Hrolbot
Multi-agent AI platform running on a Proxmox VM — autonomous Claude Code workers coordinated through an MCP mesh for code review, builds, testing, and system operations.
Hrolbot is what happens when you give a VM too much responsibility. It started as a Discord bot calling Claude Code and evolved into a full multi-agent platform where specialized workers — frontend, backend, reviewer, testing, research, devops — operate autonomously on separate Linux user accounts.
Workers coordinate through a shared MCP message mesh, write code on feature branches, review each other's work with adversarial prompting, and push to GitHub. The orchestrator receives tasks from Discord or CLI, delegates to the right specialists, and triages results from dual-dispatch builds (Sonnet + Codex in parallel).
The whole system runs on a 4-core Proxmox VM with 8GB RAM and a 150GB SSD. Workers use tmux sessions for long tasks, SQLite for state, and a second-brain knowledge base for continuity across sessions. It manages itself — heartbeat monitoring, session wrap-ups, worker health checks, and automatic recovery from failures.
Case Study
The Problem
Running a growing homelab with dozens of services, plus building multiple software projects, was creating more operational overhead than one person could handle efficiently.
The Approach
Built an autonomous AI agent platform where specialized Claude Code workers handle different aspects of development — code writing, reviewing, testing, DevOps — coordinated through a message mesh with stigmergic handoffs.
The Outcome
A system that autonomously handles code reviews, dual-provider builds, system monitoring, and operational tasks — while maintaining conversation continuity and learning from each session.