Panguard AI

AI-powered security for agents and servers — detect threats in real-time, on-device, with hundreds of open-source ATR detection rules (updated continuously).

Features

• Guard Daemon — Real-time endpoint monitoring; blocks high-confidence threats (critical / high-stable) in the default guarded posture and advises on the rest (--enforce to block on every matched verdict, --advisory to detect-only). OS-level responses (block IP / isolate file / kill process) are off by default and require arming + elevated privileges.
• Local Dashboard — Zero-telemetry security control center running on your machine
• Hundreds of Detection Rules — ATR (Agent Threat Rules), a community-driven, LLM-reviewed open standard, updated continuously
• MCP Proxy — Transparent tool-call evaluation for Claude, Cursor, and other AI agents
• Skill Auditor — Pre-deployment security audit for AgentSkills and OpenClaw
• 100% Free & Open Source (MIT) — Every Community feature is free forever. Cryptographic rule signing and SOC 2 evidence are Enterprise-only.

Quick Start

npm install -g @panguard-ai/panguard
# or run without installing:
npx @panguard-ai/panguard

# Start the guard + dashboard
pga up

# Scan your skills before deploying
pga scan

# Run interactive menu
pga

What It Does

1. Guard — Runs as a daemon, monitoring for malicious agent behavior (prompt injection, data exfiltration, unauthorized access)
2. Dashboard — Local web interface showing live threats, rules, and enforcement policies
3. Detection Rules — Based on ATR (Agent Threat Rules), an open standard with hundreds of continuously-updated rules
4. MCP Integration — Proxy tool calls through ATR before execution
5. Skill Auditor — Analyze SKILL.md files for vulnerabilities

Links

• Website: https://panguard.ai
• ATR Standard: https://github.com/Agent-Threat-Rule/agent-threat-rules
• GitHub: https://github.com/panguard-ai/panguard-ai
• License: MIT