Agents

Google Releases Agent Development Kit for Kotlin 1.0

Google has launched the Agent Development Kit for Kotlin 1.0, bringing the language to feature parity with Python and enabling developers to build on-device and hybrid AI agents for Android.

InfoQ AI2 days agoAgents
Image: InfoQ AI

Google has officially launched the Agent Development Kit (ADK) for Kotlin 1.0, a production-ready framework designed to build AI agents across Android, Kotlin, and JVM/server environments. This release establishes feature parity with Google's existing Python and Java ADKs, eliminating the need for mobile and backend developers to rely on Python for agentic logic. Built on Kotlin Multiplatform, the framework remains agnostic to specific model backends, session providers, or memory systems.

The ADK for Kotlin 1.0 introduces hierarchical multi-agent systems, context compaction, multi-turn conversation management, and session serialization. To optimize performance on mobile devices, the framework avoids runtime reflection. Instead, developers can declare tools using the @Tool and @Param annotations, allowing the Kotlin Symbol Processing (KSP) tool to generate schemas at compile time. Arjun Kumar, an Android engineer at PiNCAMP, noted that using KSP to handle tool schemas "keeps startup fast on mobile targets."

For sensitive operations like financial transactions, developers can implement human-in-the-loop workflows by setting the requireConfirmation parameter. This feature integrates with native Android persistence services, such as storing chat sessions in Room, indexing memory in AppSearch, and saving files directly to local storage. Additionally, the framework manages procedural knowledge through SKILL.md files. These are loaded dynamically via progressive disclosure, preventing the agent's context window from becoming overloaded with unnecessary playbooks.

For execution, the ADK supports both on-device and hybrid AI architectures. Local inference is powered by LiteRT-LM and a beta version of ML Kit, while cloud-based operations integrate with Firebase AI Logic. This allows applications to run lightweight agent tasks locally and seamlessly transition to cloud models when more complex processing is required. The open-source framework is currently available on GitHub.

This is our own summary of reporting by InfoQ AI

More in Agents