VibeGuard

VibeGuard is an open source proxy that sits between a developer's AI coding assistant and the model API, redacting sensitive data from outgoing requests and restoring it in the responses. It works as a man in the middle HTTPS proxy: a secret such as an API key, an internal server address or a project name is swapped for a placeholder token before the request leaves the machine, and the original value is put back when the answer returns, so the assistant keeps working while the real values never reach the model provider. It is aimed at developers and teams who use tools such as Claude, Cursor and Codex and want to keep credentials and proprietary names out of prompts. Detection is layered: rule lists, keyword matching, regular expressions and an optional named entity recognition step. The proxy only intercepts traffic from the processes it launches, so the rest of the system network is untouched, and a command line launcher starts a chosen AI tool through it. A local CA certificate must be trusted for interception to work. A browser based admin dashboard provides a real time audit log, visual analytics and rule management, and rule changes apply without restarting the proxy. Keywords and placeholder mappings are stored encrypted using the local CA key. Installers are provided for macOS, Linux and Windows, and the site is available in English and Chinese. The project is published as open source with its code on GitHub. The site has no pricing page and lists no paid plans.