diff --git a/src/api/__pycache__/routes.cpython-310.pyc b/src/api/__pycache__/routes.cpython-310.pyc
index 7fbe2fc7..465c6c9c 100644
Binary files a/src/api/__pycache__/routes.cpython-310.pyc and b/src/api/__pycache__/routes.cpython-310.pyc differ
diff --git a/src/services/__pycache__/agent_builder.cpython-310.pyc b/src/services/__pycache__/agent_builder.cpython-310.pyc
index 73c61dc1..6bb48ab7 100644
Binary files a/src/services/__pycache__/agent_builder.cpython-310.pyc and b/src/services/__pycache__/agent_builder.cpython-310.pyc differ
diff --git a/src/services/__pycache__/agent_runner.cpython-310.pyc b/src/services/__pycache__/agent_runner.cpython-310.pyc
index 8362fc0e..800abf61 100644
Binary files a/src/services/__pycache__/agent_runner.cpython-310.pyc and b/src/services/__pycache__/agent_runner.cpython-310.pyc differ
diff --git a/src/services/agent_builder.py b/src/services/agent_builder.py
index 416d3bec..9d34cd56 100644
--- a/src/services/agent_builder.py
+++ b/src/services/agent_builder.py
@@ -12,6 +12,8 @@ from sqlalchemy.orm import Session
 from contextlib import AsyncExitStack
 from google.adk.agents.callback_context import CallbackContext
 from google.adk.models import LlmResponse, LlmRequest
+from google.adk.tools import load_memory
+
 from typing import Optional
 import logging
 import os
@@ -161,11 +163,10 @@ def search_knowledge_base_function_sync(query: str, history=[]):
 
 
 class AgentBuilder:
-    def __init__(self, db: Session, memory_service: InMemoryMemoryService):
+    def __init__(self, db: Session):
         self.db = db
         self.custom_tool_builder = CustomToolBuilder()
         self.mcp_service = MCPService()
-        self.memory_service = memory_service
 
     async def _create_llm_agent(
         self, agent
@@ -184,11 +185,6 @@ class AgentBuilder:
 
         # Combina todas as ferramentas
         all_tools = custom_tools + mcp_tools
-
-        # Verifica se load_memory está habilitado
-        before_model_callback_func = None
-        if agent.config.get("load_memory") == True:
-            before_model_callback_func = before_model_callback
             
         now = datetime.now()
         current_datetime = now.strftime("%d/%m/%Y %H:%M")
@@ -204,6 +200,13 @@ class AgentBuilder:
             current_time=current_time,
         )
 
+        # Verifica se load_memory está habilitado
+        # before_model_callback_func = None
+        if agent.config.get("load_memory") == True:
+            all_tools.append(load_memory)
+            # before_model_callback_func = before_model_callback
+            formatted_prompt = formatted_prompt + "\n\n<memory_instructions>ALWAYS use the load_memory tool to retrieve knowledge for your context</memory_instructions>\n\n"
+
         return (
             LlmAgent(
                 name=agent.name,
@@ -211,7 +214,7 @@ class AgentBuilder:
                 instruction=formatted_prompt,
                 description=agent.description,
                 tools=all_tools,
-                before_model_callback=before_model_callback_func,
+                # before_model_callback=before_model_callback_func,
             ),
             mcp_exit_stack,
         )
diff --git a/src/services/agent_runner.py b/src/services/agent_runner.py
index 2a88e590..cc884ba2 100644
--- a/src/services/agent_runner.py
+++ b/src/services/agent_runner.py
@@ -42,7 +42,7 @@ async def run_agent(
             raise AgentNotFoundError(f"Agente com ID {agent_id} não encontrado")
 
         # Usando o AgentBuilder para criar o agente
-        agent_builder = AgentBuilder(db, memory_service)
+        agent_builder = AgentBuilder(db)
         root_agent, exit_stack = await agent_builder.build_agent(get_root_agent)
 
         logger.info("Configurando Runner")
@@ -51,6 +51,7 @@ async def run_agent(
             app_name=get_root_agent.name,
             session_service=session_service,
             artifact_service=artifacts_service,
+            memory_service=memory_service,
         )
         session_id = contact_id + "_" + agent_id
 
@@ -82,6 +83,15 @@ async def run_agent(
                 if event.is_final_response() and event.content and event.content.parts:
                     final_response_text = event.content.parts[0].text
                     logger.info(f"Resposta final recebida: {final_response_text}")
+        
+            completed_session = session_service.get_session(
+                app_name=root_agent.name,
+                user_id=contact_id,
+                session_id=session_id,
+            )
+            
+            memory_service.add_session_to_memory(completed_session)
+        
         finally:
             # Garante que o exit_stack seja fechado corretamente
             if exit_stack: