Proof,not promises
Security you can trust, even in the AI era: 21,739 automated tests, 625 security checks, the OWASP Top 10 answered line by line, and a scale lab with half a million products.
Security you can trust, even in the AI era
Today, AI can write software in minutes. Trust still takes proof. This is how BLiNX eCommerce and POS is tested, attacked, locked down and checked by hand before it ever meets your customers.
Trust is the new feature
Every shop system says it is secure. We counted
automated tests
Written once, guarding every change after it.
tests done by hand
Real people, real till, real browser, real phone.
security checks
Each one tries to break in, cheat or steal.
access rules in the database
The rule sits next to the data, not in the screen.
On average, 14 automated tests stand behind every one of the 1,519 features in the platform.
Tested · what is inside the number
The anatomy of 21,739 tests
Five kinds of test, stacked like a building. The wide base checks every small part. Each floor above checks more of the real thing.
Database, load and others
Crowds of shoppers, years of data, release scripts.
Security checks
Each one tries to break in, cheat or steal.
Live system checks
Questions asked of a running system, end to end.
Real-browser tests
A robot shops, pays and manages the shop like a person.
Small-part tests
Every rule and sum checked alone, in 1,923 files.
Where the tests live, and what they guard
The most tests sit where the most can go wrong
By place · each test counted once
Shop website and admin
Shop pages, checkout, the admin screens and their security tests
POS till
Selling, stock, shifts, receipts and working offline
Whole-system journeys
End-to-end runs through the real till app and a real browser
Shop app
Browsing, cart, checkout, sign-in and the assistant on a phone
Server functions
The functions behind checkout, payments, stock and the till
Shared business rules
Pricing, promotion and tax rules shared by every app
Release and build checks
Checks on the release process itself
By topic · one test can count under more than one
Sign-in, roles and access
Money, checkout and payments
Orders, returns and delivery
Products, search and catalogue
Stock, batches and purchasing
Offline selling and sync
Accounting and reports
Accessibility and screen layout
Languages and currencies
AI features and their limits
random-input tests
The computer invents odd values to find the case nobody thought of.
access checks in the permission matrix
310 data tables against 10 kinds of user. No customer reached another customer's data.
tests run by hand in every round
Each on the real product, by a named tester, with a screenshot as proof.
Protected · we play the thief first
Attacked before you are
Every online shop gets attacked. The only question is who finds the gap first. These are the break-ins the platform is designed against, and our own tests try them.
Hiding code in an order note
Harvesting customer profiles
Replaying a checkout
Hunting for payment keys
Opening a guest's order
Mapping the system from outside
Forcing free shipping
Guessing coupon codes
Changing the price
Reading someone else's order
Giving yourself staff rights
Faking a payment message
Skipping the sign-in
Using one coupon twice
Creating free gift cards
Two buyers, one last item
Peeking at another's wishlist
Promoting yourself to admin
Impossible cart quantities
Marking an order as paid
A poisoned image upload
automated security checks try to break in, cheat or steal
attack suites: checkout, payments, coupons, sign-in, staff rights, uploads and more
tests in dedicated security folders across the website, the till and the server
The world's standard list of web risks
The OWASP Top 10, answered line by line
Broken access control
An access rule on every data table (848 in all), and staff rights checked live in the database.
Cryptographic failures
AES-256 encryption, salted one-way hashes for PINs, and an always-encrypted connection.
Injection
Parameterised queries only, customer text cleaned before display, and a strict content policy in the browser.
Insecure design
The server sets every price, sales are all-or-nothing, and big refunds need a second person.
Security misconfiguration
A security gate in every build refuses risky settings before they can ship.
Vulnerable components
Libraries are checked for known flaws before a release; secrets are scanned out of the code.
Sign-in failures
Two-step sign-in enforced in the database, lockouts on wrong guesses, and stolen-token detection.
Software and data integrity
Updates, themes and licences carry a digital signature. Unsigned means refused.
Logging and monitoring
A trail of every admin change, a tamper-evident till log and a service health monitor.
Server-side request forgery
Requests the server makes to outside addresses pass a guard that blocks internal networks.
OWASP Top 10, 2021 edition. Design and reviews also follow the OWASP ASVS checklist.
The engineering, named
A stolen password is not enough
Most break-ins start with a login. So the second step is enforced inside the database itself, not only on the sign-in screen.
Two-step sign-in, enforced in the database
235 database rules demand the second step. A stolen password alone opens nothing.
Roles and permissions
12 roles and 552 separate permissions, checked live. A role change takes effect at once.
Sign in with a fingerprint or face
Passkeys on the web and in the shop app, with the second step still asked where it is on.
Risky actions ask again
Refunds, deletions and secret changes need the password and a fresh code.
Stolen-session defence
A copied sign-in token is spotted on reuse and the whole session is ended.
Wrong guesses lock the door
Repeated wrong passwords or PINs lock the account, with longer locks each time.
See every signed-in device
Sign the others out with one tap. Idle admin sessions close by themselves.
A second person for big steps
Large refunds and overrides at the till need a supervisor's own PIN.
Zero trust, in the database
The database says no by itself
In most systems the screen decides who may see what, and the database trusts the screen. Here the rule lives next to the data. Even a hacked screen cannot read what the rule refuses.
guarded data tables
Every one has row-level security switched on.
access rules
Who may read or change each row, written next to the data.
second-step rules
Rules that refuse staff data without the two-step code.
validity checks
Wrong values are refused at the door: a negative price, an impossible date.
relationship rules
No order without a customer, no line without a product.
guard triggers
Automatic guards that fire on every write, whoever sends it.
hardened functions
Each one locked against the classic hijacking trick.
separate permissions
Across 12 roles, from cashier to owner.
Built to last · tested with the future's data
Fast on day one. Still fast at half a million products
We filled a lab with years of made-up trading, then measured the shop again.
products in the test shop
orders
order lines
customers
This is where we tested, not where it stops. These times come from one small desktop-class machine. A bigger server carries more products, more orders and more years.
Time to fetch, with all that data inside
One product page
Everything the page needs, in a single fetch.
The shop's home catalogue
About two hundred times faster than before the lab tuned it.
A category of 50,000 products
With its filters, one page at a time.
Search across the whole shop
Half a million products, with filters.
Within a blink, even on a starter server
Share of everyday requests answered in a twentieth of a second with the lab throttled to the smallest server size.
Database time measured in the BLiNX scale lab. Network time comes on top.
Straight answers
Hard questions, answered
The questions boards, owners and developers ask before they sign.
No. It is 21,739 separate checks, split across the website, the till, the shop app and the server, and thousands of them sit on money and sign-in, where mistakes cost the most.
Tested in the open
Ask us for the proof
Any number on this page can be shown to you live: the tests, the hand-run round, the attack suites, the scale lab. Bring your hardest question.
The maths that guards your data
Open, proven algorithms that banks and browsers rely on. No home-made secrets.
Encrypted at rest
Sales waiting on an offline till and stored session tokens are unreadable if copied.
PINs as one-way hashes
Staff PINs are never stored. Only a slow, salted hash is.
A slow lock on the offline PIN
The PIN is a short number, so the real protection is the limit: a few wrong guesses lock it, offline hours are limited, and the server checks again when the till is back online.
Keyed hashes for codes
A database leak would not reveal live tokens or activation codes.
Signed, or refused
Updates, themes, licences and offline sales carry a digital signature.
A log that shows tampering
Each till security entry is chained to the one before it.
Always an encrypted line
Browsers are told to refuse any unencrypted connection.
Master keys stay in the vault
Platform secrets are wrapped by keys that never leave a cloud key vault.
Keys sealed by the device
On a Mac with a security chip, running the build signed by BLiNX, the till's key is kept in that chip and cannot be copied off the machine. On other computers, today including Windows, it is kept in the computer's user-account keystore. That stops other users and a copied file, but not software running inside the same account.
Card numbers never touch the shop
Cards are typed straight into the payment provider's own secure fields.
A sale happens completely, or not at all
Lost internet, a double click, two tills on the last item: the places where ordinary systems lose money.
All or nothing
Stock, payment, receipt and accounts are saved in one step. A sale can never be half-saved.
Safe to send twice
A double click or a retry after lost internet lands on the same order. Nobody is charged twice.
Two tills, one last item
The record is locked while it changes, so the same item cannot be sold twice.
Payment messages are verified
Each payment notice is signature-checked and processed exactly once.
Paid must match charged
An order is marked paid only when amount and currency match the payment.
Books that balance by rule
The database itself refuses an entry whose debits and credits differ.
Posted accounts cannot be edited
No quiet changes, no deletes, no back-dating into a closed year.
One authority for stock
Every stock move goes through one gate that counts each command once.
A missing receipt number is noticed
Bill numbers are issued in fixed blocks, and the server alerts on a gap.
A stolen till is just a box
The till keeps selling when the internet drops. That power needs its own set of locks.
The login only works on that till
Every request is signed by the till's own key. A login file copied to another computer does not work there.
Selling offline, safely
Offline sales are signed as they are made and cannot be faked or altered before they sync.
An encrypted waiting queue
Sales waiting to sync are unreadable if the computer is copied.
A deleted sale is noticed
The queue is chained and sealed. Removing or editing an entry blocks the till and alerts the server.
A locked-down program
The till app runs sandboxed, refuses injected code and checks its own files.
Only genuine updates install
Every update is signature-checked and re-checked just before it runs.
Clock tricks do not work
Turning the date back does not extend a licence.
Lost or stolen? One screen
Suspend it, block it or have it erase its own shop data from POS Master.
Limits you set
How long and how much a till may sell offline is a signed permit with an end time.
The shield around every request
What stands between the open internet and your shop, and what a change must pass before it ships.
A strict content policy
The browser runs only our own scripts and the shop cannot be framed by another site.
Customer text is cleaned
Reviews and notes cannot carry hidden code onto your pages.
Only your sites may call
The server answers browsers only from addresses on its list.
Floods are throttled
Guessing and hammering are slowed and refused across all servers.
Bots stopped at the door
A human check guards checkout and sign-in, verified on the server.
Uploads checked by content
A file is judged by what is inside it, not by its name.
No reaching inside
The server cannot be tricked into calling internal addresses.
A security gate in every build
Risky code patterns fail the build before a person even reviews it.
No keys in the code
Scanners block passwords and keys from entering the codebase.
A full green run, or no release
Dozens of checks must all pass before anything ships.
AI helps your shop sell. It never gets the keys
An AI assistant talks to strangers all day. Some of them will try to trick it. So ours is fenced in from the first message.
It knows only public facts
What the assistant knows is assembled on the server from what any visitor may see. Nothing private goes in.
Customer text is never a command
The assistant is instructed to treat messages as questions, not orders, and to keep its rules to itself.
Hidden prices stay hidden
Quote-only prices are removed before the AI ever sees them.
Bots cannot run up a bill
A bot check comes before the chat, with limits per visitor and per shop each day.
Bounded in and out
Messages and answers have fixed limits, and only valid product ids are accepted.
Shoppers choose
AI features sit behind the shopper's own privacy choices, and each can be switched off by the shop.
Their data stays theirs. Yours stays yours
Data terms follow the GDPR and Sri Lanka's Personal Data Protection Act.
A shopper can take their data
One request returns what the shop holds about them.
And have it erased
Personal details are removed; orders keep only the amounts the law requires.
No card numbers in the shop
Card entry happens in the payment provider's fields. The shop keeps only the last four digits.
You choose how long logs live
Retention is a setting, with legal minimums that cannot be undercut by mistake.
Your shop is not a row in ours
Each business runs on its own database. Your data is not mixed with anyone else's.
Clear choices
Cookies, AI and privacy choices in one place, with a real "reject".
One slow part never stops the sale
Enterprise building blocks, so the shop keeps selling while the housekeeping happens behind it.
Work waits in a queue
Emails, alerts and page updates run in the background. A slow email never slows the till.
Screens update themselves
New orders and alerts reach the admin screen live, without a refresh.
Ready-made answers
The shop reads summaries built in advance, so big catalogues stay quick.
The till carries its own catalogue
Search and selling work from the till's own store, online or not.
Pages served from nearby
Catalogue pages are cached close to the shopper and refreshed quietly.
Housekeeping runs itself
Clean-up, exchange rates and reconciliation run on a timetable.
You know when a service is down
Payments, mail and storage are checked every few minutes, with one alert per outage.
Backups that are tested
Backups are encrypted before storage and practice-restored.