openzot/arcade

A software factory that makes browser games: zot designs, builds and playtests a new one every shift, then ships it unreviewed.

6

stars

121

commits

JavaScript

primary language

Sep 6, 2026

updated

openzot.github.io/arcade/
agentic-ai
ai-agent
autonomous-agents
browser-games
canvas
game-development
generative-ai
github-actions
github-pages
html5-games
llm
openrouter
software-factory
vanilla-javascript
zot

README

zot arcade

A software factory that makes browser games.

https://openzot.github.io/arcade/

Some notes

  • Nobody reviews the games before they ship. They are model output, published as-is; expect the occasional dud.
  • The standing order is orders/new-game.yaml - 36 lines, the whole specification of what comes off the line.
  • AGENTS.md is what zot reads before every shift.
  • Setup, workflows, layout and tuning are in OPERATING.md. The short version: fork this repository, add an OPENROUTER_API_KEY secret, and run the shift workflow once.
  • Every shift's session is published as a row in the openzot/arcade dataset on Hugging Face: the full conversation that made each game, in chat shape (system / user / assistant with tool calls), plus the screenshots the model saw, the game's files, and how the run ended. If you want real, end-to-end agent trajectories for fine-tuning, training or evaluation, that is the place; how rows are produced is in OPERATING.md.
  • Why the output shape is fixed - and why that is what makes this a factory - is in PHILOSOPHY.md.

Contributors

openzot/arcade

A software factory that makes browser games: zot designs, builds and playtests a new one every shift, then ships it unreviewed.

6

stars

121

commits

JavaScript

primary language

Sep 6, 2026

updated

openzot.github.io/arcade/
agentic-ai
ai-agent
autonomous-agents
browser-games
canvas
game-development
generative-ai
github-actions
github-pages
html5-games
llm
openrouter
software-factory
vanilla-javascript
zot

README

zot arcade

A software factory that makes browser games.

https://openzot.github.io/arcade/

Some notes

  • Nobody reviews the games before they ship. They are model output, published as-is; expect the occasional dud.
  • The standing order is orders/new-game.yaml - 36 lines, the whole specification of what comes off the line.
  • AGENTS.md is what zot reads before every shift.
  • Setup, workflows, layout and tuning are in OPERATING.md. The short version: fork this repository, add an OPENROUTER_API_KEY secret, and run the shift workflow once.
  • Every shift's session is published as a row in the openzot/arcade dataset on Hugging Face: the full conversation that made each game, in chat shape (system / user / assistant with tool calls), plus the screenshots the model saw, the game's files, and how the run ended. If you want real, end-to-end agent trajectories for fine-tuning, training or evaluation, that is the place; how rows are produced is in OPERATING.md.
  • Why the output shape is fixed - and why that is what makes this a factory - is in PHILOSOPHY.md.

Contributors

Languages

JavaScript

77.9%

CSS

13.5%

HTML

8.4%