ADK is Google's code-first toolkit for building agents: an agent with tools, a runner that drives the loop, sessions and state, callbacks around every step, and workflow graphs for the parts where the order is fixed. It runs anywhere Python does, and deploys to Cloud Run, GKE or Google's managed agent runtime.
Reach for it when
- You are building on Gemini or Google Cloud
- You want fixed workflows as well as model-driven agents
- You want one toolkit from prototype to deployment
Look elsewhere when
- You want a browser-only tutorial: ADK cannot install in a browser
- You need a tiny dependency footprint
What the lessons will cover
- 01Install and a first agent
- 02A stand-in model you write in twenty lines
- 03The runner, sessions and events
- 04Function tools and what they return
- 05ToolContext, state and safety
- 06Instructions, structured output and callbacks
- 07Sub agents and transfer
- 08Sequential, parallel and loop agents
- 09Workflow graphs, and the deprecation
- 10State prefixes and memory
- 11The dev UI, the CLI and a server
- 12A support desk end to end
Others in vendor sdk
OpenAI Agents SDKOpenAI's own agent loop, with handoffs and guardrails.
Vercel AI SDKThe TypeScript way to build streaming AI apps.
Claude Agent SDKClaude Code as a library you can run yourself.Common questions
Is Google ADK free to use?
Google ADK is open source and free to run yourself. You still pay whichever model provider you point it at, and this tutorial is free with no signup.
Do I need to know Python to use Google ADK?
Basic Python is enough. Functions, dictionaries and imports cover most of what Google ADK asks of you.
When should I not use Google ADK?
You want a browser-only tutorial: ADK cannot install in a browser. You need a tiny dependency footprint.
Lessons for Google ADK are being written. Meanwhile the Agent Governance Toolkit tutorial covers the same ground: state, tools, loops, memory and human approval. Most of it carries straight over.
Start the Agent Governance Toolkit tutorial →