Skip to content
MERVENORBook a triage call

For startups — Auspicia · Apps built with AI tools

Built it with AI tools? We make it safe for real users.

Lovable, Bolt, Replit, Cursor, v0 and similar tools get a working first version in days. What they often leave behind is harder to see: database tables anyone can read, keys in the browser, access checks that only happen on screen, no tests and no reliable way to deploy. We fix vibe-coded apps the way we take over any inherited codebase: a five-day Auspicia check first, then a fixed-price Firmora sprint for what matters before launch.

AI-BUILT APP CHECKyour-appOpen database rulesneeds workreadyKeys in the browserneeds workreadyChecks only on screenneeds workreadyTests and deploysneeds workreadyCosts and limitsneeds workreadysafe to ship
The check
Five business days
Price
Fixed fee, confirmed on the call
Then
A Firmora sprint, fixed price
Your code
Stays in your accounts
Read-only access for the check. Nothing is changed until you approve the plan.

01The usual gaps

What AI tools usually leave behind

Open data

Tables anyone can read

Database tables or storage buckets that anyone with the app's public key can read, or change.
Keys in the browser

Keys anyone can copy

Secret database keys or paid API keys shipped in the app's code, where anyone can copy them and run up your bill.
Checks only on screen

Rules the server ignores

The interface hides buttons from people who should not use them, but the server still accepts the request.
No tests, no safe deploys

Every change is a gamble

No tests around the parts that matter, no test environment and no way back.
Code nobody has read

Logic nobody understands

Duplicated logic, unused packages, and sometimes packages that were suggested but are unmaintained or do not exist.
Costs that grow with traffic

Bills with no ceiling

AI calls and database queries with no limits or caching, often billed to a personal card.

02Tools

The tools we see

We work on the code these tools produce: usually React or Next.js with TypeScript, a Node.js back end, and Supabase or Firebase for data and sign-in.

Lovable and Bolt

React front ends, usually on Supabase.

Replit

Full-stack apps built and hosted on Replit.

v0

Next.js and React interfaces.

Cursor, Windsurf and Claude Code

AI-written code in a normal repository.

Base44 and other hosted builders

The check first confirms what code and data can be exported.

If your tool can export the code or connect a GitHub repository, we work there. If it cannot, the check says so on the first day.

03How it works

From first call to a safe launch

  1. Triage call30 minutes

    You show us the app and what worries you; we tell you what to check first.
  2. Auspicia checkFive business days

    A read-only review of the code, the database rules, the keys and the deploy: a written report with every finding ranked by severity, and a fixed-price plan.
  3. Firmora sprintTwo or four weeks

    The risky parts fixed in your repository, with tests around them and a deploy you can undo.
  4. Servata retainerOptional, monthly

    Senior engineering cover until your first engineering hire.

04Keep or rebuild

Do you need to start again?

Rarely. An app built with AI tools is an inherited codebase like any other: most of it can be kept, made safe and built on. If a part is cheaper to rebuild, the report says so, with the evidence.

You can keep building with your AI tool afterwards; the sprint adds tests and checks that catch a change that breaks something important. Before you launch, run the security checklist for AI-built apps and the Supabase checks yourself.

05Questions

Frequently asked questions

Not answered here? Ask on the triage call.

Can you fix a Lovable or Bolt app that works in preview but fails in production?

Yes. The gap between preview and production usually comes from environment settings, database rules, sign-in configuration or code that should run on the server; the check finds which, and the sprint fixes it.

Is our users' data exposed?

The check tests it the way an outsider would: with the app's public key and no account, and as one user trying to read another user's data. If anything is exposed, we tell you as soon as we find it, not in the final report.

Do we have to stop using AI coding tools?

No. Keep using them; the sprint adds tests and a review step, so a new generated change cannot quietly undo the fixes.

What does it cost?

The check is a fixed fee, confirmed on the triage call; the report prices the sprint that follows, with written scope and exclusions. Nothing is billed by the hour.

Will you sign an NDA?

Yes: a mutual NDA before we see the code, read-only access for the check, and all later work in repositories and accounts you own.

Guides

All guides
Checklist · Taking over a projectSecuring the Supabase back end of an AI-built app: ten checksAI-built apps on Lovable, Bolt and similar tools often leave Supabase tables open. Ten checks for row-level security, keys, grants and storage.Read the guideGuide · Taking over a projectBuilt with AI coding tools? A security checklist before you launchApps built with AI coding tools often ship with open databases, exposed keys and missing access checks. Twelve checks to run before real users arrive.Read the guideChecklist · Taking over a projectTechnical due diligence: what investors check in your code, and how to prepareWhat investors and acquirers check in a startup's software: ownership, security, architecture, scaling, costs and key-person risk, and how to prepare.Read the guide

06Start

Tell us about the application

Thirty minutes, no slides, no sales script. You leave with a view on what to check first.

The check
Five business days
Price
Fixed fee, confirmed on the call
Then
A Firmora sprint, fixed price
Your code
Stays in your accounts