Adding memory layer in adk framework

[harshit3134]

I want to have a memory layer with this framework like mem0 or langmem. Should I build a tool for it or should I use the BaseMemoryService from the memory module?

If anyone have other ideas, please share here.

[sachincse]

ADK has built-in memory support via BaseMemoryService — that is the recommended approach.

Option 1: Use BaseMemoryService (Recommended)

ADK provides InMemoryMemoryService out of the box and a BaseMemoryService you can extend:

from google.adk.memory import InMemoryMemoryService

memory_service = InMemoryMemoryService()

runner = Runner(
    agent=your_agent,
    app_name="my_app",
    session_service=session_service,
    memory_service=memory_service,
)

To build a custom backend (e.g., vector DB), extend BaseMemoryService:

from google.adk.memory import BaseMemoryService

class MyCustomMemoryService(BaseMemoryService):
    def add_session_to_memory(self, session):
        # Store session data in your preferred backend
        pass

    def search_memory(self, app_name, user_id, query):
        # Retrieve relevant memories
        pass

Integrating mem0

Wrap mem0 inside a custom BaseMemoryService:

from google.adk.memory import BaseMemoryService
from mem0 import Memory

class Mem0MemoryService(BaseMemoryService):
    def __init__(self):
        self.mem = Memory()

    def add_session_to_memory(self, session):
        for event in session.events:
            if hasattr(event, "content") and event.content:
                self.mem.add(event.content.parts[0].text, user_id=session.user_id)

    def search_memory(self, app_name, user_id, query):
        return self.mem.search(query, user_id=user_id)

Why not a Tool?

Building memory as a tool is less integrated — the agent must explicitly decide to call it. BaseMemoryService automatically injects relevant context into the agent loop.

TL;DR: Use BaseMemoryService as your integration point. It keeps memory within the ADK lifecycle rather than as a separate tool.