Koog, the Kotlin-based framework for building AI agents, has just received a significant update in version 0.4.0 from JetBrains.

This new release brings several exciting features to the table, including native structured output that ensures seamless performance in production environments. One of the major pain points addressed by Koog is when a large language model (LLM) provides precise data format requirements but then fails to execute due to structural issues. To combat this issue, Koog 0.4.0 introduces native structured output supported by some LLMs, complete with pragmatic guardrails like retries and fixing strategies.

When working with models that support structured output, the framework can utilize it directly; otherwise, it falls back to a tuned prompt-and-retry approach powered by a separate model until the payload matches expectations. This ensures that agents are observable, seamless, deployable in users’ stacks, and more predictable.

New Platforms and Models Integrated into Koog

Koog 0.4.0 has also added support for Apple’s iOS platform as part of JetBrains’ focus on Kotlin Multiplatform development. This means developers can now build an agent once and deploy it to both mobile operating systems, along with JVM back ends, using the same strategy graphs, observability hooks, and tests.

However, a specific version – Koog 0.4.1 – is required for building iOS-compatible agents. Additionally, this update includes GPT-5 model support and customizable LLM parameters that allow models to think more critically about complex parameters by adjusting settings such as reasoningEffort.

Innovative Features Improve Performance and Observability

The integration of OpenTelemetry backing in Koog 0.4.0 offers seamless compatibility with both the W&B Weave AI development toolkit and the Langfuse open source LLM engineering platform. This enables developers to see nested agent events, token breakdowns per request, as well as cost analysis.

To combat issues like model calling timeouts, tools malfunctioning, or network hiccups, Koog has introduced RetryingLLMClient with Conservative , Production , and Aggressive presets for fine-grained control. FineSeek model support is also featured in this release.

Concluding Thoughts

Koog 0.4.0 represents a significant step forward in AI agent development, thanks to its innovative features that address real-world challenges faced by developers and users alike. By integrating native structured output, iOS support, GPT-5 model capabilities, OpenTelemetry backing, and more – all within the Kotlin-based framework – JetBrains is empowering developers with powerful tools for creating seamless, observable, deployable AI agents.

Inspired by

• https://www.infoworld.com/article/4047854/jetbrains-updates-kotlin-based-ai-agent-development-framework.html

Sources

• openai.com
• opentelemetry.io
• jetbrains.com
• github.com