ml-kit-genai-prompt-api

Add the com.google.mlkit:genai-prompt dependency (version 1.0.0-beta4 or later) to your app-level build.gradle and ensure minSdk is at least 26. Then create a model with Generation.getClient() and verify its FeatureStatus is AVAILABLE before inference.

Define a data class annotated with @Generable and @Guide field descriptions, then build a typed request with generateTypedContentRequest. Refactor functions to return the typed object instead of a String, and check availability with isStructuredOutputFeatureAvailable() first.

The API requires Android API level 26 or higher, so minSdk must be updated to 26 if it is lower. If you use the genai-schema-compiler dependency, the KSP plugin must be version 2.3.6 or above.

Use the prefix caching API when your prompt exceeds 200 words. Caching the shared prefix reduces redundant processing across repeated inference calls that share the same prompt beginning.

Inference fails if the model is not fully downloaded and available on the device. Check that the model's FeatureStatus is AVAILABLE before the first generateContent call, and release instances with close() when the Activity, Fragment, or ViewModel is destroyed.