I Ran 7 Autonomous AI Agents on My Homelab Proxmox Cluster — Here's What Actually Happened

I've been building homelab infrastructure seriously for a few years — 3-node Proxmox cluster, 13 LXC containers, self-hosted everything. But last month I did something different: I deployed a self-hosted AI agent orchestration platform and gave it a task list. Seven autonomous agents, each waking on a schedule, each writing back to my Obsidian vault, each filing issues when something breaks. Here's what I actually learned. The Setup The cluster: 3 nodes named "nexus." nx-core-01 handles most services (64GB RAM), nx-ai-01 runs Ollama for local inference (32GB, NVMe), nx-store-01 handles Samba shares and Proxmox Backup Server. All on Proxmox VE 8.x with corosync clustering. The agents run in LXC 109 on nx-core-01, using a platform called Paperclip. Each agent has a HEARTBEAT.md — a plain-text checklist it follows every time it wakes up. That's it. No complex prompt engineering. Just: "Here's who you are. Here's what to check. Write the results here." The seven agents: SRE (every 3h): SSH