ADK 1.0 for Java released: Google expands framework for AI agents

Google has released version 1.0.0 of the Agent Development Kit (ADK) for Java, making its framework for AI agents available for the widely used Java platform for the first time. The open-source project, which originally started in Python, is thus growing into a multilingual ecosystem – in addition to Java, it also supports Go and TypeScript.

The ADK is aimed at developers who want to build AI agents that not only generate text but also actively interact with their environment. Such agents combine Large Language Models (LLMs) with tools, state management, and external data sources.

More tools and external data sources

Among the most important innovations are additional tools for accessing external information. For example, the GoogleMapsTool integrates location data directly, while the UrlContextTool automatically retrieves and summarizes website content. With code executors, agents can also run their own code locally (Docker) or in the cloud.

Also new is a central app structure with a plug-in system. This allows functions such as logging, context filtering, or global behavioral rules to be defined once and applied to all agents.

Context, control, and collaboration

For longer interactions, ADK introduces context management that summarizes older conversation parts to control token limits and costs. At the same time, the framework supports human-in-the-loop scenarios: agents can pause actions and request approval.

In addition, there are components for session and long-term storage, as well as support for the Agent2Agent protocol, through which agents can communicate with each other across frameworks.

More information is available in the post on the Google for Developers blog and the documentation for the tool.

(mdo)