GitHub radar
AI Agent–Built 3D Cities: Walk SF and Kyoto in Browser
Autonomous Claude Fable 5.1 agent swarms reconstructed two real city districts from open data and built them as browser-native 3D apps you can walk through.
PhiloLabs dispatched autonomous Claude Fable 5.1 agent swarms to reconstruct real city districts entirely from open geographic data and public reference imagery. The pipeline runs in four stages: parallel research agents gather OpenStreetMap geometry, USGS elevation data, transit specs, and a per-fact storefront census; offline scripts generate optimised 3D asset kits; a pure Three.js runtime assembles terrain, streets, crowds and traffic from JSON specs; and Playwright-driven QA agents screenshot fixed viewpoints and diff them against free-licensed photographs to catch errors. Two worlds are finished: Union Square, San Francisco (453 OSM buildings, 75 hand-authored façades, 129 named storefronts, 220 pedestrians on a 1,398-node nav graph, Powell Street cable cars, and two explorable interiors: Apple Union Square and Nintendo SAN FRANCISCO) and Higashiyama, Kyoto (266 buildings, 2.3 km continuously walkable route climbing 76 m from Gion to Kiyomizu-dera, rendered as a hand-painted anime background). No game engine, no proprietary 3D tiles — every element is generated from code.
Why a vibe-coder should care
Shows concretely what a Claude Fable 5.1 agent swarm can build end-to-end — from raw geographic data to a walkable 3D city ready to run in any browser. The full pipeline code lives in the repo, so you can study it or adapt the approach for your own locations. Both worlds launch in any modern browser with a single npm command.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Deploy one of the worlds from https://github.com/PhiloLabs/fable51-worlds — follow the README, run union-square-sf with npm install && npm run dev, open it in a browser, and show me a screenshot.
Runs on a regular laptop — only Node.js needed.
Open on GitHub▌ More finds