Start here

Tools

Shaping what it does

More than one agent

What it remembers

Running it for real

Making it real

Build a support agent with Google's Agent Development Kit: tools it chooses between, sub agents it hands work to, workflow agents that run steps in a fixed order, state that survives a conversation, and callbacks that watch every step. Every snippet was run against the real library, and a twenty line stand-in model means none of it needs a key.

Google ADK

Why there is no editor here

Google ADK course: a complete, practical tutorial

What you are going to build

No key yet

Install Google ADK and create your first agent

Install it and make an agent

Write it for a reader

Google ADK: start with a plain Python function

A function, before any of this

The docstring is the interface

How Google ADK turns a function into a tool schema

What ADK reads off your function

Why bother

Write a stand-in model for Google ADK with no API key

A model you can run for free

What the runner does with your question

Sessions come first

Run a Google ADK agent with Runner and a session

Running an agent

Debugging starts here

Understanding ADK events: the unit of everything

Reading the event stream

Function tools in Google ADK

A function is a tool

Return what is needed

What a Google ADK tool should return

What a tool should return

Edit descriptions, not prompts

How a Google ADK agent chooses between tools

Many tools, and how it chooses

Not everything belongs in a parameter

ToolContext in Google ADK: state and actions from a tool

Tools that see the session

Choosing between a tool and an integration

A long tool list is not free

Built-in and third-party tools in Google ADK

Tools you did not write

Keep it the same for everyone

Writing instructions for a Google ADK agent

Instructions that work

The key without the schema

Structured output in Google ADK with output_schema

Answers with a fixed shape

Where a hard rule belongs

Google ADK callbacks: run your code around every step

Callbacks: code around the loop

Especially when nobody is watching

Limit a runaway agent loop in Google ADK with max_llm_calls

When the loop will not stop

Transfers cost tokens

Sub agents and transfer in Google ADK

Handing work to another agent

How a sequence passes work along

Tools and a schema, separately

SequentialAgent in Google ADK: steps in a fixed order

Steps in a fixed order

One key each

ParallelAgent in Google ADK: run agents at the same time

Two things at once

Two agents in the loop

LoopAgent in Google ADK: repeat until good enough

Doing it again until it is right

Reading older material

ADK Workflow graphs versus sequential, parallel and loop agents

Workflows, and which shape to use

It is your data, not the prompt

Session state in Google ADK

State: the agent's scratchpad

Where the persistence actually comes from

user: is a commitment

State prefixes in Google ADK: user, app and temp

Three prefixes, three lifetimes

Which one you need

Long term memory in Google ADK with a memory service

Memory across conversations

Worth doing once

Run Google ADK agents with adk web, adk run and api_server

The dev UI, the CLI and a server

Do not commit it

Do not delete it

Connect Google ADK to Gemini and other models

Using a real model

Prove it refuses

Safety and guardrails for Google ADK agents

Safety, and what to check

Google ADK: build a complete support agent end to end

Build the support desk

Start here quiz

What happens when you create an LlmAgent?