Google has released the Agent Development Kit for Kotlin 1.0. This is a production-ready framework for building AI agents across Kotlin, Android, and JVM or server applications.
The new release brings Kotlin to feature parity with Google's ADK for Python and Java. It also adds Android-specific capabilities for on-device and hybrid artificial intelligence.
Developers using ADK for Kotlin 1.0 no longer need to rely on Python for agentic logic. They can use idiomatic Kotlin APIs for orchestration, tool support, persistence, memory, and human-in-the-loop workflows.
Built on Kotlin Multiplatform, the ADK runs on all supported platforms from server-side applications to mobile devices. Google states that its architecture is completely agnostic to specific model backends, session providers, or memory systems.
ADK for Kotlin 1.0 adds support for hierarchical multi-agent systems where a parent agent can delegate tasks to a child. It also includes context compaction and multi-turn conversation for automatic context management and history summarization to reduce token usage.
Other features include session management, which allows agent state to be paused, serialized, and restored, alongside first-class Java interoperability.
Tools can be declared using specific annotations such as @Tool and @Param. KSP uses these annotations to generate Kotlin function schemas at compile time, avoiding runtime reflection for better type-safety and performance.
PiNCAMP Android engineer Arjun Kumar noted on LinkedIn that handling tool schemas at compile time with KSP keeps startup fast on mobile targets.
Human-in-the-loop workflows help developers prevent unintended or improper tool usage by requiring explicit human confirmation before sensitive actions are executed. Developers can require confirmation by setting requireConfirmation in a tool declaration.
AI Dev Weekly newsletter maintainer Joske Vermeulen recommended starting with one resumable agent and explicit tool confirmation before reaching for a hierarchy of agents, noting that production readiness depends more on lifecycle recovery and deterministic tool boundaries than on agent count.
The framework also includes skills, allowing developers to manage procedural knowledge stored in SKILL.md files that are dynamically loaded only when needed using progressive disclosure.
For on-device inference, the framework supports both LiteRT-LM and ML Kit in beta. For cloud and hybrid artificial intelligence, the ADK integrates with Firebase AI Logic.
