Including covariant information, such as position, force, velocity or spin is\nimportant in many tasks in computational physics and chemistry. We introduce\nSteerable E(3) Equivariant Graph Neural Networks (SEGNNs) that generalise\nequivariant graph networks, such that node and edge attributes are not\nrestricted to invariant scalars, but can contain covariant information, such as\nvectors or tensors. This model, composed of steerable MLPs, is able to\nincorporate geometric and physical information in both the message and update\nfunctions. Through the definition of steerable node attributes, the MLPs\nprovide a new class of activation functions for general use with steerable\nfeature fields. We discuss ours and related work through the lens of\nequivariant non-linear convolutions, which further allows us to pin-point the\nsuccessful components of SEGNNs: non-linear message aggregation improves upon\nclassic linear (steerable) point convolutions; steerable messages improve upon\nrecent equivariant graph networks that send invariant messages. We demonstrate\nthe effectiveness of our method on several tasks in computational physics and\nchemistry and provide extensive ablation studies.\n