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.
JavaScript 3d-generationai-agentsanimationavatardigital-humangenerative-aiglbgltfimage-to-3dllmmcpmcp-servermodel-context-protocolsolanatext-to-3dthreejsweb-componentswebglwebxrx402
// curriculum.js — load the generated tour curriculum and own the cross-page
// tour state. The curriculum (what we visit + say) is built at deploy time by
// scripts/build-tour.mjs and served from /tour/curriculum.json. The live state
// (active? which stop? paused? muted?) lives in sessionStorage so the tour
// survives a full-page navigation between stops on different routes.
const CURRICULUM_URL = '/tour/curriculum.json';
const STATE_KEY = 'tws:tour:state'; // live, per-tab tour (sessionStorage)
const RESUME_KEY = 'tws:tour:resume'; // durable cross-session memory (localStorage)
// The generic anchors TourDirector._resolveTarget() falls back to when a stop
// has no hand-authored `targets` (only about a fifth of the curriculum does, see
// scripts/build-tour.mjs → TARGETS). It lives here rather than in director.js so
// the offline tools can read it without importing the whole 3D engine:
// scripts/capture-tour-atlas.mjs resolves anchors exactly the way the live tour
// does, and its report would be fiction if the two lists ever drifted.
export const TOUR_FALLBACK_SELECTORS = [
'[data-tour-target]',
'main h1, .hero h1, h1',
'a.cta, .btn-primary, button[type="submit"], main a.button, .hero a',
];
// The content root a tour anchor is preferentially resolved inside. See
// _resolveTarget() for why an unscoped lookup keeps landing on the site nav.
export const TOUR_CONTENT_ROOT_SELECTOR = 'main, [role="main"], article';
let _cache = null;
export async function loadCurriculum() {
if (_cache) return _cache;
const res = await fetch(CURRICULUM_URL, { cache: 'force-cache' });
if (!res.ok) throw new Error(`tour curriculum ${res.status}`);
const data = await res.json();
if (!data || !Array.isArray(data.stops) || !data.stops.length) {
throw new Error('tour curriculum empty');
}
_cache = data;
return data;
}
// Normalize a pathname to match curriculum stop paths ("/" stays "/", others
// lose any trailing slash). Query and hash are irrelevant to which stop we're on.
export function normalizePath(pathname = location.pathname) {
// Strip query/hash first (a caller may hand us a full href) then trailing
// slashes, so only the route path decides which stop we're on — as the
// surrounding comments promise.
const clean = String(pathname).split(/[?#]/)[0];
const p = clean.replace(/\/+$/, '');
return p === '' ? '/' : p;
}
const DEFAULT_STATE = {
active: false,
index: 0,
track: 'full',
paused: false,
muted: false,
voice: 'nova',
speed: 1,
};
export function readState() {
try {
const raw = sessionStorage.getItem(STATE_KEY);
if (!raw) return { ...DEFAULT_STATE };
return { ...DEFAULT_STATE, ...JSON.parse(raw) };
} catch {
return { ...DEFAULT_STATE };
}
}
export function writeState(patch) {
const next = { ...readState(), ...patch };
try {
sessionStorage.setItem(STATE_KEY, JSON.stringify(next));
} catch {
/* private mode / disabled storage — tour still runs within this page */
}
// Mirror durable preferences + progress so a returning visitor (new tab,
// next day) can pick the tour back up. The live sequencing still reads from
// sessionStorage; this is the cross-session memory only.
if (next.active) {
writeResume({ index: next.index, track: next.track, voice: next.voice, speed: next.speed });
}
return next;
}
export function clearState() {
try {
sessionStorage.removeItem(STATE_KEY);
} catch {
/* ignore */
}
}
// ── Cross-session memory (localStorage) ──────────────────────────────────────
// Remembers where a visitor got to and their voice/speed preferences across
// sessions, plus whether they've ever finished the tour. Never drives live
// sequencing — only the /tour page's "resume" / "completed" affordances and the
// director's preference defaults read it.
const DEFAULT_RESUME = { index: 0, track: 'full', voice: 'nova', speed: 1, completed: false };
export function readResume() {
try {
const raw = localStorage.getItem(RESUME_KEY);
if (!raw) return { ...DEFAULT_RESUME };
return { ...DEFAULT_RESUME, ...JSON.parse(raw) };
} catch {
return { ...DEFAULT_RESUME };
}
}
export function writeResume(patch) {
const next = { ...readResume(), ...patch };
try {
localStorage.setItem(RESUME_KEY, JSON.stringify(next));
} catch {
/* storage unavailable — cross-session resume simply won't persist */
}
return next;
}
export function markCompleted() {
writeResume({ completed: true, index: 0 });
}
// ── Playlists ────────────────────────────────────────────────────────────────
// A track is a view over the curriculum: an ordered list of absolute stop
// indices to actually visit. 'full' is every stop in the general site tour
// (section !== 'onboarding' — see below); 'quick' is 'full''s highlighted
// heroes; 'onboarding' is the short, hand-authored new-account walkthrough
// (section === 'onboarding', built by scripts/build-tour.mjs's
// ONBOARDING_STOPS) — its own stops are excluded from 'full'/'quick' so the
// two curricula never bleed into each other on the same visit. Pure (no
// storage) so the sequencing logic and tests can build it freely. Always
// non-empty for a curriculum that actually has matching stops — an unknown
// track or a curriculum with no highlights falls back to the full
// non-onboarding list so the tour can never strand itself with nothing to play.
export function buildPlaylist(curriculum, track = 'full') {
const all = curriculum.stops.map((_, i) => i);
const isOnboardingStop = (i) => curriculum.stops[i].section === 'onboarding';
if (track === 'onboarding') {
const onboarding = all.filter(isOnboardingStop);
return onboarding.length ? onboarding : all;
}
const general = all.filter((i) => !isOnboardingStop(i));
const base = general.length ? general : all;
if (track !== 'quick') return base;
const quick = base.filter((i) => curriculum.stops[i].highlight);
return quick.length ? quick : base;
}
export function trackMeta(curriculum, track = 'full') {
return (curriculum.tracks || []).find((t) => t.id === track) || null;
}
// Index of the first stop on a given path, or -1. Used to snap the tour back
// onto the route when a visitor navigates by hand. When `playlist` (an array
// of absolute stop indices, as returned by buildPlaylist) is given, only
// those stops are considered — otherwise the onboarding curriculum and the
// general site tour can share a path (e.g. /markets, /profile) and a
// mid-track hand-navigation could snap onto the wrong track's stop.
export function stopIndexForPath(curriculum, pathname = location.pathname, playlist = null) {
const target = normalizePath(pathname);
if (playlist) {
return playlist.find((i) => normalizePath(curriculum.stops[i]?.path) === target) ?? -1;
}
return curriculum.stops.findIndex((s) => normalizePath(s.path) === target);
}
export function sectionTitle(curriculum, id) {
return (curriculum.sections || []).find((s) => s.id === id)?.title || '';
}