Google Agent Development Kit (ADK) is a flexible, modular open-source framework for developing and deploying AI agents, optimized for the Gemini model family and Google Cloud ecosystem while remaining model-agnostic and deployment-agnostic. It solves the challenge of building sophisticated multi-agent applications by providing a code-first approach with workflow agents for predictable pipelines, LLM-driven dynamic routing for adaptive behavior, and hierarchical agent composition for complex coordination and delegation. ADK brings the same engineering discipline of traditional software development to the world of AI agents, with strong emphasis on testability, evaluation, and production deployment.
ADK provides workflow agents including Sequential, Parallel, and Loop patterns for predictable pipelines, alongside LLM-driven dynamic routing for adaptive agent behavior. The framework equips agents with diverse tool capabilities including pre-built tools for Search and Code Execution, custom function tools, third-party library integrations, and the ability to use other agents as tools. Built-in bidirectional audio and video streaming enables human-like conversational interactions, while the evaluation framework allows systematic assessment of both final response quality and step-by-step execution trajectories against predefined test cases.
Google ADK targets TypeScript and Python developers building AI agents within the Google Cloud ecosystem, including applications that leverage Gemini models, Google Search, Vertex AI, and Cloud Run for deployment. It integrates with Google Cloud Agent Engine for managed scaling, supports containerized deployment through Docker, and works with both Google and third-party model providers for maximum flexibility. ADK is particularly well-suited for enterprise teams building multi-agent applications on Google Cloud infrastructure, with built-in support for evaluation, monitoring, and production deployment patterns that align with Google Cloud best practices.