Koog 1.0 Release: What AI App Developers Need to Know
JetBrains released Koog 1.0 on May 27, 2026, announced at the KotlinConf 2026 keynote. Koog is JetBrains’ open-source framework for building AI agents in Kotlin and Java. The 1.0 label marks a specific commitment: stable modules carry a guarantee of no breaking API changes for at least one year.
This article covers what the release actually contains, what the stability guarantee means in practice, and what developers should check before deciding whether to evaluate Koog now.
What Koog is
Koog provides building blocks for agentic applications on the JVM and Kotlin Multiplatform targets: tools, workflows, persistence, memory, observability, and integrations with existing JVM and Kotlin Multiplatform projects. JetBrains introduced it at KotlinConf the previous year; 1.0 follows several public releases and internal production use.
Koog is open-source and available now via dependency updates. If you are already using an earlier version, the migration path is documented in the Koog 1.0 release notes.
What 1.0 actually delivers
According to the official JetBrains announcement, the headline items in Koog 1.0 are:
Stability commitment. Stable modules in 1.0 will not have breaking API changes for at least one year. This is the primary signal that JetBrains considers the core architecture settled. Note that beta modules are still labeled beta and do not carry this guarantee.
Redesigned Java interop. The Java interop layer has been rebuilt with a cleaner and more consistent API. This matters for teams running mixed Kotlin/Java codebases or integrating Koog agents into existing JVM services.
Decoupled HTTP transport. HTTP transport is now independent of the framework core, which makes it easier to integrate Koog into infrastructure that already uses specific HTTP clients.
OpenTelemetry support across targets. Observability via OpenTelemetry is now available across Koog’s Kotlin Multiplatform targets. Teams that run agents in mixed environments — JVM, Android, or other Kotlin Multiplatform targets — can get consistent telemetry without per-platform instrumentation.
Local Android AI. New provider integrations support running LiteRT models locally on Android devices, extending Koog’s agent capabilities to on-device inference.
Improved persistence and memory. Long-running agents can now rely on better persistence and memory support, which matters for workflows that carry state across sessions or need to resume interrupted processes.
Anthropic prompt caching support. Koog now supports Anthropic’s prompt caching, which can reduce latency and token costs for agents that repeat substantial context across calls.
What ACCESS_OR_AVAILABILITY means for this release
Koog 1.0 is available now as an open-source project. Existing users update their dependencies; new users can start from the docs. The key caveat is the stable/beta module distinction: not everything in Koog 1.0 carries the one-year stability guarantee. Before building on a specific module, verify its stability status in the release notes and docs.
The framework targets Kotlin and Java developers. If your team works in Python, Go, Node.js, or another language, Koog is not relevant to your current stack — there are no other-language bindings announced in this release.
Who should look at this now
Developers building AI agent workflows on Kotlin or Java who want a stable, JetBrains-maintained foundation with production-oriented observability will find the 1.0 release worth evaluating. The stability commitment reduces one category of adoption risk that existed in earlier versions.
Teams already using LangChain, LlamaIndex, or another agent framework in a different language, or teams that need ready-to-use AI apps rather than a development framework, can skip this release.
The full announcement and release notes are in the official JetBrains AI blog post.