We open-source the JAX implementation of the path-specific counterfactual fairness method published in Chiappa, AAAI 2019. This method allows to devise individual-level-fair decision systems in cases where a sensitive attribute might influence the decision along both fair and unfair causal pathways. The causal DAG underlying the observed data is augmented with latent variables that represent shared unobserved randomness between the factual and counterfactual worlds, and variational autoencoding is used to learn the joint distribution of the causal DAG and to perform inference on the latent variables. For a new individual, the latent variables are then inferred and used in conjunction with the causal DAG to produce a fair decision.