Google ADK Tutorial: Build and Run AI Agents on Zo

If you’re searching for “Google ADK” (Agent Development Kit), you probably want a quick, concrete path to: install it, build a minimal agent, run it, and then deploy it somewhere that won’t disappear when your laptop sleeps.

Zo Computer is a good fit for this workflow because it’s a persistent Linux server with:

  • a real filesystem (your agent code + logs live in folders)

  • a terminal for installing dependencies

  • an always-on “Services” manager so your agent can keep running

This tutorial uses ADK for Python.

Prerequisites

  • A Zo Computer

  • A model/API credential appropriate for your ADK setup (depending on the model/provider you use)

  • Basic comfort running terminal commands

Step 1: Create a project folder

Create a dedicated directory so your agent code, virtual environment, and logs stay together:

  • Create a folder like: Agents/google-adk/

(Any folder is fine. The important thing is: keep the agent and its environment together.)

Step 2: Create a Python virtual environment

From your project folder:

  1. Create a virtual environment

  2. Activate it

This keeps ADK dependencies isolated from the rest of your system.

Step 3: Install Google ADK

Install ADK from PyPI:

  • Package: google-adk ^1

If you already have ADK installed elsewhere, still prefer using a venv per agent project so upgrades are controlled.

Step 4: Write a minimal agent

Create a file like agent.py with a minimal “hello world” agent.

What you want at minimum:

  • An agent definition (name + instruction)

  • A runnable entrypoint you can test from the command line

ADK supports richer patterns (tools, workflows, memory/session services), but start minimal and add complexity only after you have an end-to-end run loop.

Step 5: Run it interactively (smoke test)

Before you deploy anything, run the agent once locally to verify:

  • imports work

  • credentials are accessible

  • the agent responds

If you get errors here, fix them now—deploying a broken agent as a background service is just harder to debug.

Step 6: (Optional) Use ADK devtools / web UI while developing

ADK provides developer tooling (CLI and web UI) that can make iteration faster—especially if you’re testing multi-step behaviors. Start here for official guidance on TypeScript devtools concepts; the workflow is similar in spirit (local dev loop, then deploy) ^2.

Step 7: Keep your agent always-on using Zo Services

Once your agent runs locally, the practical next step is making it persistent.

On Zo, you typically want one of these patterns:

  1. HTTP service (recommended): your agent exposes an API endpoint you can call from other tools/agents.

  2. Background worker: your agent runs on a schedule (Zo Agents) or loops continuously.

For “always-on”, use Zo Services (so it auto-restarts). For “runs on a schedule”, use Zo Agents. Agents documentation lives here ^3.

Step 8: Connect your ADK agent to Zo workflows

Once the agent exists, the easiest way to make it useful is to give it real inputs and outputs in your Zo environment:

  • Read/write files in your workspace (durable state)

  • Use Zo’s web tools when you need live data:

    • Fast extraction: https://docs.zocomputer.com/tools/read-webpage

    • Full browser rendering + screenshot: https://docs.zocomputer.com/tools/view-webpage

Both links are in Zo’s official tool reference ^4 ^5.

Common mistakes

  • Skipping the smoke test: if you haven’t run it once interactively, you don’t know what “working” looks like.

  • Mixing environments: install ADK in a venv per project; don’t “pip install” into your global Python.

  • No logs: always write logs to a file in your agent folder so you can debug restarts.

Summary

You now have a working Google ADK agent project on a persistent server.

The next iteration loop is:

  1. Add one tool (file IO, a simple API call, or a single web action)

  2. Add one durable state file

  3. Deploy as a Zo Service (always-on) or a Zo Agent (scheduled)