What is prompt-gate?

Prompt Gate is an open-source desktop security tool that prevents sensitive information from being accidentally sent to AI tools. It runs entirely on your own machine and operates at two layers: it blocks access to AI services your organization has not approved by intercepting their domain names before any connection is made, and it inspects text you are about to send to approved AI tools to detect secrets, credentials, or personal data before they leave your device. The inspection engine is designed to catch obfuscated secrets that would fool a basic pattern scanner. It normalizes text before checking it, which means it catches secrets where someone has swapped in visually identical characters from other alphabets, added invisible Unicode characters, or encoded the value in base64. The scanner then scores each match against where the content is going, since a credential pasted into a code repository is more concerning than the same string in a comment. Users can whitelist specific known-safe values without sending anything to a remote service. Content is captured through a browser extension for Chrome, Firefox, and Safari, or through an optional local proxy for traffic from applications outside the browser. The tool appears in the system tray and is designed to use minimal CPU and memory. Privacy is a stated priority: the only things the tool writes to disk are configuration settings, domain block-lists, and simple integer counters of how many scans and blocks occurred. No URLs, domain names, user identifiers, or specific event records are ever stored. Prompt Gate works on Windows, macOS, and Linux. The repository includes a standalone command-line scanner for testing the detection engine directly, benchmark results over a million lines of source code, and documentation covering the detection algorithm in detail. The project is MIT licensed.