from tensorflow.contrib.rnn import RNNCell

The RNNCell class is a base class for creating recurrent neural network cells in TensorFlow. It defines the interface for a cell that can be used in an RNN layer, such as a basic RNN cell, LSTM cell, or GRU cell.

To create a custom RNN cell, you can subclass the RNNCell class and implement the methods defined in the interface, such as the call method that takes input and previous state and returns output and new state.

Using the RNNCell class makes it easy to build custom RNN models in TensorFlow, as you can simply stack multiple cells together to create a multi-layer RNN.