FractalSwarm Uses Multi-Agent Hierarchy to Automate Production Incident Response

A prototype system called FractalSwarm has been demonstrated on DEV Community, showcasing a multi-agent approach to site reliability engineering (SRE) in production environments. The system uses a three-tier agent hierarchy — Incident Commander, Domain Specialist, and MCP Log Parser — to diagnose and resolve infrastructure incidents across domains such as disk, database, and network. To reduce costs and API load, a WebAssembly engine pre-clusters raw alerts using Hamming distance calculations before any LLM processing occurs. Multiple incidents can be handled concurrently through parallel asynchronous workflows, while a human-in-the-loop approval gate requires engineers to review and authorize all mitigation commands before they execute on live systems. The project is acknowledged as a prototype still requiring refinement, but its authors argue it demonstrates a viable pattern for combining structured agent hierarchies with deterministic pre-filters and strict safety controls.