This paper presents reservoirpy, a Python library for Reservoir Computing (RC) models design and training, with a particular focus on Echo State Networks (ESNs). The library contains basic building blocks for a large variety of recurrent neural networks defined within the field of RC, along with both offline and online learning rules. Advanced features of the library enable compositions of RC building blocks to create complex "deep" models, delayed connections between these blocks to convey feedback signals, and empower users to create their own recurrent operators or neuronal connections topology. This tool is solely based on Python standard scientific packages such as numpy and scipy. It improves RC time efficiency with parallelism using joblib package, making it accessible to a large academic or industrial audience even with a low computational budget. Source code, tutorials and examples from the RC literature can be found at https://github.com/reservoirpy/reservoirpy while documentation can be found at https://reservoirpy.readthedocs.io/en/latest/?badge=latest