SwarmSecurity

Multi-persona Solidity auditor using routed LLM swarm, consensus scoring & code-snippets.

SwarmSecurity

Created At

ETHGlobal Buenos Aires

Winner of

Nethermind

Nethermind - AgentArena: Build the Next Generation of Security Agents

Project Description

Swarm.Security is an advanced AI-powered security audit agent designed to automatically detect vulnerabilities in Solidity smart contracts. The system employs an intelligence architecture of multiple agents; aka a "swarm" - where multiple specialized AI personas work collaboratively to analyze code from different security "expert" perspectives.

Unlike traditional static analysis tools, Swarm.Security uses Large Language Models (LLMs) to understand contract logic, identify complex attack vectors, and provide detailed explanations of potential vulnerabilities.

The system is built with extensibility in mind, allowing new security expert personas to be easily added to the swarm. These personas can be easily updated and enhanced with the latest attack vectors leveraged use by real world threat actors.

How it's Made

Swarm.Security was made specifically for the Nethermind Agent Arena; a platform that evaluates AI agents for their ability to find security vulnerabilities in smart contracts. The agent receives tasks through a webhook and processes them automatically.

Swarm.Security has two modes of operation:

  • Server Mode — Receives audit tasks via webhook from the Agent Arena
  • Local Mode — Directly audits GitHub repositories for testing and development

The platform sends notifications when new audit challenges are available, and Swarm.Security analyzes them using a set of specialized personas. These custom personas or smart contract experts are intentionally designed to be “hack-able. They have been validated against real-world exploited smart contracts

background image mobile

Join the mailing list

Get the latest news and updates