JetBrains Open‑Sources KotlinLLM: LLM‑Driven Smart Macros for Compiled Kotlin

JetBrains announced on July 28 that the KotlinLLM prototype is now publicly open‑sourced under the Apache License 2.0.

As an IntelliJ IDEA plugin, KotlinLLM explores the use of LLM‑driven “smart macros” in Kotlin/JVM projects to enable dynamic code generation, runtime logic updates, and hot‑reloading of code.

JetBrains notes that while LLMs are commonly used for code completion, generation, and understanding during development, their application at runtime in compiled applications remains rare.

Existing attempts in the field typically suffer from three pain points:

Direct runtime delegation : slow response, nondeterministic output, high cost, and strong dependence on the stability of external LLM services.

External agent workflow : placing generated logic outside the codebase greatly increases the difficulty of review, automated testing, and version release.

Language limitations : most academic and industrial projects (e.g., byLLM, Nightjar, Healer) target dynamic languages such as Python and do not address the integration challenges of static, compiled languages like Kotlin.

To overcome these bottlenecks, KotlinLLM proposes three targeted solutions:

Explicit awareness : the call stack is marked to indicate that a function is driven by an LLM, making the involvement visible during code review.

Source‑level persistence : LLM‑generated behavior is saved as standard Kotlin source files rather than existing only in runtime memory, allowing the code to be committed, reviewed, tested, and distributed like any other source.

Zero runtime overhead : once the code is generated, it runs as native Kotlin without further plugin dependence; for covered logic the system does not re‑invoke the LLM, eliminating latency, cost, and ensuring reproducibility.

The plugin’s source code, smart‑macro API, and example projects are available in the open‑source repository.