dokimos

Worth considering if your AI product already lives in the Java/Kotlin ecosystem and you want evaluation to behave like part of normal testing. The evidence supports a fairly clear positioning: this is not a general-purpose chat model and not a hosted eval platform; it is better understood as a library for embedding LLM and agent evaluation into JUnit and CI, closer to an AI-aware testing framework than to a standalone observability product.

Its practical value is straightforward: teams can define and run evaluations for LLM, agent, and RAG behavior inside existing JVM projects. The GitHub repo and homepage consistently describe Java/Kotlin support, JUnit and CI execution, and integrations with Spring AI, LangChain4j, Koog, and Embabel. The X posts from JetBrains and Kotlin are useful as heat signals showing that the Koog integration got attention, but they do not by themselves prove eval quality, reliability, or ease of adoption. Official repo and product docs are stronger evidence for capability boundaries than social reach.