How it works

From a sentence to a shippable app in four steps

No separate build pipeline to learn. You chat, it builds, you see the cost and the security verdict before you ever commit to keeping it.

01

Describe what you want

Type a plain-English request — "a habit tracker with streaks" — or start from one of 15 free templates for an instant, zero-cost v1. Either way, you land in a chat interface next to a live preview pane, not a code editor. For most requests you get one self-contained app in seconds; for something bigger, start a multi-file React project instead and the same loop below still applies.

02

See the cost before it runs

A live estimate appears in the composer before you send. Generation is streamed from Claude and produces one complete, self-contained HTML app — styling and logic included, no partial files. If the generation fails, its cost is tracked separately and never billed to your build spend; if you report the same bug twice, the AI stops repeating itself and re-diagnoses from scratch instead.

03

Review the diff and the security scan

Every change lands as a new version with a line-by-line diff against the last one — roll back with one click if you don't like it. In parallel, the build is checked for hardcoded API keys and secrets, eval() / dynamic code execution, and other common AI-coding security mistakes, with the verdict shown as a pass/fail badge on the Security tab.

04

Download it, or ship it

Download the app as a single portable .html file you can run anywhere — no lock-in, the code is always fully visible. Or publish it to a shareable link on vibesafebuilder.com; publishing re-runs the security scan server-side and refuses to go live if a critical issue is found, so a bypassed client-side check can never leak onto the public internet. A multi-file project has two more paths — see below.

When one file isn't enough

Some ideas need more than one HTML page — separate components, a shared backend, a real repository. Click ⚛ New React project instead of building a single page, and the loop above still applies, but the AI now works across a real, multi-file codebase — components, hooks, and styles in separate files, assembled with Vite. Ask for a change and only the files that need it get rewritten. If your app needs data shared between people or across devices, the AI can reach for a built-in backend automatically, no setup required.

Two more ways to ship a project like this, on top of download and publish:

Push to GitHub

Connect your GitHub account and push the project into a repository you own, as a real commit — ordinary code, not something trapped in this builder. Pull changes back in later, too.

Deploy to your own Vercel account

Connect a Vercel token scoped to one project, and deploy — Vercel builds your project from real source, the same as if you'd pushed it yourself. Making it public runs a dedicated Publish Security Check first: it scans every file for live credentials and refuses, with no override, if it finds one.

Full detail on both in the user guide.

Two ways to pay for it

Bring your own Anthropic API key and pay Anthropic directly at their standard rates — we add no markup and never see your key or your invoice. Or take the $15/month Managed plan and we cover up to $10/month of usage on our own key, no API key required. See full details on the pricing page and the user guide.