Google ADK tutorial: support agents on Google Cloud

Krish Naik walks through a customer support application on the Gemini Enterprise Agent Platform in a tutorial sponsored by Google Cloud. Development and playground testing happen locally, but the demonstrated deployment runs on Google Cloud. This is not a guide to running model inference offline.

The setup uses a Google Cloud project with billing enabled and Google Agent CLI version 1.4.2. A coding assistant generates files and executes commands from plain English prompts. The scaffold uses Google ADK and includes Terraform deployment files. The speaker identifies the model as Gemini 3.7 Flash.

A coordinator routes requests to billing, shipping and refund specialists. Invoice lookup and order status use hardcoded sample data; the refund specialist follows instructions without a dedicated tool. These tests demonstrate routing, rather than a connection to live customer records.

The walkthrough adds a callback to persist completed turns to Memory Bank, then deploys to Agent Runtime with Agent Identity. The speaker explains that the coordinator and specialists share one deployed identity. Subsequent tests retrieve a contact preference and an invoice identifier from memory.

Evaluation and broader governance features are left for later videos. The tutorial ends by deleting deployed resources and the project to stop billing.