Someone pastes
A customer's card number goes into a chat box, with a perfectly good reason. Nothing has been sent yet.
Roger Shadelyt checks everything your people send to ChatGPT, Claude, Gemini, Copilot and 50 more AI tools, on their own PC, in about 12 milliseconds. Card numbers, IDs, keys and customer records stay inside. Everything else goes straight through.
Windows 10 and 11, Windows Server 2016 or later Chrome, Edge and Brave
Try it: this is a pretend AI chat. Type or paste anything and press Send.
Runs only in your browser: nothing you type here leaves this page. A few simplified checks; the product runs 127 detectors on the PC.
Protects your people on 54 AI tools out of the box, and any you add
AI tools made everyone faster. They also became the easiest way for company data to walk out of the door, usually by someone just trying to do their job well.
A customer list to “clean up”, a contract to “summarise”, source code to “fix”. Each one lands on a website your company doesn't control.
The text now sits on another company's servers under their terms. You can't recall it, delete it, or prove where it went next.
India's DPDP Act allows penalties of up to ₹250 crore for failing to protect personal data. GDPR goes up to 4% of worldwide turnover.
Block the sites and people switch to their phones and personal accounts. You lose the productivity and keep the leak, now where you can't see it.
What happens between someone pressing Send and the AI tool receiving anything. Scroll through it.
A customer's card number goes into a chat box, with a perfectly good reason. Nothing has been sent yet.
The browser extension pauses the paste, the Enter key, the Send button or the upload, in every frame of the page, before the AI site sees a thing.
The agent on the same PC undoes tricks like spaced, dotted or encoded digits and validates what it finds the way the issuer would: a card's check digit, an Aadhaar checksum. About 12 ms.
Block it, warn, ask for a business reason, or send it with the values replaced. The employee sees why, in their own language, and gets on with their work.
Your console gets an event: the kind of data, a masked value, the AI tool, the person. What they wrote is never stored, and never sent to us.
to check a typical paste, on the employee's own PC. People only notice Roger Shadelyt when it finds something.
127 detectors for 108 kinds of data, checked the way the issuer checks them, so a random number isn't mistaken for a card.
characters of what your people write are ever sent to us. The check happens in the PC's memory.
41 countries' ID, tax and bank formats, plus global ones.
54 AI tools recognised (74 web addresses), and any you add.
languages for the message your people see.
ready-made policy templates for GDPR, HIPAA, PCI DSS, DPDP, LGPD, PIPL and 17 more regulations.
Most of the time people don't need to be stopped. They need the card number, the ID or the password taken out, and the rest of their message sent. Switch between the two to see what the AI actually receives.
Hi! Please draft a payment reminder to Mr Rao. Card 4111 1111 1111 1111, Aadhaar 2345 6789 0124, call him on +91 98765 43210. Keep it friendly but firm.
Hi! Please draft a payment reminder to Mr Rao. Card [PAYMENT_CARD_1], Aadhaar [IN_AADHAAR_1], call him on [PHONE_1]. Keep it friendly but firm.
3 sensitive values found on the PC
Same message, same help from the AI. The values never left.
A clear message inside the AI tool, in 12 languages, saying what was found and why. People learn the rule instead of fighting it.
Upload a customer list and any of its records is recognised, from hashes only. Register a confidential document and it's recognised even reworded.
Word, Excel, PowerPoint, PDF, code, ZIP archives and the text in images. A file that can't be read is never treated as clean.
See which tools people actually use. Approve, tolerate or block each one, and review new AI-looking sites as they appear.
Which PCs are protected, which browser is missing the extension, and when someone switches protection off.
Every policy is signed and expires; every event is chained so edits and gaps show. If something fails, built-in rules still block the worst.
Deploy with Intune, Group Policy, NinjaOne, Kaseya or ConnectWise. Sync people from Entra ID or Active Directory. Alerts to Splunk, any SIEM or a webhook.
Owner, admin and analyst roles, two-step sign-in, single sign-on, and an audit log of every change, especially the ones that lower protection.
Messages to AI desktop apps that expose their message box, and scripts and SDKs that use Roger Shadelyt's local AI gateway, are checked too.
Your score and what to do first appear here as you go.
We run your console at
yourcompany.shadelyt.com. With Enterprise, everything it records is stored in your own PostgreSQL
database; during a free trial, in a database of its own on our server. Nothing to install on a server of
yours.
It knows your company name, plan and how many devices are protected. Not your events, your people or your policies.
The Shadelyt icon on every PC shows protection is on, and we give you an employee notice template to explain it.
No proxy, no network changes, and no change to how people use AI. Most teams go from first install to full protection in two weeks.
We set it up at yourcompany.shadelyt.com and you connect your own PostgreSQL database. Two-step
sign-in for every admin from the start.
One command through Microsoft Intune or Group Policy. PCs register themselves, and the browser extension is installed with the agent.
Monitor mode shows which AI tools your people use and what would have been stopped, without interrupting anyone.
Block, warn or redact, per AI tool and per kind of data, tuned to what you saw. People keep using AI; the sensitive parts stay home.
Try everything free on your own PCs. When it proves itself, move to Enterprise and keep what you set up, with your data in your own database. Nothing to install on a server.
Prove it on your own data first.
Everything, for your whole company.
Prices are in Indian rupees and exclude GST. The price per PC depends on how many PCs and for how long. A PC counts once, however many people use it.
No. It looks only at what is sent to AI tools, at the moment it is sent, and only on the employee's own PC. Your console receives what was found (the kind of data, a count and a masked value), never the text.
A typical paste is checked in about 12 milliseconds; a 100 KB paste in about a fifth of a second. People only notice Roger Shadelyt when it finds something.
No, and we'd advise against it. You decide per tool: approve it, tolerate it, or block it. Roger Shadelyt makes the tools you allow safe to use.
Standard users can't remove the agent or the force-installed extension, and a stopped agent comes back within seconds. Any attempt to switch protection off is reported to your console within minutes. We recommend keeping staff as standard Windows users.
PCs keep enforcing the last policy they received. If the agent itself stops, the browser extension still holds back card numbers, keys and main ID numbers on its own, and alerts the console.
Enterprise: in your own PostgreSQL database, with our servers only running the console. Trial: on our server, in a database of its own that no other company's console can open, until you move it to your own database or ask us to delete it. Either way, what employees write is never stored anywhere.
Windows 10 and 11 (and Windows Server 2016 or later) with Chrome, Edge or Brave, the browsers most companies standardise on. Talk to us if you need another platform.
Your console is ready the same day: we run it. The agent goes out with one command through the tool you already use, and PCs register themselves. Most teams then watch in monitor mode for a week before enforcing.
Try everything in Roger Shadelyt for 15 days, free, on as many PCs as you like. Tell us your work email and company; we set up your console and send you the sign-in within one business day. You'll see which AI tools your people use and what would have been stopped. If it's nothing, you've lost nothing.