Skip to content

Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.

1.6 KB
Raw Blame

Site snapshots

A daily visual record of how every page on three.ws looks, so we can always see what a page looked like before a drastic redesign.

Capture a snapshot

node scripts/page-snapshot.mjs            # all public pages, desktop + mobile, against https://three.ws
node scripts/page-snapshot.mjs --desktop-only
node scripts/page-snapshot.mjs / /ibm /pay   # just a few routes
BASE_URL=http://localhost:3000 node scripts/page-snapshot.mjs   # snapshot a local build

or npm run snapshot.

Pages come from data/pages.json — the same manifest that drives /sitemap, llms.txt, and the page audit. Add a page there and it's snapshotted automatically. Auth-gated pages are skipped unless named explicitly.

How it's stored (and why)

Screenshots are written to stable paths under current/ and overwritten on every run:

current/
  index.html        ← open this: a browsable gallery of the latest set
  manifest.json     ← machine-readable index, stamped with the capture date
  desktop/<slug>.jpg
  mobile/<slug>.jpg

The archive is git history. Committing snapshots/ each day captures that day; the working tree stays at ~one day's size instead of growing forever.

  • Browse today: open current/index.html
  • A page's history: git log --follow -- snapshots/current/desktop/home.jpg
  • Recover a past day: scripts/snapshot-export-day.sh 2026-06-05 (writes exported/<date>/ — gitignored, just for viewing)

Daily run

The capture script is wired and tested. The daily trigger (GitHub Action cron or a scheduled agent) is added separately once the output is confirmed.