ConcoLLMic: Agentic Concolic Execution

Language- and Theory-Agnostic Concolic Execution via LLM Agents
Code Paper

ConcoLLMic is an AI-powered concolic testing engine that helps developers detect elusive software vulnerabilities.

  • 🌐 Native multi-language support: Works on your favorite languages out of the box. Even for polyglot systems!
  • 🧩 Goodbye, environment models: No more warnings from unmodelled functions and environments.
  • 📖 Readable insights: Human-friendly constraint summaries and reproducible test cases.
  • 🚀 Covers more, and faster: In ~4h, ConcoLLMic achieves 230+%/80+% higher coverage than KLEE/AFL++ in 48h.

Quick Links

 Documentation

Complete guide for using ConcoLLMic, including installation, instrumentation, and concolic testing.

  • Installation & System Diagram
  • Examples & Quick Start
  • User Guide

 Support

Get help with troubleshooting, FAQ, and License for ConcoLLMic.

  • Troubleshooting Guide
  • Frequently Asked Questions
  • License Information

BibTeX

@inproceedings{ConcoLLMic,
    title={{Agentic Concolic Execution}},
    author={Luo, Zhengxiong and Zhao, Huan and Wolff, Dylan and Cadar, Cristian and Roychoudhury, Abhik},
    booktitle={2026 IEEE Symposium on Security and Privacy (SP)},
    pages={1--19},
    year={2026},
    publisher = {IEEE Computer Society}
}