Revert "v3"

This reverts commit d4a3ee03b66ff5bfeec8d1f10a95adaecb840e26.

app.py

@@ -543,35 +543,39 @@ async def stream_agent_response(question: str, chat_history: List[List[str]]) ->
         logger.error(f"Error updating agent memory: {str(e)}", exc_info=True)
     
     # Execute the agent with proper error handling
-    assistant_message = {"role": "assistant", "content": ""}
-    messages.append(assistant_message)
-
-    wants_chart, _desired = detect_chart_preferences(question)
-    input_text = question
-    if wants_chart:
-        input_text = (
-            "Instrucciones: Si el usuario pide una visualización y los datos ya "
-            "están en el historial, no pidas más contexto. Responde brevemente "
-            "confirmando que generarás/has generado la gráfica y, si necesitas SQL, "
-            "devuelve solo la consulta en un bloque ```sql``` sin explicaciones.\n\n"
-            f"Pregunta del usuario: {question}"
-        )
+    try:
+        # Add empty assistant message that will be updated
+        assistant_message = {"role": "assistant", "content": ""}
+        messages.append(assistant_message)
+        
+        # If the user is asking for a chart, steer the agent to be concise
+        wants_chart, _desired = detect_chart_preferences(question)
+        input_text = question
+        if wants_chart:
+            input_text = (
+                "Instrucciones: Si el usuario pide una visualización y los datos ya "
+                "están en el historial, no pidas más contexto. Responde brevemente "
+                "confirmando que generarás/has generado la gráfica y, si necesitas SQL, "
+                "devuelve solo la consulta en un bloque ```sql``` sin explicaciones.\n\n"
+                f"Pregunta del usuario: {question}"
+            )
 
-    response = await agent.ainvoke({"input": input_text})
-    logger.info(f"Agent response type: {type(response)}")
-    logger.info(f"Agent response content: {str(response)[:500]}...")
-
-    # Handle different response formats
-    if hasattr(response, 'output') and response.output:
-        response_text = response.output
-    elif isinstance(response, str):
-        response_text = response
-    elif hasattr(response, 'get') and callable(response.get) and 'output' in response:
-        response_text = response['output']
-    else:
-        response_text = str(response)
-    
-    logger.info(f"Extracted response text: {response_text[:200]}...")
+        # Let the agent use its memory; don't pass raw chat_history
+        response = await agent.ainvoke({"input": input_text})
+        logger.info(f"Agent response type: {type(response)}")
+        logger.info(f"Agent response content: {str(response)[:500]}...")
+            
+            # Handle different response formats
+            if hasattr(response, 'output') and response.output:
+                response_text = response.output
+            elif isinstance(response, str):
+                response_text = response
+            elif hasattr(response, 'get') and callable(response.get) and 'output' in response:
+                response_text = response['output']
+            else:
+                response_text = str(response)
+            
+            logger.info(f"Extracted response text: {response_text[:200]}...")
             
             # Check if the response contains an SQL query and it truly looks like SQL
             sql_query = extract_sql_query(response_text)