Kavach - Tactical Zero-Trust OS Firewall for Local AI Agents

Hi Product Hunt! 👋 I’m Akshay, the builder behind Kavach. Like many of you, I’ve been building and experimenting with local LLMs and autonomous agents. But the anxiety of giving an AI unrestricted read/write access to my file system was holding me back. One hallucinated script could wipe out a src directory or corrupt my configs. I couldn't find a good "emergency brake," so I built one. Kavach (Sanskrit for "Armor") is a tactical security layer that intercepts system calls at the OS level. It doesn't just monitor logs; it actively stops unauthorized changes. Here is what makes it tick: 👻 The Phantom Workspace: If an untrusted agent tries to modify a file, Kavach redirects the action to a hidden virtual directory. The AI thinks it succeeded, but your real code is safe. ⏳ Temporal Rollback: If a file gets mangled, our cryptographic micro-cache lets you one-click "time-travel" it back to its exact state from milliseconds prior. 🧠 The Turing Protocol: To stop multimodal models from using RPA to just click "Approve" on firewall alerts, the UI uses adversarial noise patterns to blind AI OCR. I built the backend in Rust for near-zero latency, paired with a Tauri + React frontend. It runs 100% locally with zero cloud dependencies. If you are tinkering with AI agents, I’d love for you to download the v1.0 (available for Windows & macOS), protect your perimeter, and try to bypass the Phantom Workspace. I'll be hanging out in the comments all day, let me know what you think, what features you'd want next, or if you just want to talk Rust and local LLMs! 🛡️