pymdp is an open-source Python package for simulating active inference agents in discrete state-spaces. I lead its development and maintenance.

The library provides reference implementations of the core algorithms used in active inference research — variational inference over hidden states, expected free energy computation for policy selection, and learning over Dirichlet priors — along with a clean API for building custom generative models.

Published as a paper in the Journal of Open Source Software (2022). It’s used by labs around the world for both research and teaching.