Summary

Project Muteki is an autonomous, heterogeneous multi-model AI agent swarm designed to overcome single-agent limitations in cybersecurity tasks. It coordinates agents on a shared "blackboard" for collaborative problem-solving in CTFs, penetration testing, and code auditing, offering enhanced efficiency and success rates for security professionals.

How It Works

Muteki orchestrates a swarm of diverse AI agents (e.g., Claude Code, Codex, Cursor) using a shared "blackboard" for evidence dissemination and coordination. A core loop, running approximately every 2 seconds, involves a coordinator planning, workers executing commands, and results feeding back into the blackboard. This shared-evidence approach prevents dead ends and maximizes collaborative problem-solving efficiency.

Quick Start & Requirements

Initiate with ./init.sh for setup and tests, then launch the command deck via ./run.sh web. Prerequisites include Python ≥ 3.13 (via uv), Node.js (UI), Go ≥ 1.26 (supervisor), and Docker. Users must also install and authenticate proprietary engine CLIs (Claude, Codex, Cursor), each with distinct licensing and data policies. macOS is the primary tested OS; Windows Compose mode is unvalidated.

Highlighted Details

Muteki achieved a 100% solve rate on the 200-challenge NYU CTF Bench test set within a 30-minute budget per challenge. It autonomously solved all challenges in a 3-hour RIFFHACK 2026 CTF and secured first blood on the iChunQiu Yunjing penetration-testing range in 2 hours. The architecture supports broad goal-driven multi-agent collaboration beyond CTFs.

Maintenance & Community

The project is open-sourced, encouraging issue reporting and community contributions. A discussion group is available. Future plans include authentication, container mode optimization, expanding agent support, and developing a TUI mode.

Licensing & Compatibility

Licensed under GNU AGPL-3.0, a strong copyleft license requiring derivative works to be shared under the same terms, potentially impacting integration with proprietary software.

Limitations & Caveats

As an offensive security tool, Muteki executes commands and requires use in disposable environments due to its lack of malicious challenge isolation. It is primarily tested on macOS, with Windows Compose mode unvalidated. Reliance on proprietary third-party CLIs necessitates adherence to their specific policies. The TUI mode is unfinished.