Skip to main content

Lab 13: Creating a "Memory" Agent with Tool Context Challenge

Goal​

In this lab, you will build an agent that can remember and recall a piece of information using a custom tool. You will learn how to use the ToolContext object to read from and write to the session's state, giving your tool "memory."

Step 1: Create the Memory Agent Project​

  1. Create the agent project:

    adk create memory_agent
    cd memory_agent

  2. Create the tools module:

    mkdir tools
    touch tools/__init__.py
    touch tools/memory.py

Step 2: Write the State-Aware Tool Functions​

Exercise: Open tools/memory.py. A skeleton with two functions is provided. Your task is to implement the logic for each function using the tool_context to manage the agent's memory.

# In tools/memory.py (Starter Code)

from google.adk.tools import ToolContext

def remember_name(name: str, tool_context: ToolContext) -> dict:
    """
    Remembers the user's name by saving it to the session state.
    Use this tool when the user tells you their name.
    Args:
        name: The user's name to remember.
    """
    # Note: Do not describe the 'tool_context' parameter in the docstring.
    # The LLM does not need to know about it.
    
    # TODO: Use the tool_context to save the user's `name` to the
    # session state under the key 'user_name'.
    
    return {"status": "success", "message": f"I will remember that your name is {name}."}

def recall_name(tool_context: ToolContext) -> dict:
    """
    Recalls the user's name from the session state.
    Use this tool when the user asks you what their name is.
    """
    # TODO: Use the tool_context to get the 'user_name' from the state.
    # If the name exists, return it in a success dictionary:
    # {"status": "success", "name": user_name}
    # If it doesn't exist, return a "not_found" status:
    # {"status": "not_found", "message": "I don't believe you've told me your name yet."}
    pass

Step 3: Configure the Agent in agent.py​

Exercise: Open agent.py and configure the agent to use your new tools.

# In agent.py
from google.adk.agents import LlmAgent
from google.adk.tools import FunctionTool

# TODO: Import the `remember_name` and `recall_name` functions from your tools.memory module.

# TODO: Create a FunctionTool for each of your imported functions.

root_agent = LlmAgent(
    name="memory_agent",
    model="gemini-2.5-flash",
    description="An agent that can remember and recall the user's name.",
    instruction="""
# TODO: Write an instruction that tells the agent:
# - To use the `remember_name` tool when the user provides their name.
# - To use the `recall_name` tool when the user asks what their name is.
""",
    # TODO: Add the two FunctionTool objects you created to this list.
    tools=[]
)

Step 4: Test the Memory Agent​

  1. Set up your .env file.
  2. Navigate to the parent directory (cd ..) and start the Dev UI: 
    adk web
  3. Interact with the agent:
    • Turn 1: "Hi, my name is Alex."
    • Turn 2: "What is my name?"
  4. Examine the Trace and State:
    • After the first turn, check the State tab to see if user_name was saved correctly.
    • Check the Trace view for both turns to see the remember_name and recall_name tools being called.

Having Trouble?​

If you get stuck, you can find the complete, working code in the lab-solution.md file.

Lab Summary​

You have now created a tool with "memory"! You have learned to:

  • Access the ToolContext by adding it to a tool function's signature.
  • Write data to the session state using tool_context.state['key'] = value.
  • Read data from the session state using tool_context.state.get('key').
  • Build an agent that can carry information across multiple turns of a conversation.

Self-Reflection Questions​

  • Why is it important that the tool_context parameter is not included in the function's docstring?
  • The recall_name function uses tool_context.state.get('user_name') instead of tool_context.state['user_name']. What is the difference, and why is .get() a safer choice here?
  • How could you extend this agent to forget a user's name? What would the new tool function look like?

🕵️ Hidden Solution 🕵️​

Looking for the solution? Here's a hint (Base64 decode me): L2RvYy1hZGstdHJhaW5pbmcvbW9kdWxlMTMtdG9vbC1jb250ZXh0L2xhYi1zb2x1dGlvbg==

The direct link is: Lab Solution