Interest among ML practitioners in online/streaming machine learning, where models learn continuously from each new sample, is growing quickly. To facilitate the development of online ML applications, the creators of Creme and scikit-multiflow have combined these projects into River. River enables users to develop models that update in real-time, including to adapt to concept drift. It is also designed to build production-ready systems, which are much faster and more efficient than their counterparts created with Tensorflow and PyTorch.To support the streaming context, River includes an efficient Cython extension of dictionary structures for operations on unidimensional arrays.