Google ADK

Agentic AIFrameworks and SDKsPublished By Simon Budziak

Google ADK (Agent Development Kit) is Google's open-source, code-first framework for building AI agents in Python, TypeScript, Go, Java, or Kotlin. It treats agent construction as ordinary software engineering: you compose agents into hierarchies, wire tools explicitly, and deploy anywhere from a container to managed Google Cloud runtimes.

How does Google ADK structure an agent system?

Agents are plain code objects carrying instructions, a model, and tools, composed into trees where a root agent delegates work to specialized children. This native hierarchy is ADK’s signature, so multi-agent systems need no separate orchestration layer: workflow agents handle sequential, parallel, and loop control flow deterministically, while model-driven agents route dynamically. Tools attach as plain functions, OpenAPI specs, or MCP servers, and built-in tracing records every step for debugging.

When should you pick ADK over other frameworks?

When you want an official batteries-included toolkit and may land on Google Cloud, where deploying to managed runtimes requires no code changes. Teams maximizing ecosystem breadth usually reach for LangGraph or LangChain; teams wanting a minimal typed loop compare Pydantic AI. ADK wins when first-party evaluation tooling, deployment paths, and multi-language support outweigh third-party ecosystem size.

Frequently asked questions

Is Google ADK limited to Gemini models?

No. It is optimized for Gemini but model agnostic: the same agent code runs against other hosted providers or locally running models through adapters, so the framework choice does not lock you to one vendor.

What languages does Google ADK support?

Python, TypeScript, Go, Java, and Kotlin, with the core concepts consistent across them. That multi-language coverage is unusual among agent frameworks and matters for teams whose services are not written in Python.

Summarize this page with

See this working in a system we built