Jraph provides data structures for representing graphs, utilities for graph manipulation, and a collection of pre-built GNN models. It leverages JAX’s functional transformations for flexible neural network construction and supports distributed training for large-scale graph networks with explicit edge messaging across multiple devices.