Use AI freely.Keep sensitive data yours.
Protect sensitive information before your team shares it with ChatGPT or Claude. Inspection happens on-device. Policies and activity stay centrally managed.
14-day Team trial with five protection seats; card required. Then from US$5 per seat/month, or US$48 per seat/year. Five-seat minimum. Pricing
Chrome and Firefox extensions · Windows deployment agentSupported ChatGPT and Claude web paths. Windows clipboard protection is a separate control.
Try an example
Synthetic text · checked only inside this box
Can you tighten this reply to our supplier? Their contact is marta.lindqvist@example.com (flagged: Email address) and the delivery slipped to 14 March.
Review before sending
Email address found. Edit it, redact it, or continue if your policy allows.
An illustration of the decision flow, not a test of the live extension.
A useful prompt can contain information you didn’t mean to share.
A customer email in a draft
Asking for a calmer reply to a customer complaint.
“Rewrite this so it’s less defensive — it’s going to dana.reyes@example.com.”
Detects
Email address
The address is flagged; the rest of the request is untouched.
Next
WarnEdit the draft, redact the address to [EMAIL] after reviewing it, or continue if policy permits.
An internal name in meeting notes
Turning notes into a tidy list of action items.
“Summarise the Northwind Labs pricing review into next steps.”
Detects
Company name
Matches a name your administrator configured.
Next
WarnReplace the name, redact it, or continue where your team’s policy allows.
A credential while troubleshooting
Working out why an API request keeps failing.
“Why does this return 401? I’m sending DEMO_SECRET_1234567890ABCDEF.”
Detects
Credential / API secret
Matches a supported credential pattern (synthetic example).
Next
Block Can’t be overriddenRemove the secret, or review a redacted draft, then submit again yourself.
- Allow
- No matching rule prevents this submission.
- Warn
- Review the finding. Edit, redact where available, or continue if policy permits.
- Block
- This submission can’t continue unchanged. Remove the restricted information before trying again.
Blocks for keys, secrets and confidential markers can’t be overridden. Privelin never cleans up and sends a prompt on its own: redaction is reviewed first, and you choose when to submit again.
Where Privelin runs.
- Chrome extension
- Checks supported ChatGPT and Claude web submissions on the device before they’re sent.
- Firefox extension
- The same checks and policy, in Firefox.
- Windows agent
- Deploy browser protection, keep policies cached locally, and apply clipboard checks on Windows.
Clipboard checks aren’t full desktop-app protection, and the agent doesn’t read conversations inside AI desktop apps. See coverage
On-device protection. Team-wide control.
Set your policy, invite people, assign protection seats, see which installations are reporting, and review decisions without collecting your team’s prompts.
3 of 5 trial protection seats assigned
People
| Person | Membership | Role access | Protection | Installations | Actions |
|---|---|---|---|---|---|
| Amara Okaforamara@northwind.example | Member | Can manage | Seat assigned | 3 · one seat | |
| Jonas Weberjonas@northwind.example | Member | View only | Seat assigned | 1 | |
| Priya Ramanpriya@northwind.example | Invite pending | View only | Seat assigned | 0 | |
| Lucía Fernándezlucia@northwind.example | Member | View only | Unassigned | 0 |
Amara has Chrome, Firefox and a Windows agent linked — still one protection seat.
Access shown here is illustrative. It shows how restricted views look, not how permissions are enforced.
- What happened?
- Review decisions by category, site, actor and time.
- Who is protected?
- Invite people, set their access, and assign protection seats — one per person, however many installations they have.
- Which installations are reporting?
- See each extension and Windows agent, when it last reported, and the policy version it applied.
- What happens offline?
- Checks on supported paths continue on the device, using the cached policy.
Activity shows the decisions Privelin made on supported paths. It isn’t proof that every interaction was checked or that a breach was prevented, and a recent report doesn’t mean a device is protected right now.
Built around one boundary. Everything else follows from it.
- How it worksWrite as usual. When you submit through a supported path, Privelin checks the text on your device and offers only the choices your policy allows.See the three steps
- ProtectionWhat Privelin detects, how decisions and redaction work, and exactly which submission paths are covered.Read the coverage
- PrivacyYour prompts aren’t the reporting database. Activity reporting carries decision metadata, not message text.What we collect
Common questions
Short answers here; each page goes into more detail.
No. Prompts are checked inside the browser extension, and clipboard content is checked by the Windows agent — both on the device. Activity reporting records decisions — the category, the decision, where, who and when — not prompt text, clipboard content or the values that matched.
Account and billing details are kept separately from activity reporting. A message you’re allowed to send still goes to ChatGPT or Claude as usual.
What is sent to PrivelinPrivelin has extensions for Chrome and Firefox, covering supported submission paths on the ChatGPT and Claude websites.
Separately, the Windows agent can deploy the extension to Chrome, Edge and Firefox. Deploying to a browser isn’t the same as verified protection in it — the coverage table shows what has been verified.
See the coverage tableNo full desktop-app conversation interception is provided. Windows clipboard checks are a separate control and do not cover every data path.
No. Protection seats represent assigned people. Multiple linked installations do not automatically create additional seats.
Fourteen days, a payment card, and five protection seats. It automatically converts to the selected paid subscription, with a five-seat minimum.
See pricingYes. Detection is limited to the supported categories and patterns, so some sensitive text won’t match and some harmless text will. Review your rules from time to time, and keep encouraging people to think about what they share.
Know what leaves your browser.
Try Team for 14 days with five protection seats. Set your policy, connect installations, and review activity without collecting prompt text.
Start your Team trial