Polymarket blasted it; SpaceXAI shipped the page. On July 28, 2026, Build Mode went Early Beta: tell Grok an idea, watch a working preview appear in chat, refine in plain English, then publish to grok.me or a custom domain.
Gate: SuperGrok Heavy on web, iOS, and Android.
TL;DR
| Item | Detail |
|---|---|
| Product | Grok Build Mode |
| Who | SuperGrok Heavy (Early Beta) |
| Where | grok.com · iOS · Android (mode switcher → Build) |
| Loop | Prompt → live preview → NL edits → publish |
| Share | *.grok.me or your domain |
| Not the same as | Open-source Grok Build CLI harness |
What “vibe coding” means here
Unlike “paste this React,” Build Mode keeps the artifact running in the conversation. SpaceXAI’s examples (forest driver, city sim, physics toys, beat machines) are meant to prove interactive state — not static HTML dumps.
Categories they call out:
- Websites / portfolios
- Apps with real logic (planners, calculators, trackers)
- Games (arcade → 3D)
- Dashboards with connectors for live data
How to try it (Early Beta)
- Confirm you have SuperGrok Heavy — Build Mode is gated to that tier during Early Beta.
- Open grok.com or the Grok iOS/Android app.
- Open the mode switcher and choose Build.
- Describe the product in one clear message (audience, primary action, visual tone).
- Interact with the live preview in chat — click through, break the happy path, note what feels wrong.
- Iterate in natural language: “make the CTA sticky,” “add dark mode,” “store todos in localStorage.”
- When it works, publish to a
*.grok.melink or attach a domain you own, then share with teammates or the public.
SpaceXAI’s official line is blunt: nothing to install, nothing to configure, and you never have to touch the code unless you want to. That is the product bet — chat is the IDE.
Official examples worth studying
The launch page walks through real prompts and hosted results. One published example:
Build a peaceful 3D driving game: an endless forest I can drive through freely, with no missions and no timers. Make it feel calm.
That became driver.grok.me — an endless forest drive, not a marketing landing page. Other demos SpaceXAI groups under “Made with Build Mode” include a city sim, physics toys, and a beat machine. The point of the lineup is stateful interaction: cameras, timers, audio, simulation loops — the failure modes of static HTML generators.
When you write your own first prompt, borrow that specificity:
- Constraint clarity (“no missions, no timers”) beats vague “make it fun.”
- Feel words (“peaceful,” “calm”) steer art direction when you refuse to open the CSS.
- One primary verb (drive, plan, track, chart) keeps the first build from sprawling.
Publish paths: grok.me vs custom domain
| Path | Best for | Trade-offs |
|---|---|---|
*.grok.me | Instant share, demos, internal pilots | Brand sits under SpaceXAI’s subdomain; fine for prototypes |
| Custom domain | Customer-facing sites, client work, portfolio | You own DNS/brand; still hosted in Build Mode’s publish flow |
Treat Early Beta hosting like any managed preview: confirm what happens on unpublish, whether builds are versioned, and whether secrets in client code (API keys in the browser) are your problem — because they are. If you ever paste credentials into a build surface, read explainx.ai’s Grok Build secrets / upload security notes.
Prompt patterns that survive Early Beta
Landing page
Build a one-page site for a B2B MCP analytics tool called explainx.ai. Hero: product name large, one sentence value prop, primary CTA “Browse MCP servers,” secondary CTA “Read the blog.” Dark, technical, no purple gradients. Include a short feature strip (connectors, loops, skills) and a footer with docs links.
Internal utility
Build a team on-call rota app: week view, assignee dropdown, Slack webhook field (client-side only for now), export CSV. Persist rota in localStorage. Mobile-friendly.
Dashboard with connectors
Build a live dashboard that pulls the metrics I connect and shows filterable charts for the last 7 / 30 days. Default empty state should explain how to attach a connector.
Game / toy
Build a browser puzzle: 5×5 lights-out, undo stack, shareable seed in the URL, keyboard and touch controls.
Bad prompts (“make an app”) produce generic shells. Good prompts name entities, persistence, and the first user action.
Build Mode vs Lovable / v0 / Bolt / Cursor
| Tool | Loop | Hosting | Best when |
|---|---|---|---|
| Grok Build Mode | Chat + live preview + publish | grok.me / custom domain | Heavy subscriber; want shareable artifact without a repo |
| Lovable / v0 / Bolt | Prompt → UI codebase | Their hosts / export | Design-system-ish frontends, component libraries |
| Cursor / Claude Code | Agent in a real repo | Your deploy | Monorepos, tests, PR review, long-lived products |
Build Mode competes with prompt-to-UI + host, not with open vs closed agent harnesses on a production monorepo. If your job is “ship a demo by Friday,” Build Mode is in the race. If your job is “merge without breaking billing,” stay in an IDE agent — Cursor Start India is the price story for that path; What is Cursor? is the onboarding path.
Limits of Early Beta (honest checklist)
- Quality variance — Early X reactions already split between “MVP in minutes” and “tried it, nonsense.” Run your own prompt suite before promising clients.
- Code ownership — Opening the code is optional; exporting a durable git workflow may still be weaker than IDE agents. Confirm export before you bet a product on it.
- Connectors — Dashboards with live data only work as well as the connector surface; treat “pull business data” as beta until you verify auth scopes.
- Compliance — Publishing customer PII into a chat-built app without a review is still a compliance failure, even if the UI looks polished.
- Mobile parity — Officially on iOS/Android; still verify touch targets and keyboard behavior on your build, not just desktop.
Where this sits in the SpaceXAI stack
Build Mode is the consumer chat surface. Grok Build open-sourced is the developer harness. Grok 4.5 + Cursor and the SpaceX–Cursor acquisition story are the IDE / company layer. Same family; different jobs. Mixing the names in a pitch deck is how you confuse buyers.
Build Mode vs Grok Build (don’t mix the brands)
| Build Mode | Grok Build | |
|---|---|---|
| Surface | Chat product | Terminal agent / TUI |
| Audience | Heavy subscribers, creators | Developers |
| Output | Hosted preview + link/domain | Repo edits, Excel, etc. |
| Source | Closed product feature | Open-sourced |
Security note if you ever paste secrets into any Grok build surface: see repository upload / secrets risks.
Who this threatens (and who it doesn’t)
Build Mode competes with Lovable / v0 / Bolt-class prompt-to-UI tools, not with Claude Code on a monorepo. For serious agent harnesses, compare open vs closed harnesses and Cursor Start India for IDE agents.
Early X reactions already split: “MVP in minutes” vs “tried it, bullshit.” Treat Early Beta quality as uneven until your own prompt suite passes.
A practical evaluation suite (30 minutes)
Before you standardize on Build Mode for client demos, run five fixed prompts and score each 1–5 on: first-preview quality, edit fidelity (“do what I said”), mobile layout, publish reliability, and reload persistence.
- Marketing one-pager with two CTAs
- CRUD list with localStorage
- Simple arcade game with score
- Form with client-side validation
- Chart dashboard with fake data (then connectors if available)
Anything under ~18/25 total means you still need a human designer or an IDE agent for that class of work. Score in a shared sheet so the team argues from the same artifacts — not from one viral clip.
Pricing and access reality
During Early Beta, SuperGrok Heavy is the gate. Pricing tiers move; confirm the entitlement in your Grok account rather than quoting a third-party screenshot. If your org only has standard SuperGrok, you may be waiting on a wider rollout — plan demos accordingly.
Also separate subscription cost from opportunity cost: a Heavy seat that produces three shareable prototypes a week can be cheaper than a designer sprint; a Heavy seat that produces broken games nobody can maintain is expensive theater.
Related on explainx.ai
- Grok Build open-sourced
- Grok Build secrets / upload security
- Grok 4.5 + Cursor
- SpaceX acquires Cursor
- Top open & closed agent harnesses
- What is Cursor?
Official
Bottom line for builders
If you need a shareable interactive demo tonight, try Build Mode on Heavy. If you need a maintained codebase with PRs, stay in Grok Build or an IDE agent. The July 28 launch is SpaceXAI saying vibe coding belongs in the same chat as Grok — not only in a separate SaaS. Follow @explainx_ai for how the Early Beta holds up against Lovable-class tools week to week. When Build Mode ships connector-backed dashboards that survive a real auth review, that is the moment it stops being a toy and starts competing with hosted app builders on serious internal tools.
Availability and pricing tiers change — confirm SuperGrok Heavy entitlements in your Grok account as of July 29, 2026.