Stop sensitive data before you hit send.
Prompt Guard watches the compose box on ChatGPT, Claude, Gemini, Copilot and Perplexity. The instant a secret, credential, customer record or internal identifier appears, it flags it — and can redact it in place before the request ever leaves your browser.
- On-device detection
- 40+ built-in detectors
- Works across 12 AI tools
2 sensitive values detected
AWS access key · Customer email address
Debug this upload script — creds are AKIA4YF2QX7NLEXAMPLE and notify dana@acme.io
- Built-in detectors
- 40+
- Typical scan latency
- <10 ms
- Supported AI platforms
- 12
- Bytes of prompt text sent to us
- 0
Built-in detectors
Typical scan latency
Supported AI platforms
Bytes of prompt text sent to us
One pasted config file is all it takes.
AI assistants have become the fastest way to debug a stack trace or rewrite a customer email. They are also the fastest way to leak a production key, an unreleased contract or a spreadsheet of real customer data — because the paste happens in one keystroke and there is nothing standing between it and send.
Secrets pasted into a chat are retained by the provider and may sit in logs well beyond the conversation.
Once sent, there is no recall — rotating a leaked key is the only remedy, and only if you noticed.
Policy documents do not stop a paste at 6pm on a Friday. A control in the compose box does.
What Prompt Guard actually does
Real-time scanning that catches secrets, credentials and PII in your AI prompts before they are ever sent.
Detection that runs before send
Prompt Guard inspects the compose box as you type and paste. The check completes locally in single-digit milliseconds, so the warning appears while you can still act on it.
40+ detectors out of the box
AWS, GCP and Azure keys, GitHub and Slack tokens, private keys, JWTs, database connection strings, credit cards with Luhn validation, emails, phone numbers, national IDs and more.
Your own rules
Add regex or keyword detectors for internal project codenames, customer identifiers, ticket formats or repository paths — whatever counts as sensitive at your organisation.
Redact, don't just warn
Replace a match with a stable placeholder in one click. The prompt stays coherent and useful to the model, while the real value never leaves the browser.
Your prompts never leave your machine
Detection is entirely local — there is no inference endpoint, so prompt text is never transmitted to ExtHive or anyone else. Paid plans do talk to our servers, but only for licensing, optional settings sync and Team policy; never for the text being scanned.
A record you can audit
A log of what was flagged and what you did about it, with counts by detector. It stays on your device on Free and Pro; Team plans sync detector names and counts — never the matched text — so an administrator can report on the control.
Three steps, then it gets out of the way
No onboarding tour and no configuration you have to finish before it becomes useful. An account only enters the picture if you buy a paid plan.
- 1
Install and pin
Add Prompt Guard to your browser. It activates automatically on supported AI sites. The Free plan works immediately with no account; paid plans are unlocked once by pasting the licence key we email you.
- 2
Type as you normally would
Detection runs locally against every keystroke and paste. Clean prompts are untouched and you will not notice it is there.
- 3
Resolve, then send
When something is flagged, redact it, allow it once, or add it to your allowlist. Your choice is recorded locally so you are not asked twice.
Prompt Guard plans
Start free and stay free if that covers you. Upgrade when you need the full detector set, custom rules or team policy.
Annual billing — save up to 20%
Free
For individuals who want the essential safety net on their own machine.
Free
Free forever. No card and no account required.
Install free1 device
- 20 core detectors — included
- 5 supported AI platforms — included
- Inline warnings before send — included
- Manual redaction — included
- Local allowlist — included
- Works fully offline, no sign-in — included
- Custom detector rules — not included
- Audit log & CSV export — not included
- Priority support — not included
Pro
Most popularFor professionals handling client data, credentials and source code every day.
$4.83/month
$58 billed annually
14-day free trial. Account required. Cancel any time.
Start 14-day trialUp to 3 devices
- All 40+ detectors — included
- Every supported AI platform, plus custom sites — included
- Unlimited custom detector rules — included
- One-click and automatic redaction — included
- Local audit log with CSV export — included
- Sensitivity tuning per detector — included
- Optional encrypted settings sync across your devices — included
- Priority email support — included
Team
Per seatFor organisations that need one policy applied consistently across everyone.
$4/month per seat
$48 billed annually
Per seat. Minimum 3 seats. Managed accounts. Annual invoicing available.
Talk to usUnlimited devices per seat
- Everything in Pro — included
- Centrally managed detector policies — included
- Shared organisation allowlist — included
- Seat management & SSO-ready deployment — included
- Organisation-wide audit export (metadata, never prompt text) — included
- Managed browser policy deployment — included
- Dedicated support channel — included
- Onboarding assistance — included
Prices are in USD and exclude any sales tax or VAT, which is calculated at checkout based on your location. Payments are processed by Paddle, our Merchant of Record. Paid plans require an ExtHive account to hold the licence; the Free plan does not.
14-day money-back guarantee
If Prompt Guard is not right for you, email us within 14 days of your purchase and we will refund it in full — no interrogation. Read the refund policy.
Nothing hidden in the permissions dialog
Here is every permission Prompt Guard requests, why it needs it, where your data goes, and whether any of it involves an account or our servers.
What it asks for, and why
- storage
- Stores your detector settings, allowlist and local activity log on your device.
- Host access to supported AI sites
- Lets the extension read and annotate the compose box on the AI platforms you enable. It is not granted for any other site.
- scripting
- Injects the scanner and the inline warning UI into supported pages.
- Host access to api.exthive.com
- Used by paid plans only, to activate your licence and — where you have enabled them — to sync settings or fetch your organisation's Team policy. The Free plan makes no requests to this host.
Where your data goes
Prompt text and detector matchesstays on your device
Scanned in memory inside your browser and discarded immediately after the scan. Never transmitted to ExtHive or any third party, on any plan, including Team.
Settings, custom detectors and allowlistsonly if you turn it on
Stored in local extension storage by default. If you enable Pro settings sync, they are encrypted and stored by ExtHive so your own devices stay in step. On Team plans, the administrator's policy and shared allowlist are fetched from ExtHive; your personal entries stay yours.
Activity logonly if you turn it on
Records the detector name, timestamp and action taken — never the matched value or surrounding prompt. Stays on your device on Free and Pro. On Team plans this metadata is reported to your organisation's audit export.
Account, licence and subscription dataonly if you turn it on
Paid plans only. Your email, licence key, plan and subscription status are processed to operate the licence. Payments are handled by Paddle as Merchant of Record; ExtHive never sees your card details. The Free plan creates none of this.
Accounts and what runs on our serversAccount for paid plans
The Free plan has no sign-up at all: no email, no identifier, no network calls to us. Pro and Team need an ExtHive account because a paid licence has to belong to someone — pasting the licence key we email you creates it, keyed to the address you purchased under. That account holds the licence, enforces the device limit, and is what makes optional settings sync and Team policy possible. Cancelling drops you back to Free rather than locking you out, and you can delete the account afterwards.
Licence activationPart of the plan
Paid plans check your licence key and subscription status against our servers so the paid features can unlock, and periodically thereafter.
Settings sync (Pro)Opt in
Off unless you turn it on. Copies your detector configuration, custom rules and allowlist between your own devices, encrypted in transit and at rest.
Team policy and seats (Team)Part of the plan
Your administrator's detector policy, shared allowlist and seat assignments are stored by us and distributed to the seats in your organisation.
Organisation audit export (Team)Part of the plan
Seats report which detector fired, when, and what the user did — never the matched value or the prompt around it — so administrators can report on the control.
Questions people actually ask
Still have a question?
We answer support email ourselves, usually within 1 business day.
Contact supportAdd Prompt Guard to your browser
Installs from the official extension store for your browser. It works straight away on the free plan; paid plans are unlocked once with the licence key we email you.
- ChromeComing soon
- EdgeComing soon
- BraveComing soon
- FirefoxComing soon
Try Prompt Guard today
The free plan needs no card and no account. Upgrade only when you hit a limit that matters to you — a paid licence adds an account, and nothing else changes about what stays on your device.