The C++ libraries currently provide:
A command line interface written in C++ which executes an inference.
An example of how to use C++ API with a trained model.
We are still preparing a well-formatted C++ API reference manual, however you can read through the header files where most of classes and functions are documented in Doxygen format. The header files can be found under include directory.
The example MNIST runtime is a good starting point to understand how to use C++ API for neural network inference.