Preview

Building Agents with Google ADK

A Developer's Reference for Production Agent Development

Load this book when:

Building, debugging, or deploying a Google ADK agent (Python or Java).

Choosing an orchestration pattern (Sequential / Loop / Parallel / Workflow graph) or a Gemini model for an ADK agent.

Migrating an ADK 1.x project to 2.0, or resolving an import/dependency error after an upgrade.

Setting up local testing, evaluation, or deployment to Cloud Run / Vertex AI Agent Engine.

Audience: marketplace reference. Last updated: 2026-06-14. Scope of validity: ADK Python 2.2.0 (2026-06-04, latest stable on the 2.x line) — note the 1.x line is still maintained in parallel and is at v1.35.0 (2026-06-10), so 1.27.2 (referenced in Ch 17) is NOT the 1.x head; ADK Java 1.4.0. Refresh cadence: every minor ADK release (~bi-weekly) — this is volatile/tactical API content (see Writing Books for AI Agents, Ch 14, RULE 14.4).

Version coverage: v1.27.2 (March 2026) · ADK 2.0 Alpha notes in Chapter 14 corrected 2026-06-14: ADK 2.0 reached GA on 2026-05-19; latest stable is v2.2.0 (2026-06-04). See Chapter 15.

Add to library to read more

Table of Contents

Chapter 2: Decision Matrix

Persist and resume sessions

Chapter 7: Memory Services

Development

Production (v1.26+: supports memory consolidation)

Store a session as memories programmatically

Chapter 8: Callbacks

Chapter 9: Plugins

Basic run

Run with a persistent session backend

Save session on exit (for replay or debugging)

Resume a previously saved session

Basic launch

Custom port and host

Record real agent interactions as test fixtures

Cloud Run

Setup

Dev

Eval & Optimize

Point adk web at the agents/ directory — it finds all agents inside

Point adk run at the specific agent directory

Cloud Run target

agent.py — load env first

❌ Broken in 1.27.2

✅ Option A — explicit skills list

Appendix: Rules & Gotchas

Building Agents with Google ADK

About

Preview