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.

3.6 KB
Raw Blame

/everything: browse every feature by goal

three.ws has about 400 product pages. The nav shows roughly 30 of them to a first-time visitor on purpose (see onboarding-tier.md), and /sitemap lists all of them by section. Neither answers the question a visitor actually has: "I want to animate my avatar, where do I go?"

/everything answers it. Every product page is assigned one goal (a job), and the page is a searchable directory of those goals.

What a visitor gets

  • Twelve goals, each with a one-line promise, a "Start here" button and its pages: make a 3D model or avatar, animate and perform, build an AI agent, put it on your site, sell and earn, launch a token, trade and track markets, wallets and on-chain identity, worlds/AR/play, discover, developer tools, your account.
  • Search across title, path, description and tags. Press / anywhere on the page to focus it. ?q=wallet deep-links a search, #launch deep-links a goal.
  • Goal chips with live counts. They filter in place and update the URL hash.
  • Pins: the star on any row pins it to a "Pinned" block at the top. Recently opened tracks the last six pages opened from here. Both live in localStorage only (tws:fav, tws:recent); nothing is sent to a server.
  • New this month: pages whose added date in data/pages.json is within 45 days.
  • Designed states: skeleton while loading, a retry panel (with a /sitemap fallback) if features.json fails, an empty-search panel that suggests how to rephrase.

Where the data comes from

data/pages.json ──┐
                  ├─ scripts/build-page-index.mjs ──> public/features.json
scripts/lib/feature-jobs.mjs ─┘                       (pages[].job, jobs[])
                                                          │
                      pages/everything.html + src/everything.js (the page)
                      public/search.js (⌘K labels each result with its goal)

scripts/lib/feature-jobs.mjs is the single classification. A page gets its job from, in order: an exact-path pin in OVERRIDES, the first matching regex in RULES, then its pages.json section default. A new page therefore shows up on /everything with no extra step; if the rules put it in the wrong goal, pin it in OVERRIDES. Each goal also lists featured flagship paths that lead its card.

tests/feature-jobs.test.js fails if a product page resolves to no goal, a goal is empty, an override or featured path does not exist, or a core journey (/forge, /create-agent, /widgets, ...) lands under the wrong goal.

Other discovery surfaces added with it

  • Nav: "Browse by goal" leads Discover > Start here, and "Everything" is a top-level link. Both are in the lite tier (the lite menu stays at or under 30 links; /pocket moved to the advanced tier to make room).
  • ⌘K palette: a "Browse by goal" quick action and suggestion, and every page result is labelled with its goal instead of its site section.
  • What next? strip (public/next-steps.js, loaded by nav.js): under a dozen tool pages (/forge, /image-to-3d, /create/*, /animations, /widgets, /marketplace, /launch, ...) a short strip suggests the natural next steps, dismissible per session. Opt a page out with <html data-next-steps="off">. tests/next-steps.test.js checks every link resolves to a real page. It complements public/feature-discovery.js, which fires its own card after a tws:feature-done event.
  • /sitemap links to /everything in its header.

/everything used to be a guessable alias that 308'd to /sitemap; that alias was removed from vercel.json and vite.config.js.