Is TensorFlow Lite a Native API?

If I want a custom model or if I need to do custom training, I'm going to use TensorFlow on the cloud to create that model. And then once it's on the device, then I have a choice. I could actually use Tensor Flow Lite directly. We don't necessarily recommend that people go directly there because it sounds like we have value add in either the neural network API stuff for performance reasons and an ML kit. If you happen to be writing your app in Java or Kotlin, as well as bundled models that you could use instead of the custom model.