@pixelkit-labs/mlkit

Expo module (Kotlin) for the on-device Google ML Kit surface:

| Family | What it covers | | :--- | :--- | | GenAI | Gemini Nano through the Prompt API on AICore, plus summarization, proofreading, rewriting | | Vision | barcode, face, face mesh, text recognition, image labeling, object detection, digital ink, pose, selfie and subject segmentation, document scanner | | Natural language | language identification, translation, smart reply, entity extraction | | Embeddings | on-device float vector generation (512/768-dim) and cosine similarity running locally |

This is the expensive one. It pulls in 19 ML Kit artifacts, compiles with -Xskip-metadata-version-check, and pins every kotlin-stdlib in the consuming build, because genai-prompt requires Kotlin 2.3.21 while Expo 57 compiles with 2.1.20. That is why it is a separate package from @pixelkit-labs/native: if you only want CPU clocks and battery temperature, you should not pay for any of this.

Android only. Requires a development build, and Gemini Nano additionally requires a device with AICore and the model downloaded. The TypeScript bridge uses requireOptionalNativeModule, so it resolves to null rather than throwing when the native side is absent, letting callers report unavailable.

Normally installed as a dependency of @pixelkit-labs/sdk, which wraps it in useGeminiNano, useGenAITasks, useVisionAI, useNaturalLanguageAI, and useEmbeddings.

npx expo install @pixelkit-labs/mlkit

Documentation

https://pixelkit-labs.github.io/pixelkit-docs/ — every hook with its inputs, outputs and a contract for each function it exposes.

Licence

MIT