AI SDK Kotlin/ai.torad.aisdk/LowLevelLanguageModelApi

LowLevelLanguageModelApi

@RequiresOptIn(level = RequiresOptIn.Level.ERROR, message = "Direct LanguageModel execution is a low-level provider API. Use Agent.generate/Agent.stream or a high-level SDK API for application prompts, or opt in with @OptIn(LowLevelLanguageModelApi::class).")

@Target(allowedTargets = [AnnotationTarget.CLASS, AnnotationTarget.FUNCTION, AnnotationTarget.PROPERTY, AnnotationTarget.PROPERTY_GETTER, AnnotationTarget.PROPERTY_SETTER, AnnotationTarget.CONSTRUCTOR, AnnotationTarget.TYPEALIAS])

annotation class LowLevelLanguageModelApi(source)

Marks direct LanguageModel execution as low-level provider access.

Application prompts should normally go through an Agent or another high-level SDK API so tool loops, middleware, telemetry, persistence, and output handling stay in one place. Calling the model directly is still supported for provider authors, tests, and deliberately low-level integrations, but it must be explicit via @OptIn(LowLevelLanguageModelApi::class).

Public concrete LanguageModel implementations should put this marker on their execution overrides so direct calls on the concrete type stay gated too.