T-1 HTML Pages sits at /workbench/text/html, and it's the corner of x-rush I keep coming back to. It's an online HTML5 workspace: start from a template library covering 12 categories, or skip the library entirely and let the AI generate a page from a blank file. Both routes drop you into the same editor. I've used it for a client landing page, a friend's portfolio, and a pixel-art game I never finished - the workflow holds up. And the payoff is the part most people walk past: hit Publish and the page goes live on its own public URL that anyone can open with no login, ready to share, embed, or post.
This is the 2026 version of the doc. I'll cover how the template library is laid out, what the preview modal actually tells you, the fork-vs-generate decision, and how the editor's file tree, Monaco, AI panel, and live preview fit together - then what Publish actually hands you (a standalone page, not a sandbox link). Where there are sharp edges, I'll call them out - I'd rather you know now than mid-deadline.
The Template Library
Open the page and the first thing you see is Featured - three Editor's Picks cards sorted by popularity. These rotate, and the picks are honestly decent; the team curates them rather than auto-ranking by clicks. Below that is Recently Used: your last four templates, persisted in localStorage so they survive a refresh but not a different browser. Useful when you're bouncing between two projects.
Then come the 12 category tabs, each with a count badge so you can see at a glance how many templates live under "Portfolio" versus "Game". The search box top-right takes ⌘K (or Ctrl+K) to focus - type a few letters and it filters across names and tags. Next to it, a sort dropdown with four options: Most Popular, Newest, Recently Updated, Alphabetical. I default to Recently Updated; "Newest" and "Recently Updated" sound the same but aren't - one is creation date, the other is last edit. Small thing, but it matters when you're hunting for something actively maintained.
Preview Before You Commit
Clicking Preview on any card opens a fullscreen modal - not a side panel, fullscreen, which I appreciate because half-width previews lie about spacing. Four device tabs run across the top: Desktop, Tablet (768px), Mobile (375px), and Code (raw source). The pixel widths are fixed on purpose; 768 and 375 are the breakpoints most templates were designed against, so what you see is what you ship.
The right-side info panel shows usage count, a difficulty rating, tags, and a short list of related templates. I check the difficulty rating every time - a "beginner" template with 200 lines of inline CSS is going to fight you the moment you try to rebrand it. Related templates is the underrated feature here: I've found better starting points through "related" than through the category tabs more than once.
Fork a Template or Generate from Scratch
Two ways in. Use Template calls forkTemplate - it copies the template into a fresh workspace and auto-redirects you to /workbench/text/html/[newId]. Your fork is independent; the original stays untouched. This is the fast path and the one I take 80% of the time.
+ Create from Scratch is the AI route. You fill in a prompt, pick a type (landing / product / blog / personal / portfolio / game), choose a style (modern / minimal / cyber / glass), and toggle features (responsive, dark-mode, animations, contact-form, seo, icons). Then a four-step generation animation plays - mostly for show, but it does reflect real progress through prompt-parse -> scaffold -> fill -> finalize. When it finishes, Open Editor takes you in. Generation runs on x-rush's smart routing, which taps the industry's top-tier text models - Claude Fable 5, GPT-5.6, Gemini 3.5, Qwen3.7 Max and the rest - and routes each request to the one that best fits the task. The pool tracks the landscape and updates as new models land, so you're not stuck on last year's release. For HTML it's tuned to treat the type/style/feature flags as hard constraints, not vibes.
Honest take: AI-from-scratch is great for a first draft on something unusual (a game landing, a weird portfolio), but for a standard SaaS landing page the template library gets you further faster. Generate when the library doesn't have what you need.
Inside the Editor
The editor runs in what the team calls "clean mode" - no left sidebar, no chrome, just the work. Four zones:
- Left, file tree. New file, upload assets, right-click for rename/delete/copy/download, and drag to reorder. Reordering matters because the live preview loads files in tree order - put your
styles.css above script.js and it just works.
- Center, Monaco. Same editor VS Code uses, so your muscle memory transfers. Syntax highlighting, multi-cursor, find-and-replace, the works.
- Right, AI assistant panel. Collapsible - click the handle to shove it off-screen when you need width for a long file.
- Bottom, live preview iframe. Refreshes on save. Controls for fullscreen, popout (opens the preview in a real browser window - great for DevTools), device switch, and a console error count badge. If that badge reads anything above zero, fix it before you keep going - a single uncaught error can quietly break downstream scripts.
The AI Assistant
The AI panel isn't a generic chatbot bolted on. When you ask it something, it reads your current file's contents and the file tree structure first, then answers in that context. "Make the hero section sticky" gets you a real diff against the file you have open, not a generic tutorial.
Code changes come back as a Diff view - additions and deletions side by side. You Apply to write it in, or Reject to discard. There's no partial-apply, which is a deliberate tradeoff: it keeps history clean. While the AI is thinking you get a three-dot cycling animation; responses usually land in 1–3 seconds for small edits, longer for "rewrite this whole section" asks. It's the same smart routing behind generation - Claude Fable 5, GPT-5.6, Gemini 3.5, Qwen3.7 Max - so the assistant is reading your actual file with a top-tier model, not phoning it in.
One thing I rely on: undo always gets you back to the original template. Every fork starts with an undoSnapshot, so no matter how badly an AI edit goes sideways, one undo and you're at the clean fork state. This is why I apply AI diffs aggressively and undo just as aggressively - the cost of a bad apply is one keystroke.
Save, Publish, Version History
Saving is two-mode. Manual is Ctrl/Cmd+S. Auto is useAutoSave, which writes on a debounce. When there are unsaved changes, a dot lights up next to the workspace name - that's your "dirty" indicator. There's also a beforeunload guard so closing the tab with unsaved work throws the browser's "are you sure" prompt. Don't ignore it; I have, and I lost fifteen minutes.
Publish is where T-1 stops being a sandbox and turns into a real publishing tool - and honestly this is the feature I'd put on the billboard. Hit Publish and you get a public URL at /p/{id}. The thing to get straight: the page is completely standalone. It lives at its own URL, loads on its own, and anyone can open it with zero login - no x-rush account, no "sign in to view," no gate of any kind. Send the link in a chat, drop it in an email, paste it into a WeChat moment (朋友圈); the person on the other end just sees a working web page. They never have to know x-rush exists.
That /p/{id} URL is backed by a share/[id] page that renders the published snapshot, so it's a real hosted page - not a preview iframe tucked behind your workspace. Two upsides I lean on constantly:
- Take a page down and re-publish, and the URL doesn't change. I use this for client landing pages all the time: pull it offline while I rework the copy, push it back live when it's ready, and the link I already sent still resolves. No awkward "sorry, here's the new URL" email.
- Embed it anywhere. Because it's a standalone page on its own URL, you can drop it into an
<iframe> on another site, link it from a newsletter, or share it on social. It behaves like any hosted HTML page - because that's exactly what it is.
Version history shows publish snapshots only (not every save), and keeps the most recent 10. You can preview any snapshot or restore it. Restoring doesn't delete the current state; it creates a new snapshot - so you can A/B between two published versions without losing either.
Two more things worth knowing. The ShareModal lets you push a page to the community under four categories: landing, portfolio, auto, chat. And the Explore page lets other users fork your published work - when they do, they see a ForkedBanner showing the origin workspace, so credit flows upstream. I've had strangers fork a portfolio template, and the banner linking back is genuinely how I found my next client.
Pro Tips and Sharp Edges
- Undo is your safety net.
undoSnapshot means you can always return to the clean fork. Apply AI diffs freely; undo is one keystroke away.
- A published page is a standalone page, not a preview link. Anyone opens
/p/{id} with no login - share it, embed it in an iframe, post it to social. Pull it down and re-publish and the URL holds steady.
- Switching templates with unsaved work triggers a
ConfirmSwitchModal with three choices - save, discard, or cancel. Read it before you click. The default is not "discard", which has saved me more than once.
- Explore lets you fork anyone's published work. Same fork rules as templates - good starting points, no setup.
- Version history is publish-snapshots only, most recent 10. If you need to keep more, publish intentionally rather than relying on autosave.
- Mobile (under 768px) shows a
MobileEditorGuard telling you to switch to desktop. The editor genuinely doesn't fit on a phone - code editing on a 375px screen is a bad time. The guard is correct; listen to it.
- The console error badge in the preview is not decorative. Zero errors is the bar. If it's red, your page is broken in a way the visual preview might not show.
That's T-1. It's not the flashiest module, but it's the one I'd hand someone on day one - the template library gets you shipping in minutes, the AI assistant plus clean editor scale up to real client work, and Publish hands you a standalone page on its own URL that you can share or embed without dragging anyone into a sign-up flow. No different tool required.