Covasim and Starsim

Covasim (COVID-19 Agent-based Simulator) builds a population of simulated people with ages and household sizes drawn from a country's demographic data.^1,2 It links them in contact networks for households, schools, workplaces, long-term care homes and the wider community, and the virus passes along those links.² Each run projects indicators such as infections and peak hospital demand.¹ Its main use is comparing interventions: distancing, school closures, testing, contact tracing, quarantine and vaccination, including delays and people lost to follow-up.^1,2

The Institute for Disease Modeling (IDM) built it, with the University of Copenhagen, the Burnet Institute, GitHub and Microsoft.¹ The model paper appeared in PLOS Computational Biology in July 2021.² According to its makers it was used to inform policy decisions in the US, UK and Australia, among more than a dozen countries; published studies cover test-trace-quarantine, school reopening in the UK and border reopening in Vietnam.¹ It is written in pure Python, and a realistic scenario runs on a laptop in under a minute.²

Covasim is now part of Starsim, a general framework for agent-based disease models.^1,3 IDM, the Burnet Institute and other collaborators develop it, and version 1.0 launched in July 2024.³ Starsim handles several diseases at once, both infectious and not, and other health states such as pregnancy.⁴ Models built on it cover HIV, HPV and syphilis, tuberculosis, RSV, Ebola, cholera, family planning and malnutrition.⁴ It runs in Python or R.³ Covasim 4.0.1 came out in October 2026, and the README offers help to anyone moving from version 3.^1,5

Both install with pip and come with tutorials, including how to calibrate a model to case data and how to build a custom population.^1,3 Covasim comes with no promise of support or that the code works as intended.¹ A model is only as good as its contact networks and parameters, and these are generic unless the user supplies local data. Both are MIT-licensed.^1,3