This agent sample showcases the art of the possible using the ADK and Application Integration, Google Cloud’s low-to-no code integration platform for building Agentic workflows. Application Integration will be used in this sample to orchestrate a deterministic set of operations, including the following:

• Order processing/storage via BigQuery
• Orchestrating email confirmations once the order data has been written to BigQuery
• Invoking a “Human in the Loop” approval pattern if the order quantity exceeds 100

The agent will utilize the following tools:

• order_processing_tool: Deterministically processes order details depending on the size of the order, etc

This tool was created through an Application Integration process and is referenced in the agent code using the ADK’s built in ApplicationIntegrationToolset. This pattern/workflow lets an agent developer turn a deterministic set of operations into a tool through a few lines of code!

• Python 3.12+
• Poetry for dependency management and packaging
  • See the official Poetry website for more information. To install Poetry run:

  pip install poetry

• Google Cloud Project with the following roles assigned
  • Application Integration Admin
  • Connector Admin
  • Secret Manager Admin
  • Storage Admin
  • Service Usage Consumer
  • Logs Viewer

Once you have created your project, install the Google Cloud SDK. Then run the following command to authenticate:

gcloud auth login

You need to enable certain Google APIs. Run the following command to enable:

export PROJECT_ID=<project_id>
gcloud services enable integrations.googleapis.com connectors.googleapis.com secretmanager.googleapis.com aiplatform.googleapis.com compute.googleapis.com --project "$PROJECT_ID"

For this sample you must also provision Application Integration and Integration Connectors in your project.

The assets and additional prerequisite instructions are available in the Application Integration Samples repo here: Order Processing.

If you already have Application Integration and Integration Connectors provisioned in your project, you can simply deploy the assets by following the quickstart below.

Follow the instructions in this GCP Cloud Shell tutorial.

1. Clone the repository:

   git clone https://github.com/google/adk-samples.git
   cd python/agents/order-processing

   For the rest of this tutorial ensure you remain in the python/agents/order-processing directory.

2. Install the dependencies:

   Note for Linux users: If you get an error related to keyring during the installation, you can disable it by running the following command:

   poetry config keyring.enabled false

   This is a one-time setup.

   poetry install

3. Configure settings:

   • Set the following environment variables. You can set them in your .env file (modify and rename .env.example file to .env) or set them directly in your shell.

You can run the agent locally using the adk command in your terminal:

To run the agent from the ADK web UI:

adk web

Then select the order-processing from the dropdown.

Create a service account and assign it Vertex AI User and Application Integration Invoker roles

PROJECT_ID=<project_id>

gcloud iam service-accounts create "adk-order-processing" --project "$PROJECT_ID"

gcloud projects add-iam-policy-binding "$PROJECT_ID" \
    --member="serviceAccount:adk-order-processing@${PROJECT_ID}.iam.gserviceaccount.com" \
    --role="roles/aiplatform.user"

gcloud projects add-iam-policy-binding "$PROJECT_ID" \
    --member="serviceAccount:adk-order-processing@${PROJECT_ID}.iam.gserviceaccount.com" \
    --role="roles/integrations.integrationEditor"

Add the service account email in the .env file

The agent can also be deployed to Agent Runtime using the following commands:

poetry install --with deployment
python3 deployment/deploy.py

When the deployment finishes, it will output the resource ID of the remote agent deployment, for example:

Created remote agent: projects/<PROJECT_NUMBER>/locations/<PROJECT_LOCATION>/reasoningEngines/<AGENT_ENGINE_ID>

For more information on deploying to Agent Runtime, see here.

The deployment script adds the AGENT_ENGINE_ID to your .env file. To test the remote agent, simply run:

python3 deployment/test_deployment.py

You may then interact with the deployed agent from the shell. You can type quit at any point to exit.

Here's a quick example of how a user might interact with the agent:

User:

Hello

Agent:

Hello! How can I assist you today? I can help with creating orders.

User:

I would like to order 5 Google Pixel phones please

Agent:

No problem! To finalize your order for 5 Google Pixel phones, I need a little more information. Could you please provide:

Your Name:

Shipping Address:

User:

Sure, my name is Test User and my Shipping Address is 15 Testing Drive

Agent:

Great! Your order for 5 Google Pixel phones has been placed successfully. Your order will be shipped to 15 Testing Drive.

Note: At this point an email will be sent to the previously provided email address confirming your order details (order id, quantity/type of order, etc)

User:

Can I place another order?

Agent:

Yes, absolutely! What would you like to order this time? Please let me know the product, quantity, your name, and the shipping address.

User:

Please use the same name and shipping address as last time, I would like to order 150 Chromebooks please

Agent:

Your order for 150 Chromebooks has been placed successfully using the same name (Test User) and shipping address (15 Testing Drive). The order status is currently "In Progress".

Note: At this point the Human in the Loop process is invoked, where the email address previously provided has to approve a process run before the order is fulfilled (as the order quantity is >100)

Once approved, you will receive an email address confirming the order details