Show HN: InBox Wall – A local-first, open-source email sanitizer for AI agents (github.com)

InBox Wall, a local-first open-source Python application, has been released to sanitize incoming email HTML before it's processed by AI assistants. This application is significant for the AI/ML community as it addresses a critical vulnerability: unfiltered email HTML can contain hidden instructions that manipulate AI behavior, potentially leading to security breaches. InBox Wall removes harmful components like invisible characters, prompt injection patterns, and hidden text through a multi-stage sanitation process, ensuring that the AI only processes safe text. Key technical features include its ability to connect to email via IMAP in a read-only mode, a thorough sanitation pipeline that evaluates CSS and HTML for malicious content, and dual-database isolation for safety. It enforces strict resource limits and privacy controls, while providing a local REST API for AI tools. Unique mechanisms like background-aware detection and inheritance-aware parsing help prevent attacks disguised within benign-looking emails. With privacy-focused design, InBox Wall serves as a valuable defense-in-depth layer for safeguarding AI email pipelines against misuse, highlighting the need for enhanced security measures in AI applications.