memory1

app.py

@@ -291,8 +291,10 @@ def create_agent(llm, db_connection, *, run_test: bool = True):
         #         test_result = agent.run(test_query)
         #         logger.info(f"Agent test query successful: {str(test_result)[:200]}...")
         #     except Exception as e:
-        #         logger.warning(f"Agent test query failed (this might be expected): {str(e)}")
-        #         # Continue even if test fails, as it might be due to model limitations
+        #         logger.warning(
+        #             f"Agent test query failed (this might be expected): {str(e)}"
+        #         )
+                # Continue even if test fails, as it might be due to model limitations
         
         logger.info("SQL agent created successfully")
         return agent, ""
@@ -488,6 +490,7 @@ def convert_to_messages_format(chat_history):
         return []
         
     messages = []
+    prev_messages = []
     
     # If the first element is a list, assume it's in the old format
     if isinstance(chat_history[0], list):
@@ -522,9 +525,13 @@ async def stream_agent_response(question: str, chat_history: List[List[str]], se
     # Add previous chat history in the correct format for the agent
     for msg_pair in chat_history:
         if len(msg_pair) >= 1 and msg_pair[0]:  # User message
-            messages.append(HumanMessage(content=msg_pair[0]))
+            hm = HumanMessage(content=msg_pair[0])
+            messages.append(hm)
+            prev_messages.append(hm)
         if len(msg_pair) >= 2 and msg_pair[1]:  # Assistant message
-            messages.append(AIMessage(content=msg_pair[1]))
+            am = AIMessage(content=msg_pair[1])
+            messages.append(am)
+            prev_messages.append(am)
     
     # Add current user's question
     user_message = HumanMessage(content=question)
@@ -564,7 +571,11 @@ async def stream_agent_response(question: str, chat_history: List[List[str]], se
         # Execute the agent with proper error handling
         try:
             # Let the agent use its memory; don't pass raw chat_history
-            response = await active_agent.ainvoke({"input": question})
+            # Pass explicit chat_history as well to maximize memory usage
+            response = await active_agent.ainvoke({
+                "input": question,
+                "chat_history": prev_messages,
+            })
             logger.info(f"Agent response type: {type(response)}")
             logger.info(f"Agent response content: {str(response)[:500]}...")
             
@@ -654,7 +665,10 @@ async def stream_agent_response(question: str, chat_history: List[List[str]], se
                             "Devuelve SOLO la consulta SQL en un bloque ```sql``` para responder a: "
                             f"{question}. No incluyas explicación ni texto adicional."
                         )
-                        sql_only_resp = await active_agent.ainvoke({"input": sql_only_prompt})
+                        sql_only_resp = await active_agent.ainvoke({
+                            "input": sql_only_prompt,
+                            "chat_history": prev_messages,
+                        })
                         sql_only_text = str(sql_only_resp)
                         sql_query2 = extract_sql_query(sql_only_text)
                         if sql_query2 and looks_like_sql(sql_query2):