Show HN: Tokenhush – keeps your secrets out of what Claude Code sends (github.com)

Tokenhush has emerged as a crucial tool for developers working with AI coding platforms by offering a local, reversible secret redaction mechanism. It functions as a loopback-only HTTP gateway that intercepts requests from AI tools before they reach the vendor API, replacing detected secrets with placeholders while allowing developers to receive the original values in the response. This is significant because it safeguards sensitive information such as API keys or email addresses from being sent out insecurely, mitigating risks associated with unintentional data exposure during AI-assisted coding tasks. Key technical features of Tokenhush include six built-in detectors capable of identifying various types of secrets—such as API keys, JSON web tokens, and credit card numbers—and replacing them with session-scoped placeholders during transmission. Notably, Tokenhush does not rely on root certificates or man-in-the-middle techniques, making it a secure solution that binds solely to the local machine. Its simplicity and portability, developed in pure Go, allow easy installation across multiple operating systems without requiring admin rights. The ability to customize detection rules and seamlessly integrate with various AI coding tools enhances its utility within the AI/ML community, ensuring that developers can work efficiently while maintaining strict confidentiality of their secrets.