Google Agent Development Kit for Kotlin Reaches Feature Parity with Python, Supports On-Device AI

Google has unveiled the Agent Development Kit (ADK) for Kotlin 1.0, a production-ready framework designed for constructing AI agents across Kotlin, Android, and JVM/server applications. This version brings Kotlin to parity with Google's existing ADK for Python and Java, while introducing Android-specific functionalities for on-device and hybrid AI.

With ADK for Kotlin 1.0, developers can now perform agentic tasks using Kotlin APIs, eliminating the need for Python. This includes orchestration, tool support, persistence, memory, and human-in-the-loop workflows. The framework operates on all supported platforms, from server-side applications to mobile devices, owing to its architecture that is completely agnostic to specific model backends, session providers, or memory systems.

Key enhancements in this version include support for hierarchical multi-agent systems, where a parent agent can delegate tasks to a child; context compaction and multi-turn conversation for efficient context management and history summarization; session management allowing agents to pause, serialize, and restore their state; and first-class Java interoperability.

Google's approach to tools and human-in-the-loop workflows in ADK for Kotlin 1.0 emphasizes compile-time schema generation using KSP, enhancing type-safety and performance. This method avoids runtime reflection, which speeds up startup on mobile targets. For high-impact tools, developers can require explicit human confirmation before execution, ensuring safe handling of sensitive operations such as initiating bank transfers.

A practical example is provided, demonstrating how to enforce confirmation for tools involved in sensitive transactions while leveraging Android's persistence services like Room, AppSearch, and Android storage.

The ADK for Kotlin also incorporates Android-native features for building agents that integrate on-device and cloud-based AI. It supports LiteRT-LM and ML Kit for on-device inference, with the latter currently available as a beta feature. For cloud and hybrid AI, ADK integrates with Firebase AI Logic, enabling Android apps to run agents locally while seamlessly leveraging cloud-based models when necessary.

The ADK for Kotlin is open source and accessible on GitHub, making it available for developers to leverage.