Tactical AI Workspace Monitor & EDR

Kavach is an open-source Endpoint Detection and Response (EDR) tool designed to monitor, restrain, and remediate autonomous AI agents and local LLMs operating on a user's machine. It addresses the inherent volatility of AI agents that can cause accidental data loss, code corruption, or sensitive information exfiltration. Kavach provides a crucial safety layer for developers and power users running AI models locally, offering immediate remediation and forensic capabilities.

How It Works

Kavach operates as a high-performance, asynchronous file system observer using Userland APIs, built with Rust for the core engine and React/TypeScript for the UI, bridged by Tauri. Its architecture prioritizes local execution with zero cloud dependencies for maximum performance and privacy. It employs application-layer defenses, such as intercepting destructive file operations into a "Phantom Workspace" and automatically terminating unresponsive agents after a 60-second delay via the "Auto Enforcer." While not a strict synchronous sandbox in its current version, it offers features like cryptographic file caching for temporal rollback and honeypot tripwires to detect malicious activity.

Quick Start & Requirements

Zero-config deployment is achieved by downloading the latest installer for your operating system from the Releases Page and running the executable. Windows requires running as Administrator, and macOS requires Full Disk Access. After launching, users select their workspace directory in the UI, and Kavach immediately begins monitoring. The project is built using Rust, React, and Tauri.

Highlighted Details

• Phantom Workspace: Intercepts and quarantines destructive file operations (Modify, Delete, Create).
• Auto Enforcer: Automatically issues an OS termination signal to agents if intercepted actions remain pending for over 60 seconds.
• Temporal Rollback: Provides one-click restoration for files under 50MB via cryptographic caching of modifications.
• Honeypot Tripwire: Deploys decoy files (e.g., system_auth_tokens.json) that trigger alarms upon access.
• Gag Order (PII Sanitizer): Scans outbound strings in real-time for sensitive data like API keys, credit cards, and emails.
• Faraday Clipboard Guard: Overwrites clipboard contents with a decoy if high entropy (potential secret) is detected.
• Cryptographic Ledger: Ensures audit logs are immutable using an FNV hash chain.
• Supply Chain Auditor: Scans package.json for known malware dependencies in real-time.

Maintenance & Community

The project is developed by Akshay Sharma. Users can report anomalies or suggest features via the in-application "Direct Line" communication module, which routes encrypted feedback directly to the developer's email.

Licensing & Compatibility

Kavach is licensed under the GNU General Public License v3.0 (GPLv3). This license requires that all forks, distributions, and derivatives remain open source under the same terms, and permanent attribution to Akshay Sharma is mandatory.

Limitations & Caveats

Kavach v1.1.0 is an EDR monitor, not a strict synchronous sandbox; true system call blocking requires kernel drivers slated for the v1.2 roadmap. Large files (over 50MB) are dynamically flagged as ineligible for Temporal Rollback. Elevated privileges (Administrator on Windows, Full Disk Access on macOS) are necessary for full functionality.