OTT-Cybersecurity-LLC/lyrie-ai

What it solves

Lyrie is an autonomous security agent designed to automate complex cybersecurity tasks that typically require manual expertise. It addresses the need for continuous, automated penetration testing, red-teaming of LLM endpoints, and a standardized way to verify the identity and trust of AI agents.

How it works

Lyrie operates as a multi-component system consisting of a Python CLI (lyrie-omega), a TypeScript SDK for the Agent Trust Protocol (@lyrie/atp), and a Rust-based security engine (Lyrie Shield). It uses a 7-phase pipeline (recon, fingerprint, scan, exploit, PoC, and report) for autonomous pentesting. For LLM red-teaming, it employs several adversarial attack strategies such as Crescendo and TAP. It also implements the Agent Trust Protocol (ATP), a cryptographic standard using Ed25519 signatures to manage agent identity and authorization.

Who it’s for

This tool is for security researchers, penetration testers, and organizations looking to automate their security audits, protect their AI agents from rogue behavior, and implement cryptographic identity for their AI workforce.

Highlights

• Autonomous Pentesting: A full 7-phase pipeline that works on live URLs and local source trees.
• LLM Red-Teaming: Built-in strategies like Crescendo, TAP, PAIR, GCG, and AutoDAN to test LLM endpoints.
• Agent Trust Protocol (ATP): An open cryptographic standard for AI agent identity and authorization.
• Self-Healing Loop: Automated detection, remediation, and revalidation of security vulnerabilities.
• Binary Analysis: Tools for static analysis, ROP gadget search, and SMT constraint analysis.
• Governance & Compliance: Built-in NIST AI RMF assessments and tool permission auditing.