https://github.com/ikkeseb/vivarium
Client-side artificial-life sandbox: a deterministic, seedable gallery of cellular-automata and life systems. TypeScript + Vite, no backend, no GPU.
https://github.com/ikkeseb/vivarium
artificial-life canvas cellular-automata conways-game-of-life lenia simulation typescript vite
Last synced: about 15 hours ago
JSON representation
Client-side artificial-life sandbox: a deterministic, seedable gallery of cellular-automata and life systems. TypeScript + Vite, no backend, no GPU.
- Host: GitHub
- URL: https://github.com/ikkeseb/vivarium
- Owner: ikkeseb
- License: mit
- Created: 2026-06-20T09:48:36.000Z (15 days ago)
- Default Branch: main
- Last Pushed: 2026-06-20T16:59:36.000Z (15 days ago)
- Last Synced: 2026-06-20T17:17:51.519Z (15 days ago)
- Topics: artificial-life, canvas, cellular-automata, conways-game-of-life, lenia, simulation, typescript, vite
- Language: TypeScript
- Size: 1 MB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# vivarium
A client-side **artificial-life sandbox** — a gallery of deterministic
cellular-automata and life systems, each a self-contained module with a seedable
initial state and a deterministic step function. Pure TypeScript (strict) + Vite,
no backend, no GPU, no runtime dependencies. Light runtime, heavy build.
**[▶ Live demo](https://vivarium.nuez.no)**
## Systems
Eight distinct systems, each with deep parameter control, presets, seeding, and
canvas painting:
| System | What it is |
| --- | --- |
| **Conway's Life** | The original `B3/S23` automaton, with a hand-verified pattern library (glider, Gosper gun, spaceships, pulsar, pentadecathlon, acorn). |
| **Life-like Rules** | Any two-state `B/S` rule, live-editable: HighLife, Day & Night, Seeds, Replicator, Maze, Coral, Anneal, Diamoeba… |
| **Generations** | Multi-state life with refractory "dying" states (`S/B/C`): Brian's Brain, Star Wars, Frogs, Bombers. |
| **Cyclic CA** | States on a cycle; cells advance when enough neighbours hold the successor state → spiral waves and "demons". |
| **Elementary 1D** | All 256 Wolfram one-dimensional rules; history scrolls upward. Rule 30, 90, 110, 184… |
| **Langton's Ant** | Agent-based turmites with `L/R/U/N` turn-string rules and multiple ants. |
| **Lenia** | Continuous CA — a smooth `[0,1]` field convolved with a radial bump kernel and a bell-shaped growth function. |
| **Particle Life** | Thousands of coloured particles driven by a per-species attraction/repulsion matrix; emergent cells and predators. |
## Architecture
Everything talks to two interfaces in [`src/core/types.ts`](src/core/types.ts):
- **`SystemDef`** — static metadata, a declarative parameter schema, presets, and
a `create(params, seed, preset)` factory.
- **`Simulation`** — mutable state with a deterministic `step()`, a `render()`
returning one of three render models (`cells` / `field` / `particles`), a
`hash()` of the full state, and optional `paint()` / `clear()`.
```
src/
core/ types (the contract), mulberry32 PRNG, FNV-1a state hash, registry
systems/ one self-contained module per system (+ a shared Life-like engine)
render/ canvas-2D renderer: typed arrays → ImageData, nearest-neighbour blit
ui/ framework-free gallery, auto-generated controls, transport, painting
```
**Determinism** is enforced: all randomness flows from a seeded `mulberry32`
PRNG, never `Math.random`. The same seed + step count always produces the same
state hash — pinned per system and at the registry level in the test suite.
**Cheap rendering**: grid/field systems paint into an offscreen `ImageData` at
native cell resolution and blit it scaled with nearest-neighbour; particle
systems draw squares grouped by species. No per-frame allocation. The whole app
is ~46 kB of JS (~15 kB gzipped).
## Develop
```bash
pnpm install
pnpm dev # Vite dev server
pnpm typecheck # tsc --noEmit (strict)
pnpm test # Vitest: known-outcome + determinism tests
pnpm build # static bundle into dist/
pnpm screens # build + Playwright screenshots into screens/
```
## Tests
Each system pins a **known outcome** (a glider returns to itself shifted by
`(1,1)` after 4 steps; Rule 30 matches its verified triangle; Langton's ant draws
a 2×2 block and returns home after 4 steps; Brian's Brain's isolated cell dies in
2; Lenia's empty field stays empty; particles stay in bounds) plus **determinism
snapshots** (seed + N steps → grid hash). 129 tests in total.
## Controls
Pick a system from the gallery, then play/pause/step, adjust speed, randomize the
seed, choose a preset, edit parameters live, and click-drag on the canvas to
paint cells (or particles). Keyboard: `space` play/pause, `s` step, `r` reset,
`n` randomize, `c` clear.
## License
[MIT](LICENSE).