Cambie la memoria por una version anterior

memory.py

@@ -1,26 +1,47 @@
-from langchain.vectorstores import Chroma
-from langchain.embeddings import HuggingFaceEmbeddings
-from langchain_core.documents import Document
+from collections import deque
+import pandas as pd 
+from config import Config
 
-
-class Memory:
-    def __init__(self):
-        self.embedding_model = HuggingFaceEmbeddings(
-            model_name="sentence-transformers/all-MiniLM-L6-v2"
-        )
-        self.vector_store = Chroma(
-            persist_directory="./chroma_db", embedding_function=self.embedding_model
-        )
+class ConversationMemory: 
+    def __init__(self, max_history: int = Config.MAX_HISTORY):
+        self.history = deque(maxlen=max_history)
         self.schema_cache = None
+
+    def add_interaction(self, question: str, sql: str, result: str): 
+        self.history.append({
+            "question": question, 
+            "sql": sql, 
+            "result_summary": self._summarize_result(result)
+        })
+
+    def _summarize_result(self, result) -> str: 
+        """Resumen ejecutivo para memoria de contexto"""
+        if isinstance(result, pd.DataFrame):
+            # Enfocado en datos CLAVE no en metadatos
+            if len(result) == 1:
+                return f"Único resultado: {result.iloc[0].to_dict()}"
+            elif 'Cliente' in result.columns:
+                top = result.nlargest(3, 'Neto') if 'Neto' in result.columns else result.head(3)
+                return f"Top clientes: {top['Cliente'].tolist()}"
+            else:
+                return f"Filas: {len(result)}, Columnas: {list(result.columns)}"
+        return str(result)
+
+    def get_context(self, current_question: str) -> str: 
+        if not self.history:
+            return ""
+        last_relevant = []
+        for interaction in self.history:
+            if "producto" in interaction['question'].lower() and "producto" in current_question.lower():
+                last_relevant.append(interaction)
+            elif "cliente" in interaction['question'].lower() and "cliente" in current_question.lower():
+                last_relevant.append(interaction)
         
-    def add_interaction(self, question: str, answer: str, sql: str):
-        document = Document(
-            page_content=f"Pregunta: {question}\nRespuesta: {answer}\nSQL: {sql}",
-            metadata={"source": "interaction"},
-        )
-        self.vector_store.add_documents([document])
-    
-    def get_relevant_interactions(self, question: str, top_k=3):
-        results = self.vector_store.similarity_search(question, k=top_k)
-        return "\n".join([d.page_content for d in results])
-    
+        context = ""
+        for i, interaction in enumerate(last_relevant[-1:], 1):  # Solo la última relevante
+            context += (
+                f"Interacción #{i}: {interaction['question'][:50]}...\n"
+                f"SQL: {interaction['sql'][:70]}...\n"
+                f"Resultado: {interaction['result_summary']}\n\n"
+            )
+        return context