Redaxa

Catches personal data and secrets in your prompt before you send it to ChatGPT or Claude.

Review sensitive information before sending a prompt to an AI assistant. Redaxa adds a prompt-checking workflow to supported ChatGPT, Claude, Gemini, Copilot and Perplexity pages. It flags potential personal information, recognizable credentials, financial identifiers and your own protected terms, then offers a redacted version for you to review. Detection includes: - Email addresses, phone numbers and supported name/address patterns. - Recognizable API keys and tokens, JWTs, private keys and labelled secrets. - Card-number candidates checked with Luhn and IBAN candidates checked with mod-97. - Custom terms such as client names and project codenames. New in 0.4.0 A redesigned popup separates Prompt protection and Repository check, with clearer account status and plan access. Repository check with Pro and Business Paste a public GitHub repository URL in the extension to open the Redaxa workspace and launch Repository check in Redaxa 0.4.0 or later for Windows. The Windows app is required and installed separately; the extension does not scan repository contents inside Chrome. A separate workspace or desktop sign-in may be needed. The Windows app helps you review potential exposed API keys, other secrets and sensitive data, with a severity summary, grouped repeated findings, scan progress and an estimated remaining time. Values are hidden by default and can be revealed individually or by section. Export a redacted text report or prepare an email draft for review. Coverage and any scan limits are shown in the result. Pro combines prompt protection, custom protected terms and access to the Windows repository workflow. Business adds shared protected terms and workspace policies for teams. A Redaxa account and an active eligible trial or subscription are required. Compare current plans: https://promptshield-beta.vercel.app/ How scanning works Your prompt is sent to Redaxa's backend for scanning. It is not processed entirely on your device. The scanning logic does not send your prompt to a third-party AI provider. Scan audit events contain metadata, such as detection categories and policy decisions, rather than prompt content. Repository scanning runs locally in the Windows app. See the privacy policy for data-handling details: https://promptshield-beta.vercel.app/privacy.html For Business workspaces Workspace administrators can configure shared protected terms and per-category policies. Block policies prevent sending through the extension's protected workflow until the detected issue is addressed. Detection uses patterns and category-specific checks. It can miss sensitive information or flag harmless text; a valid checksum does not prove a match is genuine. Repository findings are potential exposures, not confirmed vulnerabilities. API key validity is not tested. This is a review aid, not a guarantee that every sensitive item or sending route is covered. Review the result before sending or sharing. Get started and compare plans: https://promptshield-beta.vercel.app/ Windows download and documentation: https://github.com/AurelioAvila/redaxa