adk-python

Google's Agent Development Kit (ADK) is a Python framework for building AI agents — programs that can receive a goal, break it into steps, use tools like web search or code execution, and work toward completing that goal autonomously. The problem it addresses is that creating capable AI agents is complex: you need to coordinate between AI models, define what tools they can use, handle multi-step conversations, and manage how multiple agents hand off tasks to each other.

ADK takes a code-first approach, meaning you define your agent's behavior directly in Python rather than through a visual interface or configuration wizard. A single agent is defined by pointing to an AI model (it works well with Google's Gemini models but is model-agnostic), giving it instructions, and attaching tools it can call — like a web search function. Multi-agent systems go further: you can create a coordinator agent with specialist sub-agents nested inside it, and the framework handles routing tasks to the right agent automatically. It also ships with a built-in development UI for testing and debugging agent behavior, a command-line tool for running evaluations against test sets, and supports deployment to Google Cloud's infrastructure. It can also participate in agent-to-agent communication with other services via the A2A protocol.

ADK is installed via pip and is suited for developers who want to build sophisticated AI workflows — automated research assistants, customer service bots, code-execution pipelines, or any task that benefits from an AI that can plan and act over multiple steps.