Lab 10 Solution: Building a "Memory" Agent with Stateful Tools

Goal

This file contains the complete code for the agent.py and tools/memory.py files using ADK 2.0 standards for stateful tools.

memory_agent/tools/memory.py

from google.adk.tools import ToolContext

def store_name(name: str, tool_context: ToolContext) -> str:
    """
    Saves the user's name to the session memory.
    Use this tool when the user tells you their name.
    """
    # Write to session state
    tool_context.state["user_name"] = name
    return f"Got it, {name}! I've saved your name."

def recall_name(tool_context: ToolContext) -> str:
    """
    Retrieves the user's name from the session memory.
    Use this tool if the user asks who they are or what their name is.
    """
    # Read from session state
    name = tool_context.state.get("user_name", "Stranger")
    return f"Your name is {name}."

memory_agent/agent.py

from google.adk import Agent
from tools.memory import store_name, recall_name

root_agent = Agent(
    name="memory_agent",
    model="gemini-3.5-flash",
    description="An agent node that remembers users.",
    instruction="""
    You are a friendly assistant. 
    Use 'store_name' if the user introduces themselves.
    Use 'recall_name' if they ask for their name.
    """,
    tools=[store_name, recall_name]
)

Self-Reflection Answers

1. Why is it more reliable to store data in the session state rather than just relying on the LLM's chat history?

   • Answer: Chat history is "noisy" and has a limited context window. As a conversation gets longer, older messages (like an introduction) might be pushed out. Session state is structured and persistent; it acts as a reliable "database" for key facts that the agent can access deterministically.

2. What would happen if you used the same key (e.g., "user_name") for two different users?

   • Answer: Nothing bad! The ADK's Runner automatically isolates sessions based on the user_id. Each user has their own independent state dictionary, so Alice's "user_name" will never overwrite Bob's "user_name."

3. How could you extend this agent to remember other things?

   • Answer: You can create more tools that follow the same pattern (e.g., store_preference(key, value), recall_preference(key)). Because tool_context.state is just a Python dictionary, you can store any JSON-serializable data there.