Google Launches ADK for Kotlin and Android: Building AI Agents On-Device

AI Agents Go Mobile#

Google has launched ADK for Kotlin and ADK for Android 0.1.0, bringing agentic AI workflows to mobile and backend development. The releases follow 1.0.0 versions of ADK for Java and Go, plus a Python 2.0 beta.

ADK is an open-source framework for developing and running AI agents. The Kotlin variant targets backend projects, while Android brings specialized on-device optimization for privacy-enhancing applications.

Edge AI at Scale#

Gemini Nano—Google's lightweight model—now runs on over 140 million Android devices. As developers seek faster, cheaper, and privacy-preserving apps, the ability to execute AI models directly on hardware has become critical. ADK orchestrates the complexity: "ADK removes that friction by managing all the complex orchestration, context handling, and error handling for you."

Hybrid Orchestration#

With a few lines of Kotlin, developers can:

• Swap models dynamically based on task requirements
• Mix on-device and cloud models in a single multi-agent system
• Share session state across agents seamlessly
• Use cloud-based orchestrators that delegate tasks to on-device sub-agents
• Build sequential agents for multi-step workflows
• Access local documents via on-device retrieval agents, keeping data offline

In a demo trip-assistant, a cloud orchestrator handles user interaction while delegating booking verification to on-device sub-agents. Retrieval agents extract data from locally stored documents using Gemini Nano, and a validation agent compares results—all while sensitive data stays offline.

Getting Started#

To add ADK to an Android app, include:

implementation("com.google.adk:google-adk-kotlin-core-android:0.1.0")

For Kotlin backends:

implementation("com.google.adk:google-adk-kotlin-core:0.1.0")
ksp("com.google.adk:google-adk-kotlin-processor:0.1.0")