Kotlin gets a new AI agent framework

Looking to make Kotlin a first-class language for AI, JetBrains has introduced Koog, an open source agentic framework for building AI agents within the Java virtual machine (JVM) ecosystem via a Kotlin domain-specific language (DSL).

Introduced May 22 and found on GitHub, Kotlin-based Koog leverages a modern DSL and provides developers with tools to build intelligent, autonomous agents with the productivity offered by Kotlin, JetBrains said. Koog was built because JetBrains believes Kotlin developers should have an AI framework as “powerful and flexible” as Kotlin itself. Prior to Koog, there has been no comprehensive Kotlin-native agentic framework solution, the company said. 

Koog is designed to tackle obvious challenges with features such as fast onboarding, simplified agent creation, predefining strategies, and seamless Model Context Protocol (MCP). More advanced capabilities also are addressed, such as response streaming and efficient handling of long contexts and query histories. Key features of Koog include:

A pure Kotlin implementation to build and run AI agents in Kotlin without external service dependencies

A modular feature system to extend agent capabilities

Tool integration to create custom tools to give agents access to external systems and resources

Out-of-the-box solutions for situations such as streaming from large language models (LLMs) and invoking multiple tools on the fly from one LLM request

Forcing LLMs to provide specific results

Customizable workflows

Traceable workflows