Authentication and permissions
Generated code usually checks that a user is logged in. It far less often checks that this user may see this record. Row-level authorization is the single most common hole we find.
For apps built with Cursor, Lovable, Bolt, v0, Copilot and friends
You built it fast with AI.
We make it safe to run with real users.
Shipping an MVP with AI tools was the right call. Now it has paying customers, and the parts the tool was never asked about — permissions, secrets, backups, load — are what break next. We audit the code, the data and the hosting, fix what matters, and keep it running.
Read-only access. Nothing is changed until you say so.
Sound familiar?
- It works in the demo but something breaks every week in production
- Nobody on the team can explain why a given piece of code is there
- There are API keys in the repository and you are not sure which are live
- A customer sent a security questionnaire and you do not know the answers
- The database has never been backed up, or never been restored from a backup
- Deploys happen from one person's laptop
- You found out about the last outage from a user
- The AI tool keeps 'fixing' things by rewriting files it should not touch
Three or more of these and the audit will pay for itself in the first week.
What the tool was never asked about
Six places AI-built apps break, in the order we find them
None of these show up in a demo. All of them show up with a few hundred real users, a curious customer, or one bad deploy.
Secrets and environments
Keys pasted into source, one .env shared between development and production, no rotation. We move secrets to a managed store, split environments and rotate anything that was ever committed.
Data integrity
Missing unique constraints, no foreign keys, migrations run by hand. The app looks fine until two requests race, then you have duplicate orders and a support queue.
Backups and recovery
A backup that has never been restored is a hope, not a backup. We set automated backups with a retention you choose and rehearse a restore so the number in your runbook is real.
Performance under load
N+1 queries, no indexes, everything loaded into memory. Fine at ten users. We find the queries that will fall over at a thousand and fix them before you get there.
Deployment and observability
No CI, no tests, no rollback, no alerting. We add a pipeline that runs tests and deploys the same way every time, and monitoring that pages a human before users notice.
How a rescue works
Audit first, fix second, then keep it that way
Each step stands on its own. Most clients start with the audit, take the fix list, and decide from there.
- 1 · Audit
Production-Readiness Audit — $2,500, 5 business days
Read-only access to the repository and the cloud account. You get a written report, a fix list ranked by risk to users and effort to fix, and a 60-minute walkthrough. You can execute it without us.
What the audit covers → - 2 · Rescue sprint
Fixed-price fixes for the items that matter
We quote the top findings as a fixed-price sprint, typically two to four weeks. Every change ships with tests, a written plan and a way to roll it back. Your product keeps running throughout.
How project work is quoted → - 3 · Run it
DevOps Retainer — from $1,200 / month, cancel any time
A senior-led team that knows your system, on call for patching, incidents and the small changes nobody has time for. Three tiers; most rescued apps stay on this so they do not drift back.
Retainer details →
The same checklist we run on our own production systems
We operate a US healthcare platform under SOC 2 remediation and a two-country marketplace with Stripe Connect payments. The audit uses the list we run before any of those goes live — not a generic scanner report.
- Senior-led: 12+ years of production engineering on every review
- Written report in plain language, ranked by risk and effort
- Every production change ships with tests, a plan and a rollback
- 3–4 hours of daily overlap with US Eastern and CET, written English
Questions founders ask
We built the whole thing with AI. Will you judge us for it?
No. We use AI tools on every project and say so on our home page. Shipping fast was the right call; the audit is about what the tools were never asked to think about.
Do we have to rewrite the app?
Almost never. Generated code is usually fixable in place. A rewrite is on the table only if the data model is wrong at the core, and we will tell you that in the audit rather than discover it mid-sprint.
Which tools and stacks do you work with?
Apps started in Cursor, GitHub Copilot, Claude Code, Lovable, Bolt, v0, Replit and similar — on Node.js, NestJS, Next.js, Laravel/PHP, Python, Flutter and React Native, hosted on AWS, Vercel, Supabase, Railway or similar. If yours is different, ask; we will say honestly whether we are the right team.
What access do you need?
For the audit, read-only: a repository invite and a read-only cloud role. Nothing is changed. For the rescue sprint we work in your repository through pull requests you can see and review.
What if we only want the report?
That is fine and common. The audit is priced to stand alone, and the fix list is written so your own team can act on it.
How fast can you start?
The audit usually starts within a week of the call and takes five business days. Book a 20-minute call and bring the repository link.
Bring the repository link. Leave with a plan.
In 20 minutes we will tell you whether the audit fits, what we expect to find, and when we can start. No slides.