Panguard AI is an open-source, self-hosted security platform designed to provide real-time protection and audit-ready evidence for AI agents. It addresses the unique security challenges posed by AI agents, such as prompt injection, poisoned tools and skills, malicious MCP servers, and over-autonomy, which traditional security standards and tools do not adequately cover. The platform offers a Skills Audit that inspects every skill before execution, detecting both known threats using community-contributed Agent Threat Rules (ATR) and unknown threats through AI-driven analysis. When new threats are identified, new rules are shared across the community to enhance collective protection.
The tool operates with a local-first, deterministic engine that runs offline, ensuring zero telemetry and privacy for its users. It can be installed with a single command and does not require any signup, enabling rapid deployment—typically within 30 seconds. Panguard AI automatically detects a wide range of AI agents and tools, including Claude, Code, Claude Desktop, Cursor, Hermes Agent, OpenClaw, Codex CLI, WorkBuddy, NemoClaw, ArkClaw, Windsurf, QClaw, Cline, VS Code Copilot, Zed, Gemini CLI, and Continue Roo Code. The platform scans every skill and MCP server, guarding agent actions at runtime and providing signed, audit-ready output.
Panguard AI leverages the ATR rulebook, an open, executable standard for AI agent security, currently comprising 747 executable detection rules. These rules are designed for deterministic, local execution and are crosswalked to ten recognized frameworks, including the EU AI Act, NIST AI RMF, ISO/IEC 42001, OWASP Agentic, OWASP LLM, MITRE ATLAS, SAFE-MCP, CWE, Five Eyes, and CISA KEV. The platform also features an Agentic Threat Detection (ATD) taxonomy, which serves as a knowledge base for attack techniques and is mapped to MITRE ATLAS and OWASP.
The platform is MIT-licensed, free to use, and open source. Its ATR rules have been adopted and shipped in production by organizations such as Microsoft and Cisco. Panguard AI is positioned as a security tool for developers, enterprises, and anyone deploying or managing AI agents who require robust, up-to-date protection against evolving threats in the agent ecosystem.
In the Security & compliance platforms space, Panguard AI takes a focused approach. It helps developers secure AI agents by detecting prompt injection, tool poisoning, and supply chain threats using open ATR rules. It is built as an open-source project for ai developers. Panguard AI is open source under the MIT license. Panguard AI is available on the command line.
Behind Panguard AI is Panguard AI, and the product first shipped in 2026. Development happens publicly on GitHub with 50 stars and 425 commits in the last 90 days. Key capabilities include agent skill scanning, prompt injection detection, and supply chain threat detection. The interface is available in English and Chinese.
Latest indexed changes and source events
Other apps tracked under the same category.
