MemeLab — a contagion simulator in your browser

MemeLab is a free, deterministic epidemic simulator that runs entirely in your browser: a SEIR cellular automaton where every cell is one person and every tick is one day. I built it so you can watch outbreaks spread, evolve strains, switch defenses on and off, and fit R₀ to real outbreak curves — on a CPU, WebAssembly or WebGPU engine.

The simulator needs JavaScript. Enable it and reload to run it; the text below describes what it does.

What you can simulate

• SEIR outbreaks — Susceptible, Exposed, Infectious, Recovered and Dead compartments, with waning immunity (SEIRS) and optional rebirth.
• Five geometries — square, triangular, hexagonal and Voronoi lattices, plus a mean-field (well-mixed) mode.
• Four defenses — masks, vaccines, lockdown and reactive quarantine, each with its own uptake, compliance and effectiveness sliders.
• Strain evolution — mutation and natural selection over a multi-strain pool, so you can watch phenotypes drift under pressure.
• R₀ estimation — paste an observed cumulative cases or deaths curve and let the estimator search for the strain parameters that reproduce it.
• Reproducible permalinks — every run is seeded; a permalink replays the same simulation exactly.

Three engines

The same model runs on three backends. The plain CPU engine works in any modern browser. The WebAssembly engine is a faster drop-in for the same step. The WebGPU engine moves the transmission pass onto your graphics card for large grids; it needs a browser with WebGPU enabled and a real hardware adapter — software renderers are treated as unavailable. If a backend can't start, the simulator falls back to the CPU engine automatically and tells you why.

Questions I get asked

Is it free?

Yes. There is no account, no paywall and no in-app purchase. The simulation runs on your machine.

Does it run offline?

Mostly. Once the page has loaded, the simulator makes no network requests — all computation happens locally. It isn't installed as an offline app, though, so you still need a connection to load it the first time.

Is it deterministic?

Yes. Every run is driven by a seeded pseudo-random generator, and the permalink encodes the seed along with the grid size, disease genes, defenses, theme and speed. Anyone opening that link replays the same run.

Is it a real epidemiological model? How accurate is it?

It is a real SEIR compartment model expressed as a stochastic cellular automaton, with R₀ computed analytically from the strain's parameters and an effective R measured from the run. It is built for intuition and teaching, not forecasting: the population is a lattice of identical people with local contacts, and the bundled outbreak datasets are sparse and illustrative. Treat the numbers as qualitative.

How is it different from an agent-based model?

It is close to one. Each cell is an individual with its own state, strain and defense flags, so it is agent-based in that sense — but agents don't move and don't have schedules or heterogeneous behaviour. Contact structure comes from the lattice geometry (or the mean-field mode, which approximates a classic well-mixed ODE model). That keeps it fast, deterministic and easy to reason about.

From the blog

I write about outbreaks, memetics and the engineering behind MemeLab. Read the blog.

• Armchair epidemiology: the four-hour model that held up — 2026-09-02