Skip to content

You built a client portal with AI. Check that one client can’t see another client’s files.

App369 checks and fixes client portals that owners built with Lovable, Replit, Bolt, ChatGPT or Claude, starting with the $450 AI App Launch Check: 3 business days from read-only access, testing sign-in, roles and who can read what. Every problem found comes back in writing, ranked by severity with evidence.

Page updated . Price checked .

Short answer

Start with the $450 AI App Launch Check if the portal mostly works and clients are about to log in. If it runs on Firebase and the only worry is who can read what, the Firebase Security Rules Audit is $950. For one bug you can name, the App Fix Ticket starts with a $450 diagnosis. If it needs a rebuild, the Custom CRM Build starts with a $450 scoping.

Why one client can see another client’s files

A client portal has one job the others don’t: every client signs in to the same app, and each must see only their own files, invoices and messages. The screens can look right while the database behind them hands out everything.

The Launch Check covers portals that keep their data in Supabase or Firebase. In Supabase, row-level security is the database setting that decides which rows each signed-in person may read. In Firebase, security rules do the same job. If those rules are missing, or were written once and never tested with two real accounts, the portal hides other clients’ data on screen but still gives it to anyone who asks the database directly.

A preview tried with one account, yours, can’t show this. Nothing looks wrong until a second client signs in.

In February 2026, The Register reported a security researcher’s claims about one app hosted on Lovable. The article says the app itself was a platform for creating exam questions and viewing grades. It was not a client portal, but it has the same shape: many people’s records in one app, each meant to be private. By the researcher’s count, 18,697 user records were exposed, and an unauthenticated attacker could “trivially access every user record”. The article puts the cause in general terms: when the builder, AI or human, “fails to explicitly implement crucial security features like Supabase’s row-level security and role-based access”, the code that comes out “looks functional but in reality is flawed”. Its example is an access check written the wrong way round, “essentially blocking authenticated users and allowing access to unauthenticated users”. The article also carries Lovable’s response, from its chief information security officer, Igor Andriushchenko: “Any project built with Lovable includes a free security scan before publishing,” “Ultimately, it is at the discretion of the user to implement these recommendations. In this case, that implementation did not happen,” and “This project also includes code not generated by Lovable and the vulnerable database is not hosted by Lovable.”

Source: The Register, “Lovable-hosted app littered with basic flaws exposed 18K users, researcher claims”, published , by Connor Jones, reporting the claims of researcher Taimur Khan. Read . These are the article’s facts; what to do about them is App369’s view.

Test it yourself with two test client accounts

You can run this without reading any code. Use made-up test clients, not real ones. Run it on the published portal, at the address your clients use, not in the builder’s preview.

  1. Create two test clients, A and B, each with their own sign-in, and give each one a file or an invoice with a made-up name you’ll recognise.
  2. Sign in as client A. Open A’s file or invoice and copy its address from the browser’s address bar. If there is a download button, copy that link too.
  3. Sign out. Sign in as client B and paste each address into the address bar.
  4. Still signed in as B, look at every list and search box: search for A’s made-up name, and check any “all files”, “recent” or dashboard count.
  5. Still signed in as B, open one of B’s own files or invoices. If its address ends in a number, change the number by one and press Enter. Seeing another of B’s own items is fine; seeing A’s is a fail.
  6. Open a private browser window, where nobody is signed in, and paste the same addresses again.

When it fails. It fails if B, or the private window, shows anything of A’s: the file, its name, the invoice amount, or a count that includes A’s items. A blank page, a sign-in screen or a “not found” message is what you want to see. One result needs a second look, described next.

What can look like a fail. A download link that opens in the private window is not always a fail. Some file storage gives each file its own long link ending in a random code, and that link is built to open for anyone who has it, the way an “anyone with the link” share works in Google Drive or Dropbox. An invoice page on your payment provider’s own address is the same case: Stripe’s documentation says each invoice gets a unique link with a long, random code in it, for you to send to your customer. Client B can’t find A’s link on their own. But anyone the link is forwarded to can open the file without signing in, so decide whether that is acceptable for what your clients upload. Two things are a fail every time: the page address from step 2 opening for B or in the private window, and a download link short enough to guess, such as one that ends in a file name or an invoice number. If you see A’s file as B, reload the page once before you count it, because a browser can show a copy it saved while A was signed in.

What a pass doesn’t prove. A pass doesn’t prove the database is closed. The screens can hide A’s files from B while the database still hands them to anyone who asks it directly. That is the kind of gap The Register’s article describes: by the researcher’s account, someone who wasn’t signed in could reach every user record. Only a check of the access rules themselves shows it.

If it fails. If real clients already use the portal, stop adding files and inviting clients until it is fixed. If the private window opened a page of A’s, take the portal offline until it is fixed, because anyone who has the address can open it without signing in. Taking the screens down may not close the database behind them, which is why the next step checks the access rules themselves. Then start with the $450 AI App Launch Check, on Supabase or Firebase: it checks who can read what across the whole portal and ends with a fixed quote for the fixes. If the portal runs on Firebase and this is the only worry, the $950 Firebase Security Rules Audit includes the rule fixes. See the paths and prices

Where AI-built portals break

Four places, in the order they hurt.

Who can read what

One client can open another client’s files, invoices or messages, or a client can change their own role and see the admin screens.

A new client’s invite never arrives, the emailed sign-in link loops back to the sign-in page, or the password reset link doesn’t work.

File uploads

Large files fail to upload, uploaded files end up at links that open without signing in, and storage costs can grow with every upload.

The everyday gaps

Clients can’t see where their project stands, the portal is hard to use on a phone, and so clients keep emailing you instead of logging in.

Which path fits your portal?

Pick the row that sounds like you. Each is a separate path with its own price, and nothing starts without your yes.

If this sounds like youStart withWhat you getWhat comes next
If it mostly works and clients are about to log in, or already have: AI App Launch Check $450 3 business days from read-only access. Pay $450 for a Launch Check Sign-in and roles, data access (Supabase row-level security, policies and grants, or Firebase rules), payments and deploys, checked, with every problem found ranked by severity with evidence and a fixed quote for the fixes. Rescue Sprint: $1,950, or $1,500 after the $450 credit, for up to 5 critical or high issues. Only with your yes.
If it runs on Firebase and the only worry is who can read what: Firebase Security Rules Audit + Fix PR $950 5 business days. Pay $950 for a Rules Audit A written role and tenant access model (who may read and change what), a test suite that tries cross-user and cross-tenant reads and writes and role escalation (one client reaching another client’s data, or a user making themselves an admin), a report ranked by severity, and a pull request (a code change you review and deploy yourself) with rule fixes that passes the suite. Everything runs on the emulator or test data only, never your production records. The emulator is a test copy of Firebase that runs apart from your live app. Firebase only: Firestore and Storage rules, plus up to 15 Cloud Functions, in one project. For a Supabase portal, the Launch Check is the step.
If there’s one bug you can name, such as an invite that never arrives or a reset link that loops: App Fix Ticket: Diagnosis $450 3 business days from access to the code and the steps to reproduce the bug. Pay $450 for a Diagnosis The root cause, with evidence, and a fixed quote for the repair. The $450 is credited toward the fix. The App Fix Ticket is a $450 diagnosis, then $750 for a tested fix: $1,200 in all. The fix ships as a pull request with a test: a code change you review and deploy yourself.
If it needs a rebuild: Custom CRM Build: Scoping $450 3 business days from your stages (the steps each client moves through in your portal), a sample export and the list of tools it must connect to. Pay $450 for CRM Scoping What the portal must do, written down: who signs in, what each client sees, the screens and the tools it connects to, which path fits (Simon’s own Workspace369 backend, or fully custom) and a fixed price and timeline for the build. The build: from $6,000, at the fixed price in the written quote, with the scoping credited in full against the build. Only with your yes. Not included: your Workspace369 plan, hosting and third-party service costs.

The Launch Check covers the code AI builders produce: React, Next.js or Vite front ends on Supabase or Firebase, with Stripe. If your app is outside that, you’ll hear so before you pay. Check fit by email first if you’re not sure.

Simon builds and runs Workspace369, which uses Firebase, not Supabase. On Supabase apps, the same checks are done with Postgres row-level security, grants and policies.

Full refund if Simon can’t take it on once he sees your project, or if you cancel before work starts. Refund terms.

No issues worth fixing, no charge. If the Launch Check finds nothing rated medium or higher, Simon refunds the full $450. You keep the report either way. This applies to the first 10 Launch Checks bought after October 30, 2026.

Prices in USD, paid upfront. Price checked .

Or buy a portal instead

If your clients need to sign in, see their proposals and invoices, sign and pay online, a ready-made portal may cost less than fixing a custom one. Workspace369, Simon’s own product, has a branded client portal on every plan, from $29 a month for one seat, with proposals clients approve and sign online and invoices they pay online. For a one-time fixed fee, Simon sets it up for you.

Workspace369 doesn’t fit a portal that is itself the product you sell, or one built around a workflow no ready-made tool has. If your clients only need to download files, one shared folder per client in Google Drive or Dropbox may be enough.

Keep your own portal when it does something no product does. Then one of the paths above fits.

Solo Setup $450
Your branding on the workspace and client portal, up to 100 clients imported from a CSV file, proposals clients approve and sign online, an invoice with online payment and one recurring invoice.
Runs on your own Workspace369 Cadet plan: $29/month for 1 seat, after a 14-day trial. Public intake forms are on every plan, Cadet included.
Team Setup $1,500
Up to 250 clients imported, proposals and invoices clients sign and pay online, a public intake form that creates the client record, and a booking page on the Specialist plan or higher.
Runs on your own Workspace369 Voyager plan ($74/month, 2 seats) or Specialist plan ($149/month, 10 seats), after a 14-day trial. Online booking needs Specialist or higher.

You pay for your own Workspace369 plan. The setup fee is one-time.

See what each setup includesTry Workspace369 as it is

Who does the work?

Simon, a senior developer in Miami, does all the work himself, with no account manager and no hand-off.

He also built and runs Workspace369, his own product, which has a client portal and is live on the web and the App Store.

Questions owners ask first

Who can fix a client portal I built with Lovable or ChatGPT?
Simon Dziak, a senior developer in Miami who founded App369 and does all the work himself. He built and runs Workspace369, which has its own client portal. A portal that mostly works starts with App369’s $450 AI App Launch Check, done in 3 business days from read-only access, which tests sign-in, roles and who can read what.
How do I know if my clients can see each other’s files?
Make two test clients, A and B. Sign in as A, copy the address of one of A’s files or invoices, sign out, sign in as B and paste it; then paste it into a private window with nobody signed in. If either shows anything of A’s, the portal fails, with one exception: a long download link that is built to open for anyone who has it. A pass doesn’t prove the database is closed, so App369’s Launch Check also checks the access rules themselves.
My portal runs on Supabase. Which App369 step fits?
The AI App Launch Check, $450. App369’s Firebase Security Rules Audit is for Firebase apps only. On Supabase apps, the Launch Check does the same access checks with Postgres row-level security, grants and policies.
How much does it cost to fix an AI-built client portal?
At App369, the AI App Launch Check is $450, and the Rescue Sprint after it is $1,950, or $1,500 after the $450 credit, for up to 5 critical or high issues. The Firebase Security Rules Audit is $950. One named bug is a $450 diagnosis, then $750 for a tested fix: $1,200 in all. A rebuild starts with the $450 Custom CRM Build scoping, credited in full against the build, and the build is from $6,000. None of these is billed by the hour.
Should I fix my AI-built portal or buy one?
At App369 the aim is to keep what works and fix what doesn’t. A portal that mostly works starts with App369’s $450 AI App Launch Check, which reads your code, ranks each problem it finds by severity and ends with a fixed quote for the fixes. If your clients only need to sign in, see proposals and invoices, sign and pay, a ready-made portal such as Workspace369, Simon Dziak’s own product, may cost less than a custom one: its client portal is on every plan, from $29 a month.
Do you need my production passwords?
No. For the App369 Launch Check, access is read-only: a repo invite, a test login and a schema export with no user data (a list of the database’s tables and fields). No production keys or passwords. Fixes ship as pull requests you deploy.
Do I need a call before I pay?
No. At App369 the scope, price and delivery time of each step are published, and everything happens in writing. If you want to talk live, calls cost $100 per 15 minutes, paid upfront.

Describe it first, or pick your path

Not sure which path fits? Send what you built, who uses it and what breaks, before you pay anything. Simon replies by email, usually within one business day.

Check fit by email

Full refund if Simon can’t take it on once he sees your project, or if you cancel before work starts. Refund terms.

No issues worth fixing, no charge. If the Launch Check finds nothing rated medium or higher, Simon refunds the full $450. You keep the report either way. This applies to the first 10 Launch Checks bought after October 30, 2026.

No call needed. Ask a question first by email at simon@app369.com or with the contact form; Simon replies by email, usually within one business day.