Blog

AI application review: what we check in an app from Lovable, Bolt or Cursor

An audit of the code, security and infrastructure, a list of risks by severity and a plan of fixes. Exactly what we check and how the review goes.

An AI application review is an independent audit of an application you built in Lovable, Bolt, v0, Cursor, Replit or Claude Code. You get an audit of the code, security and infrastructure, a list of risks ranked by severity and a concrete plan of fixes. Below we describe exactly what we check, how the review goes and what you get out of it.

Why have an AI application checked

Vibecoding tools can build an application that looks finished within a few hours. What you cannot see – who has access to what, where the keys are stored, what happens during an outage – is exactly where AI most often cuts corners. The application works until someone starts trying to break it.

A review tells you the truth before an attacker or a customer finds it out.

Who the review is for

Anyone who has an app built with AI and:

  • wants to release it to real users or customers,
  • handles personal data, payments or company data in it,
  • cannot say for sure whether it is secure.

What we check

The review covers three areas:

1. Code

The application logic, the architecture and whether the code does what it should even in situations the AI did not anticipate.

2. Security

Among other things we look at the questions vibecoded applications most often fail on:

  • Are there API keys or other secrets in the frontend or in the repository?
  • Does the database (for example Supabase) have row-level security (RLS) switched on?
  • Does the server verify sign-in and permissions on every request, or only the frontend?
  • Is there rate limiting on sign-in and the API?

3. Infrastructure and operation

  • Are backups made – and has anyone tried restoring them?
  • Who updates the application and who watches for errors?

What you get

  • An audit of the code, security and infrastructure – written so you understand it, not just a tool’s output.
  • A list of risks by severity – you know what to fix now and what can wait.
  • A plan of fixes – concrete steps your team can handle, or we do them.

How the review goes

  1. Contact. You write what application it is and what it was built in. An e-mail or a phone number is enough.
  2. Feedback. We go through the scope: what the review should cover and what we will need access to.
  3. The quote. The review is offered as a one-off; the price depends on the scope. You find the quote in the client zone and accept it with one click.
  4. The review. A human checks every finding – it is not an automated report.
  5. Results and a plan of fixes. You get a list of risks by severity and a plan for dealing with them.

What happens after the review

You have two options and both are fine:

  • your team makes the fixes according to the plan,
  • we take over the fixing, securing and deployment – that is the Rescue and getting it running service.

Frequently asked questions

Do you review applications from Lovable, Bolt, v0, Cursor, Replit or Claude Code? Yes, the review is made exactly for applications built with these tools.

How much does a review cost? It is offered as a one-off in a quote according to the scope.

Do I then have to use your fixes too? No. The plan of fixes is yours and you can give it to anyone.

Next step

Write to us what your application was built in and what it does. The contact form takes a minute. More about the service: Review of AI applications and infrastructure.