Mesa

Mesa is a toolkit for writing agent-based models in Python. An agent-based model describes individuals, people, animals or firms, and the rules each follows, then runs them together to see what patterns emerge from their interaction. Mesa supplies the common parts: agents and a way to step them, spaces such as grids and networks, tools to record each step, and an interactive view in a web browser or a Jupyter notebook.^1,2 The project presents it as the Python alternative to NetLogo, Repast and MASON.²

It is an open-source project that welcomes contributions.¹ The paper describing version 3, by Ewout ter Hoeven and six co-authors, appeared in the Journal of Open Source Software in March 2025.^1,4 Version 3.5.1 came out in March 2026, version 4 is in pre-release, and the project is very active, with commits in October 2026.^5,6 It takes part in Google Summer of Code.^1,2

The same project makes Mesa-Geo, which adds a geographic space. Its agents carry a real geometry, a Shapely shape, with a coordinate reference system, and can be loaded from shapefiles, GeoJSON or GeoPandas tables.³ With it, agents can stand on real streets and parcels instead of an abstract grid.

Both install with pip.^2,3 Models are plain Python, so they can use Python's data tools for input and analysis.¹ Mesa ships with no population or disease of its own; everything specific to a place has to be built by the modeller. For a visual, lower-code start, see NetLogo; for a platform built around GIS from the outset, see GAMA. Mesa and Mesa-Geo are Apache-2.0.^2,3