ADK Python practice

Practice workspace for Google’s Agent Development Kit (ADK) for Python, following the Python quickstart. The repo includes two sample agents you can run from the CLI or the ADK web UI: a quickstart agent with one tool, and a multi-tool agent that exercises several tools on the same Agent.

Prerequisites

• Python 3.10 or later (this repo was set up with Python 3.14)
• pip

Setup

1. Virtual environment (recommended)

   python3 -m venv .venv

   Activate it:

   • Windows (PowerShell): .venv\Scripts\Activate.ps1
   • Windows (CMD): .venv\Scripts\activate.bat
   • macOS / Linux: source .venv/bin/activate

2. Install ADK

   pip install google-adk

   The quickstart uses pip install google-adk; a requirements.txt can be added later if you want pinned versions.

Project layout

1-Quickstart/practice_agent/
  agent.py      # Defines root_agent and tools
  __init__.py   # Package entry (imports agent)
  .env          # GOOGLE_API_KEY (local only; not committed)

2-Multi-tool/multi_tool_practice_agent/
  agent.py      # root_agent with multiple tools
  __init__.py
  .env

Quickstart (1-Quickstart/practice_agent/) follows the layout from adk create + edits to agent.py. The root_agent is the entry point ADK loads. It uses Gemini (gemini-2.5-flash) and a single get_current_time tool that returns the current local time (a stand-in for city-specific time in the quickstart).

Multi-tool (2-Multi-tool/multi_tool_practice_agent/) uses the same pattern with three tools on one agent: get_powershell_commands, get_bash_commands, and get_cmd_commands. The instructions steer the model to pick the right shell and tool for the user’s task (terminal command learning / practice).

API key

The agents use the Gemini API via a Google AI Studio key.

1. Create a key: Google AI Studio — API keys

2. Put it in each agent folder you run (same variable in each):

   • 1-Quickstart/practice_agent/.env
   • 2-Multi-tool/multi_tool_practice_agent/.env

   GOOGLE_API_KEY="YOUR_API_KEY"

Keep .env out of version control (this repo’s .gitignore ignores *.env).

Run the agents

Run commands from the repository root.

CLI (adk run)

Quickstart (single tool):

adk run 1-Quickstart/practice_agent

Multi-tool:

adk run 2-Multi-tool/multi_tool_practice_agent

You get an interactive session; type exit to quit.

Web UI (adk web)

adk web --port 8000

Open http://localhost:8000, pick the agent in the UI, and chat. The picker lists agents discovered under the current working directory (running from the repo root exposes both packages).

ADK Web is intended for development and debugging, not production.

How this maps to the quickstart

Quickstart step	Status here
Install google-adk	Done (pip install google-adk)
adk create …	1-Quickstart/practice_agent created with Gemini + Google AI backend
root_agent + a tool in agent.py	get_current_time + instructions/tools wired on Agent
GOOGLE_API_KEY in .env	1-Quickstart/practice_agent/.env
adk run / adk web	Run from repo root as above

The multi-tool example builds on the same ideas (Gemini, root_agent, tools in tools=[...]); see 2-Multi-tool/multi_tool_practice_agent/agent.py for wiring multiple callables on one agent.

For model and auth options beyond the default Gemini + API key, see: AI models for ADK agents.

Next steps

Use the ADK build guides and the rest of the ADK documentation to add workflows, more tools, or deployment patterns. The multi-tool folder is a concrete starting point for agents that must choose among several tools per turn.