google/adk-docs

What it solves

It simplifies the process of building, evaluating, and deploying AI agents by treating agent creation more like traditional software development. It provides a structured, code-first approach to managing agentic architectures, ranging from simple tasks to complex, multi-agent workflows.

How it works

ADK is a modular, model-agnostic, and deployment-agnostic framework. It allows developers to define agent logic and orchestration directly in code across multiple languages (Python, TypeScript, Go, Java, and Kotlin). It enables the integration of diverse capabilities through a tool ecosystem that supports custom functions and OpenAPI specs, and allows for the composition of specialized agents into scalable hierarchies.

Who it’s for

Software developers who want more flexibility, control, and testability when building AI agents, particularly those working within the Google ecosystem but seeking a model-agnostic toolkit.

Highlights

• Code-First Approach: Logic and orchestration are defined in code for better versioning and testability.
• Modular Multi-Agent Systems: Supports building scalable applications by composing specialized agents into hierarchies.
• Rich Tooling: Integrates pre-built tools, custom functions, and OpenAPI specifications.
• Observability: Includes built-in tracing and monitoring for debugging and optimizing workflows.
• Flexible Deployment: Supports containerization for Cloud Run, GKE, or scaling via Agent Runtime.