This library implements Cartesian genetic programming (e.g, Miller and Thomson, 2000; Miller, 2011) for symbolic regression in pure Python, targeting applications with expensive fitness evaluations. It provides Python data structures to represent and evolve two-dimensional directed graphs (genotype) that are translated into computational graphs (phenotype) implementing mathematical expressions. The computational graphs can be compiled as Python functions, SymPy expressions (Meurer et al., 2017) or PyTorch modules (Paszke et al., 2017). The library currently implements an evolutionary algorithm, specifically (mu + lambda) evolution strategies adapted from Deb et al. (2002), to evolve a population of symbolic expressions in order to optimize an objective function.