLM instance.
+
+        This creates a reference to the other instance's history, meaning both
+        instances will share the same memory object and any changes to history
+        in either instance will be reflected in both.
+
+        Args:
+        other: The AugmentedLLM instance to share memory from
+
+        Raises:
+        ValueError: If other is None or not an AugmentedLLM instance
+        """
+        if other is None:
+            raise ValueError("Cannot share memory from None")
+        if not isinstance(other, AugmentedLLM):
+            raise ValueError("Can only share memory from another AugmentedLLM instance")
+        self.history = other.history
+
     @abstractmethod
     async def generate(
         self,