Project

Parallax

An AAA-scope open-world game, built by AI agents, that installs, launches, and runs entirely on the web platform.

What this is

Parallax is a research project with a playable product. It builds the hardest kind of game the web could host — a high-fidelity, streaming open world with a multi-gigabyte local install, on-device AI NPCs, and native-title launch behavior — using only what the newest Chrome ships: WebGPU, WebAssembly (threads and memory64), the Origin Private File System, SharedArrayBuffer workers, and the built-in AI Prompt API.

The point is the question underneath: can the web platform host an AAA-class game — and where exactly does it fall short? Every browser bug, spec gap, and performance cliff found along the way is documented with reproductions and measurements, as input to making the platform better.

The deliverables

The findings

An evidence-backed catalog of where Chrome, WebGPU, V8/wasm, storage, and the built-in AI APIs limit native-class games — with reproductions, not opinions.

The harness

Automated build → deploy → launch → measure → diff infrastructure for web games. Think “WebPageTest for games.”

The game

A visually striking open-world slice — two districts, seamless streaming, conversational NPCs — installable and playable offline in the browser.

Status

Current milestone: M0 — harness + skeleton

Early days: the measurement infrastructure and platform spikes come first, the world after. Everything — plans, decisions, findings — is developed in the open. Chrome (latest release or newer) only, by design.

Follow along

View on GitHub github.com/pmeenan/parallax