Check a prompt for sensitive information and suspicious instructions
Paste a prompt or draft to look for supported personal-information patterns, common secrets such as API keys, and a limited set of suspicious instruction patterns. Review a redacted text preview before deciding what to share.
How the prompt checker works
- Paste text. Check up to 15,000 characters of a prompt or draft.
- Give consent. The text is sent to OpenWebStore’s backend for scanning; it is not sent to an AI provider.
- Review findings. Check the detected categories, inspect the redacted preview, and decide whether it is appropriate to share.
The scan uses the open-source turbocloak package, not an AI model. OpenWebStore processes the scan in memory and does not persist the text or report.
Frequently asked questions
What does the prompt safety checker look for?
It scans pasted text for supported patterns such as email addresses, phone numbers, SSN-shaped values, common API-key or secret formats, and some suspicious instruction patterns. The exact categories depend on turbocloak's detection rules; names and street addresses are not detected.
Does it use AI, and where does my text go?
No AI model is used. After you check the consent box and start a scan, the text is sent to the OpenWebStore backend for an in-memory turbocloak scan. The request and result are not stored by OpenWebStore, and the text is not sent to an AI provider.
Does the redacted preview remove prompt injection instructions?
No. The checker can flag suspicious instruction patterns, but visible instructions remain in the preview. Review and remove or rewrite them yourself before sharing text with another system.
Does a scan prove that text is safe?
No. This is a heuristic check, not a security audit or guarantee. It may miss names, street addresses, custom or unsupported credentials, sensitive context, and new instruction-attack patterns. It can also report false positives and remove invisible formatting characters from the preview. Always review the original and redacted text yourself.