Test an app built with AI: one guide per tool
This hub is for anyone who built a web app with an AI tool and wants to know whether it is ready to show, share or sell. The builders differ in how they create and host the app, but the failures are alike: it works in the builder's preview and not on the live address, sign-in breaks on a new domain, data does not persist, one user sees another user's records, or a page is blank on a phone. Choose your tool below for a guide with a pre-launch plan, then use the common checklist and the independent test.
The tools, and where the app usually lives
Where an app ends up depends on the tool and on how you publish it. The facts below come from each tool's own site or documentation.
Lovable
A platform for building and deploying web apps from natural-language descriptions, with a built-in backend (Cloud) for database and authentication. Published apps get a lovable.app address; paid plans can add a custom domain. The published app is a snapshot of the last publish. Guides: test a Lovable app and Lovable app not working.
Bolt
An AI builder for websites, apps and prototypes at bolt.new, with Bolt Cloud for hosting, database, authentication and custom domains. Published projects get a free .bolt.host address by default; projects can also be published to Netlify. Guide: test a Bolt app.
Replit
Build and publish apps from Replit. A published app runs at a .replit.app address, separate from the Preview, with deployment types such as Autoscale and Static. Replit documents that the published filesystem is not persistent, so data belongs in a database or storage service. Guide: test a Replit app.
Hercules
A platform for building apps and websites by chatting with AI, with built-in backend, database and authentication. Each app gets one free subdomain in the form [myapp].onhercules.app. Hercules has its own browser tests, which can run on publish. Guide: test a Hercules app.
Base44
Described in its documentation as a no-code AI platform for building full-stack apps, websites and AI agents, with managed backend, authentication, integrations and hosting built in. Its docs also cover testing with a separate test database. Guide: test a Base44 app.
v0
Deploys projects to Vercel. Each Vercel project gets a stable production address on a vercel.app domain, and preview deployments have their own URLs. Guide: test a v0 app.
Claude Code
An agentic coding tool that reads your codebase, edits files and runs commands, available in the terminal, IDE, desktop app and browser. It writes code in your repository; where the app is hosted is your choice. Guide: test a Claude Code app.
Claude Cowork
Brings the agentic abilities of Claude Code to knowledge work without a terminal: you describe an outcome and Claude plans and carries out multi-step tasks. Hosting is again whatever you publish to. Guide: test a Claude Cowork app.
Cursor
An AI coding agent that turns ideas into code and can run tasks autonomously and in parallel. Like Claude Code, it produces code you host yourself. Guide: test a Cursor app.
Windsurf
An AI IDE; its documentation now lives under the name Devin Desktop and mentions one-click App Deploys. No guide of its own in this series: the checklist below and the vibe-coded app guide apply.
Pre-launch checklist for any AI-built app
- Test the published address, in a private window, not the builder's preview.
- Confirm what is live equals what you last edited (many builders require a manual publish or update).
- Load every page; look for blank screens, broken images, console errors.
- Check a phone-width screen.
- Sign up with a new address, confirm emails arrive and links open the live app.
- Sign in, out and in again; try a wrong password.
- Check redirects after sign-in on every address you use, including a custom domain.
- Use two accounts and make sure one cannot read the other's data.
- Create, edit, delete a record, reload, and check it persisted.
- Open private pages signed out; you should be turned away.
- Submit forms with empty, long and unusual input.
- Check secrets: no key visible in the page or its network calls.
- Check payments in test mode before taking real money.
- Check the custom domain: secure connection, bare and
wwwversions. - Check legal pages and placeholder text.
- Re-test after every fix, because fixes by an AI can disturb other parts.
Independent, read-only testing
Every builder above offers some checking of its own, and you should use it. Builders typically test from inside the project they built. An independent test starts from the live address, does not see your code, and can be repeated after every change to spot regressions. It is a second pair of eyes, not a substitute for your judgement. The general method is in How to test a vibe-coded app.
Test your AI-built app with Avalyz
Avalyz tests a web app from the outside, in real browsers, and returns a GO, NO-GO or INCONCLUSIVE verdict with evidence.
- Free test, no sign-up: the free test page, 3 tests a day. Or open
https://avalyz.com/essai?url=YOUR_APP_URL. - Read-only by default: nothing is created or changed on your app.
- Optional test account: signing in is the sole form submitted, then the signed-in pages are explored.
- Report you can paste back into the tool that built the app: findings ranked by severity with annotated screenshots, plus a PDF sign-off report.
- Plugs into your tool: a link, a badge, a CLAUDE.md block, an MCP server for Claude Desktop and Cowork, and GitHub Actions; see the method page.
- Plans: see pricing, with a 14-day trial and no card.
Avalyz proves what it saw and does not promise the rest: it is not a security audit, an accessibility conformity audit or legal advice.
FAQ
Can an AI test an app built by an AI?
Yes, if it is independent of the one that wrote the app and judges from the outside, like a visitor.
Which address should I test?
The public address of the published app.
Does my builder's preview count?
No. The preview can differ from the live app.
How often should I test?
After each change that goes live. A read-only test can be re-run by hand, by script or on a schedule.