Google releases ADK for Kotlin 1.0, enabling production AI agents across mobile and server

Google has made the Agent Development Kit (ADK) for Kotlin officially available as version 1.0, bringing a production-ready framework for building AI-powered agents on both Android and server-side JVM environments. The release aligns Kotlin developers with capabilities already present in ADK for Python and Java, while adding Android-specific features for on-device and hybrid cloud deployments.

What's in the 1.0 Release#

ADK for Kotlin 1.0 achieves feature parity with the core ADK platform while introducing a suite of Android-first extensions. Developers can now run fast, private on-device agents using LiteRT-LM and ML Kit (beta), orchestrate workflows across hybrid environments via Firebase AI Logic, and persist agent state using Room database and AppSearch for indexed memory. The framework supports both Android applications and enterprise server-side workloads written in Kotlin.

The toolkit is built on a Kotlin Multiplatform (KMP) foundation that remains agnostic to specific model backends, session providers, or memory systems. This design allows teams to mix local on-device models with cloud-hosted models as needed.

Core Capabilities#

The 1.0 release includes several advanced patterns for multi-agent systems:

• Hierarchical multi-agent orchestration — chain specialized child agents and delegate tasks
• Context management — summarize conversation history to stay within token limits for long-running sessions
• Human-in-the-loop workflows — pause execution, request user confirmation for sensitive actions, then resume
• Type-safe function calling — use Kotlin annotations (@Tool, @Param) to generate function schemas at compile time via KSP (Kotlin Symbol Processing), eliminating runtime reflection
• Session resumability — pause and restore agent interactions across user sessions
• Enterprise integration — native support for Vertex AI session services, RAG memory, and memory banks

Example: Production Incident Triage Agent#

The announcement includes a detailed walkthrough of an incident response agent that investigates database alerts. The agent discovers and loads a standard operating procedure from a file, then autonomously runs diagnostics by calling tools that query infrastructure metrics, fetch recent deployments, and notify the on-call team. The example demonstrates how KSP generates compile-time function definitions, eliminating reflection overhead and providing type safety for functions.