Your app works.
Is it ready for real users?
Sentinel is an independent review for apps made with Lovable, Bolt, Cursor, Claude Code and other AI tools. It checks your live site and your code, gives you fixes in plain English, proves them with a rescan, and awards a trust badge your customers can verify.
ILLUSTRATION · A REVIEW, SPED UP
Built for apps made with
AI builds fast. It does not check its own work.
Apps built with prompts ship in a weekend. Many ship with a secret key anyone can read, a database anyone can open, or no limit on what an attacker can try. The builder can't vouch for its own output, and your customers know it.
live AI-built apps scanned in one public study
high-impact vulnerabilities found among them
exposed secrets such as API keys
cases of exposed personal data
Source: Escape's "State of Security of Vibe-Coded Apps" study. Read the study ↗
"It works" and "it's ready" are different sentences. Sentinel is what stands between them.
Review. Fix. Prove. Show.
Think of a building inspector. They walk the outside, read the plans, list what must change, come back to confirm, and only then put a plate on the door.
Review
We check your live site and your code together, plus your database settings if you connect them.
Required fixes
Every problem in plain English, with a ready-to-paste prompt written for the tool you built with.
Rescan
We check again. A fix only counts when the rescan proves it. Two rescans are included.
Badge
Pass every required check and get a public verify page and an embeddable badge with a signed record.
Fixes you can paste. Not a wall of jargon.
Every issue is a short card: what's wrong in everyday words, where it is, and a prompt for your AI tool that fixes it. Below is an example card from a made-up app.
SEC-KEY-001SEC-DB-001CAP-RATE-002SEC-HDR-004Prompts always ask your tool to work on a branch, change as little as possible, keep tests passing and never switch a check off.
Four things a serious buyer asks about
Each gets a score from 0 to 100. Scores explain the picture. The badge is decided only by whether every required check passes, so a good average can never hide a serious flaw.
- Secrets and keys: in your code, your git history and the scripts your site ships
- Who can do what: does the server check who's asking? Can users read each other's data?
- Database rules: Supabase row-level security, Firebase rules, open storage buckets
- Payments and AI features: signed webhooks, server-side confirmation, model keys kept off the browser
- Live site: HTTPS, cookies, headers, exposed files, known-vulnerable packages, fake packages
- Will it stay up? error tracking, timeouts, sensible retries, backups
- Will it cope with growth? queries with limits, indexes, rate limits, AI spend caps
- Pressure rehearsal: a ready-made load-test script you run on your own staging site, turned into a plain-English capacity card
- Cost at scale: an estimate of your monthly bill at 1,000, 10,000 and 100,000 users
- Privacy basics: is a privacy policy linked? Do trackers load before consent?
- Data map: what personal data you collect, where it goes and which third parties get it
- Readiness for UK/EU GDPR, PECR, the EU AI Act and accessibility (WCAG 2.2 AA)
- Cyber Essentials and PCI payment-page readiness for the parts a website review can see
- Security questionnaires: answers pre-filled with evidence links, ready to send to a buyer
- "Are you building this the right way?" about 30 sound-practice checks, sized to your app
- Structure: tidy modules, errors handled not swallowed, inputs validated
- Testing and delivery: tests that cover sign-in and payments, CI on every push
- Maintainability: setup docs, unused code, oversized files
- Never "rewrite it". Advice always names the smallest change that fixes the problem
Rings above are illustrative. Some checks arrive in stages; the progress section below shows what is built.
Three levels. Each one published and checkable.
Every level is a public list of controls. Your badge links to a verify page showing what was reviewed, when, and against which version. It stays live only while re-checks keep passing.
Launch Ready
Safe to put real users on it.
- Secrets and keys protected
- Server-side access checks
- Users can't read each other's data
- Database and storage rules
- Payments and AI-feature safety
- HTTPS, cookies, dependencies, privacy policy
Scale Ready
Keeps working as users arrive.
- Everything in Launch Ready
- Error tracking, timeouts, backups
- Tests and CI on every push
- Bounded queries and indexes
- Rate limits and AI spend caps
- Page-speed budget
Business Ready
Ready for a business customer's review.
- Everything in Scale Ready
- Data map and sub-processors
- Trackers and consent checked
- AI transparency
- Questionnaire answers with evidence
- 30 days of clean re-checks
Above these sits an optional human-verified tier: a booked review by a qualified tester. It is the only thing we ever call a penetration test.
If you ship with AI, this is for you
Built so that someone who has never opened a terminal can onboard, understand the report and fix every issue by pasting prompts. If that works, it works for everyone.
Prompt-first builders
You build in Lovable, Bolt, v0, Replit or Base44. You want to know if it's safe and get fixes you can paste. No jargon, no terminal.
"Just tell me what to fix and how."
AI-assisted developers
You work in Cursor, Claude Code, Codex, Antigravity, Copilot, Gemini Code Assist or Windsurf. Checks run inside your editor while you build.
"Catch it before I push."
Agencies and freelancers
You build client apps with AI. Hand over a signed review letter, a badge and a verify link. White-label reports, client invites and bulk status.
"Prove the work before I invoice."
Founders facing a buyer
A customer's security questionnaire. An investor's due diligence. A payment provider's review. Get evidence they can check for themselves.
"Can we trust this app?"
Not another quick scan
Quick scanners give you a score. Sentinel gives you a decision you can defend, based on a published standard.
| Sentinel | Typical quick scanner | |
|---|---|---|
| How the badge is decided | By code, from every required control of a public standard | By a score threshold, which can hide a serious flaw |
| What is checked | Live site + code + platform settings, cross-checked | Often the URL only |
| Fixes | Prompts for your tool, proven by a rescan | Suggestions, unproven |
| Can users see each other's data? | Tested. "No data to compare" never counts as a pass | Rarely tested |
| The badge itself | Signed record, last-verified date, expires if not re-checked | A static image |
| Beyond security | Reliability, capacity, privacy and engineering quality | Security only |
| Where it works | Inside Cursor, Claude Code, Codex and more | A website you visit afterwards |
| Independence | A third party, not the tool that wrote the code | Sometimes the builder itself |
Prevent it. Catch it. Keep it honest.
Sentinel doesn't wait until you've finished. It works alongside the AI that writes your code, from the first prompt to every push.
Prevent
A free rules pack teaches your AI tool to write safer code from the start: "never put the service key in browser code", "turn on row-level security for every table".
Suggest
Your tool asks Sentinel in its own chat: "what's the next fix?" and gets the prompt back. A local mode checks changed files privately, offline.
Enforce
Pull-request checks look at every push. If something serious slips back in, the badge changes and you get an email.
Autopilot fix mode: your AI tool works through the fixes on a separate branch, one commit per issue, and stops to ask before touching anything about sign-in, payments or deleting data.
A security tool that has to earn your trust
You're giving us access to your code, so the rules we hold ourselves to are written down and built into the product.
- ✓Read-onlyWe look, we never change anything. No load is generated against your app.
- ✓Your code is never runIt's read inside a locked-down container and deleted after the review.
- ✓Secrets never leaveKeys are redacted the moment they're found. They never reach a report, a log or an AI model.
- ✓Every claim has evidenceIf a report says it, it links to proof you can open.
- ✓Code decides the badge, not AINo model can award a badge or wave away a failed check.
- ✓You prove ownership firstDeep checks only run on domains you've verified. Unverified sites get public checks only.
We're building in the open
The review engine already exists and is tested. What's left is the friendly layer around it. Here's where things stand.
Estimates as of 29 September 2026.
Join first. Shape what we build.
We're opening in waves. People on the list go first.
Things people ask
Is this a penetration test?
No. The automated review is a readiness review and we never call it a pen test. If you want a human-led test, an optional human-verified tier will let you book a qualified tester. That is the only thing we call a pen test.
Does the badge mean my app is secure?
No, and we won't say it does. The badge says the app met every required control of a published standard on a stated date, and it shows when it was last re-checked. No review can promise an app can't be breached.
Do I need to be technical?
No. Every step is a button or a single prompt to paste into the tool you already use. Technical terms are always explained in brackets, and there's an "explain it simpler" option on every issue.
Will you touch or change my app?
No. Checks are read-only. We don't generate load against your app, and your code is never run. Any pressure test is a script that you choose to run on your own staging site.
What do you need access to?
At most: your live URL (with proof you own it), read-only access to one repository, and optionally read-only access to your database platform or two test accounts. Every extra is optional, and skipping one shows the affected checks as "not assessed" instead of passing them silently.
Which tools do you support?
Lovable, Bolt, v0, Replit, Base44, Cursor, Claude Code, Codex, Antigravity, GitHub Copilot, Gemini Code Assist and Windsurf, plus hosting and backends like Vercel, Netlify, Cloudflare, Supabase and Firebase.
Does it cover compliance like GDPR?
It checks readiness for the parts a review can see: privacy policy, trackers before consent, data map, sub-processors, deletion path, AI transparency, accessibility, Cyber Essentials and payment-page scripts. Some of these arrive with Business Ready. It is readiness, never a legal certification.
How much will it cost?
There will be a genuinely useful free tier. Paid options will include a one-off review and monthly plans. Final prices are announced to the waitlist first.
When does it launch?
We're targeting a public launch in early 2027, with early access opening to the waitlist before that. Dates can move, so join the list and we'll tell you the moment it does.
Ship it knowing it's ready.
Join the early-access list and get the first invitation.