Google Kotlin Agent Kit Reaches Full Feature Parity, Enables On-Device AI

Google has shipped version 1.0 of its Agent Development Kit for Kotlin, bringing a complete, production-grade framework for building autonomous AI agents directly in Kotlin and Java environments—no need to drop into Python.

The framework now matches the capabilities of Google's Python ADK while introducing mobile-first features. Built on Kotlin Multiplatform, it runs across servers, desktop JVMs, and Android devices, and Google designed it to remain independent of any particular AI model, session backend, or memory layer.

What's New in 1.0#

The release adds several production features:

• Hierarchical multi-agent coordination: Parent agents can delegate subtasks to child agents for more complex workflows
• Context compaction and multi-turn memory: Automatic summarization of conversation history to reduce token overhead
• Session persistence: Agents can pause their state, serialize it, and resume later
• Seamless Java interoperability: First-class support for existing Java codebases

Compile-Time Safety for Performance#

A core design choice involves how tools (the functions agents can invoke) are declared. Rather than using runtime reflection to discover schemas, ADK for Kotlin relies on Kotlin Symbol Processing (KSP) annotations like @Tool and @Param to generate function signatures at compile time. This approach gains stronger type checking and faster startup—a critical advantage on mobile devices where reflection overhead matters.

Human-in-the-loop workflows let developers require explicit confirmation before sensitive operations execute, such as bank transfers. Setting requireConfirmation = true on a tool definition lets the generated agent pause and await user approval before proceeding.

Domain-Specific Knowledge Without Context Bloat#

ADK introduces skills—procedural knowledge stored in SKILL.md files that agents load only when needed. Rather than loading an entire playbook into the model's context at every turn, the framework uses progressive disclosure to provide agents access to domain-specific procedures on demand.