Google ships ADK for Kotlin 1.0 — Android apps can now run AI agents on-device without touching Python
Google's Agent Development Kit for Kotlin has reached 1.0 and full feature parity with the Python version, adding Android-first, on-device agent tooling via LiteRT-LM and ML Kit — meaning production AI agents can now be built natively into an Android app rather than bolted on through a separate Python service.
28 September 2026
Google has released Agent Development Kit (ADK) for Kotlin 1.0, and the significant part isn’t the version number — it’s that Kotlin now has full feature parity with ADK’s Python core, plus a set of Android-first extensions Python never had. Developers building agentic features into an Android app no longer need to stand up a separate Python service and call out to it; the agent logic, tool definitions, and multi-agent coordination can live directly in the Android codebase.
What actually changed
ADK for Kotlin 1.0 brings hierarchical multi-agent systems, multi-turn conversation handling, human-in-the-loop confirmation flows, and long-running tool support to idiomatic Kotlin. On the on-device side, it adds LiteRT-LM and ML Kit for running fast, private agents locally on the phone, Firebase AI Logic for hybrid cloud/on-device workflows, and Room/AppSearch integration so agent state survives a process restart. Tool schemas are generated at compile time via annotations rather than runtime reflection, which is a meaningful performance and type-safety difference for anything shipping to production.
Why this is a commissioning decision, not just a developer tool
For anyone scoping an Android build with an AI feature — a support assistant, an in-app copilot, personalised recommendations that need to work offline — this closes a gap that used to force an architectural compromise: either run the agent server-side and eat the latency and data-handling questions, or maintain a second Python codebase alongside the Android one. Native, on-device agent support means faster response times, features that keep working without connectivity, and one less service to secure and maintain. It also strengthens the case for going native or cross-platform Kotlin-first on Android rather than routing AI features through a web wrapper.
So what
If your product roadmap includes an AI-powered feature on Android — on-device or hybrid — this is the point where “we’ll figure out the AI architecture later” stops being a safe default, because the native tooling to do it properly now exists. See our iOS & Android work for how we scope agentic features into a build from day one, or get in touch to talk through what an on-device AI feature would actually take for your app.