Develop an Agent Development Kit agent

This page shows you how to develop an agent using the Agent Development Kit template (the AdkApp class in the Vertex AI SDK for Python). The agent returns the exchange rate between two currencies on a specified date.

Use the following steps:

Define and configure a model
(Optional) Define and use a tool
(Optional) Manage sessions
(Optional) Manage memories

Before you begin

Make sure your environment is set up by following the steps in Set up your environment.

Define and configure a model

Define the model version:

model = "gemini-2.0-flash"

(Optional) Configure the safety settings of the model. To learn more about the options available for safety settings in Gemini, see Configure safety attributes. The following is an example of how you can configure the safety settings:

from google.genai import types

safety_settings = [
    types.SafetySetting(
        category=types.HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT,
        threshold=types.HarmBlockThreshold.OFF,
    ),
]

(Optional) Specify content generation parameters:

from google.genai import types

generate_content_config = types.GenerateContentConfig(
   safety_settings=safety_settings,
   temperature=0.28,
   max_output_tokens=1000,
   top_p=0.95,
)

Create an AdkApp using the model configurations:

from google.adk.agents import Agent
from vertexai.agent_engines import AdkApp

agent = Agent(
   model=model,                                      # Required.
   name='currency_exchange_agent',                   # Required.
   generate_content_config=generate_content_config,  # Optional.
)
app = AdkApp(agent=agent)

If you are running in an interactive environment, such as the terminal or a Colab notebook, you can run a query as an intermediate testing step:

async for event in app.async_stream_query(
   user_id="USER_ID",  # Required
   message="What is the exchange rate from US dollars to Swedish currency?",
):
   print(event)

USER_ID: Choose your own user ID with a character limit of 128. For example, user-123.

The response is a Python dictionary similar to the following example:

{'actions': {'artifact_delta': {},
             'requested_auth_configs': {},
             'state_delta': {}},
 'author': 'currency_exchange_agent',
 'content': {'parts': [{'text': 'To provide you with the most accurate '
                                'exchange rate, I need to know the specific '
                                'currencies you\'re asking about. "Swedish '
                                'currency" could refer to:\n'
                                '\n'
                                '*   **Swedish Krona (SEK):** This is the '
                                'official currency of Sweden.\n'
                                '\n'
                                "Please confirm if you're interested in the "
                                'exchange rate between USD and SEK. Once you '
                                'confirm, I can fetch the latest exchange rate '
                                'for you.\n'}],
             'role': 'model'},
 'id': 'LYg7wg8G',
 'invocation_id': 'e-113ca547-0f19-4d50-9dde-f76cbc001dce',
 'timestamp': 1744166956.925927}

(Optional) Define and use a tool

After you define your model, define the tools that your model uses for reasoning.

When you define your function, it's important to include comments that fully and clearly describe the function's parameters, what the function does, and what the function returns. This information is used by the model to determine which function to use. You must also test your function locally to confirm that it works.

Use the following code to define a function that returns an exchange rate:

def get_exchange_rate(
    currency_from: str = "USD",
    currency_to: str = "EUR",
    currency_date: str = "latest",
):
    """Retrieves the exchange rate between two currencies on a specified date.

    Uses the Frankfurter API (https://api.frankfurter.app/) to obtain
    exchange rate data.

    Args:
        currency_from: The base currency (3-letter currency code).
            Defaults to "USD" (US Dollar).
        currency_to: The target currency (3-letter currency code).
            Defaults to "EUR" (Euro).
        currency_date: The date for which to retrieve the exchange rate.
            Defaults to "latest" for the most recent exchange rate data.
            Can be specified in YYYY-MM-DD format for historical rates.

    Returns:
        dict: A dictionary containing the exchange rate information.
            Example: {"amount": 1.0, "base": "USD", "date": "2023-11-24",
                "rates": {"EUR": 0.95534}}
    """
    import requests
    response = requests.get(
        f"https://api.frankfurter.app/{currency_date}",
        params={"from": currency_from, "to": currency_to},
    )
    return response.json()

To test the function before you use it in your agent, run the following:

get_exchange_rate(currency_from="USD", currency_to="SEK")

The response should be similar to the following:

{'amount': 1.0, 'base': 'USD', 'date': '2025-04-03', 'rates': {'SEK': 9.6607}}

To use the tool inside the AdkApp, add it to the list of tools under the tools= argument:

from google.adk.agents import Agent

agent = Agent(
    model=model,                     # Required.
    name='currency_exchange_agent',  # Required.
    tools=[get_exchange_rate],       # Optional.
)

You can test the agent locally by performing test queries against it. Run the following command to test the agent locally using US dollars and Swedish Krona:

from vertexai.agent_engines import AdkApp

app = AdkApp(agent=agent)
async for event in app.async_stream_query(
    user_id="USER_ID",
    message="What is the exchange rate from US dollars to SEK on 2025-04-03?",
):
    print(event)

where USER_ID is the user ID you defined. For example, user-123.

The response is a sequence of dictionaries that's similar to the following:

{'author': 'currency_exchange_agent',
 'content': {'parts': [{'function_call': {'args': {'currency_date': '2025-04-03',
                                                   'currency_from': 'USD',
                                                   'currency_to': 'SEK'},
                                          'id': 'adk-e39f3ba2-fa8c-4169-a63a-8e4c62b89818',
                                          'name': 'get_exchange_rate'}}],
             'role': 'model'},
 'id': 'zFyIaaif',
 # ...
}
{'author': 'currency_exchange_agent',
 'content': {'parts': [{'function_response': {'id': 'adk-e39f3ba2-fa8c-4169-a63a-8e4c62b89818',
                                              'name': 'get_exchange_rate',
                                              'response': {'amount': 1.0,
                                                           'base': 'USD',
                                                           'date': '2025-04-03',
                                                           'rates': {'SEK': 9.6607}}}}],
             'role': 'user'},
 'id': 'u2YR4Uom',
 # ...
}
{'author': 'currency_exchange_agent',
 'content': {'parts': [{'text': 'The exchange rate from USD to SEK on '
                                '2025-04-03 is 9.6607.'}],
             'role': 'model'},
 'id': 'q3jWA3wl',
 # ...
}

(Optional) Manage sessions

AdkApp uses in-memory sessions when running locally and uses cloud-based managed sessions after you deploy the agent to Vertex AI Agent Engine. This section describes how to configure your ADK agent to work with managed sessions.

(Optional) Customize your sessions database

If you want to override the default managed session service with your own database, you can define a session_service_builder function as follows:

def session_service_builder():
  from google.adk.sessions import InMemorySessionService

  return InMemorySessionService()

Pass your database to AdkApp as session_service_builder=:

from vertexai.agent_engines import AdkApp

app = AdkApp(
   agent=agent,                                      # Required.
   session_service_builder=session_service_builder,  # Optional.
)

Use the agent with sessions

When you run the AdkApp locally, the following instructions use in-memory sessions:

Create a session for your agent:

session = await app.async_create_session(user_id="USER_ID")
print(session)

The session is created as the dictionary representation of an ADK session object.

List sessions associated with your agent:

await app.async_list_sessions(user_id="USER_ID")

Get a particular session:

session = await app.async_get_session(user_id="USER_ID", session_id="SESSION_ID")

where

USER_ID is the user ID you defined. For example, user-123.
SESSION_ID is the ID for the particular session you want to retrieve.

Query the AdkApp using sessions:

async for event in app.async_stream_query(
    user_id="USER_ID",
    session_id=SESSION_ID, # Optional. you can pass in the session_id when querying the agent
    message="What is the exchange rate from US dollars to Swedish currency on 2025-04-03?",
):
    print(event)

The agent might respond with a request for information like the following:

{'author': 'currency_exchange_agent',
 'content': {'parts': [{'text': 'I need to know the Swedish currency code to '
                                'provide you with the exchange rate.'}],
             'role': 'model'},
 'id': 'wIgZAtQ4',
 #...
}

You can send a response (for example, "SEK") on behalf of USER_ID within the session corresponding to session by specifying:

async for event in app.async_stream_query(
    user_id="USER_ID",
    session_id=session.id, # Optional. you can pass in the session_id when querying the agent
    message="SEK",
):
    print(event)

You should receive a continuation of the conversation like the following sequence of dictionaries:

{'author': 'currency_exchange_agent',
 'content': {'parts': [{'function_call': {'args': {'currency_date': '2025-04-03',
                                                   'currency_from': 'USD',
                                                   'currency_to': 'SEK'},
                                          'id': 'adk-2b9230a6-4b92-4a1b-9a65-b708ff6c68b6',
                                          'name': 'get_exchange_rate'}}],
             'role': 'model'},
 'id': 'bOPHtzji',
 # ...
}
{'author': 'currency_exchange_agent',
 'content': {'parts': [{'function_response': {'id': 'adk-2b9230a6-4b92-4a1b-9a65-b708ff6c68b6',
                                              'name': 'get_exchange_rate',
                                              'response': {'amount': 1.0,
                                                           'base': 'USD',
                                                           'date': '2025-04-03',
                                                           'rates': {'SEK': 9.6607}}}}],
             'role': 'user'},
 'id': '9AoDFmiL',
 # ...
}
{'author': 'currency_exchange_agent',
 'content': {'parts': [{'text': 'The exchange rate from USD to SEK on '
                                '2025-04-03 is 1 USD to 9.6607 SEK.'}],
             'role': 'model'},
 'id': 'hmle7trT',
 # ...
}

(Optional) Manage memories

By default, AdkApp uses an in-memory implementation of agentic memory when running locally and uses Vertex AI Agent Engine Memory Bank after you deploy the agent to Vertex AI Agent Engine.

When developing your ADK agent, you can include a PreloadMemoryTool that controls when the agent retrieves memories and how memories are included in the prompt. The following example agent always retrieves memories at the start of each turn and includes the memories in the system instruction:

from google import adk
from vertexai.agent_engines import AdkApp

agent = adk.Agent(
    model="gemini-2.0-flash",
    name='stateful_agent',
    instruction="""You are a Vehicle Voice Agent, designed to assist users with information and in-vehicle actions.

1.  **Direct Action:** If a user requests a specific vehicle function (e.g., "turn on the AC"), execute it immediately using the corresponding tool. You don't have the outcome of the actual tool execution, so provide a hypothetical tool execution outcome.
2.  **Information Retrieval:** Respond concisely to general information requests with your own knowledge (e.g., restaurant recommendation).
3.  **Clarity:** When necessary, try to seek clarification to better understand the user's needs and preference before taking an action.
4.  **Brevity:** Limit responses to under 30 words.
""",
    tools=[adk.tools.preload_memory_tool.PreloadMemoryTool()],
)
app = AdkApp(agent=agent)

(Optional) Customize your memory service

If you want to override the default memory service, you can define a memory_service_builder function that returns a BaseMemoryService as follows:

def memory_service_builder():
  from google.adk.memory import InMemoryMemoryService

  return InMemoryMemoryService()

Pass your database to AdkApp as memory_service_builder=:

from vertexai.agent_engines import AdkApp

app = AdkApp(
   agent=agent,                                    # Required.
   memory_service_builder=memory_service_builder,  # Optional.
)

Use the agent with memories

Test your ADK agent with memories:

Create a session and interact with the agent:

initial_session = await app.async_create_session(user_id="USER_ID")

async for event in app.async_stream_query(
    user_id="USER_ID",
    session_id=initial_session.id,
    message="Can you update the temperature to my preferred temperature?",
):
    print(event)

Since there are no available memories during the first session and the agent does not know any user preferences, the agent may reply with a response such as "What is your preferred temperature?" You can respond with the following command:

async for event in app.async_stream_query(
    user_id="USER_ID",
    session_id=initial_session.id,
    message="I like it at 71 degrees",
):
    print(event)

The agent might return with a response such as "Setting the temperature to 71 degrees Fahrenheit. Temperature successfully changed." The agent's response may vary depending on the model you used.

Generate memories from the session. To store information from the session for use in future sessions, use the async_add_session_to_memory method:
```
await app.async_add_session_to_memory(session=initial_session)
```
Test that the agent has retained memory of the session (using PreloadMemoryTool) by creating a new session and prompting the agent:
```
new_session = await app.async_create_session(user_id="USER_ID")
async for event in app.async_stream_query(
    user_id="USER_ID",
    session_id=initial_session.id,
    message="Fix the temperature!",
):
    print(event)
```
The agent might return a response such as "setting temperature to 71 degrees. Is that correct?" The agent's response may vary depending on the model and memory service provider you used.

Use the async_search_memory method to display the agent's memories:

response = await app.async_search_memory(
    user_id="USER_ID",
    query="Fix the temperature!",
)
print(response)