What this is

A Claude agent that lives in our Proxmox cluster.

It reaches the whole estate, runs real work, and remembers everything across sessions. Here's what it manages — and what's been built on it.

Where it runs

monolith · CT 114 — Claude Code on the cluster

The cluster

3-node Proxmox + Ceph — DaveBowman · HAL9000 · DiscoveryOne

It operates

Proxmox · Plex · Dispatcharr · networking · backups · cameras

Built on it

HomeDash · ClearDash

Proxmox + Claude

Watch it work · a real incident

A live channel died mid-World-Cup. No human touched it.

root@dispatcharr — journalctl -u dispatcharr -f

19:59:21 ERROR ts_proxy [in#0/mpegts] demuxing: Input/output error

19:59:21 WARN ts_proxy Server closed connection — cf.goldpremium.online

19:59:21 INFO ts_proxy trying next stream for channel: Telemundo

19:59:21 WARN ts_proxy No alternate streams available — Found 0

19:59:21 ERROR ts_proxy channel state → ERROR ← Plex quits to guide

▸ diagnosed: single upstream feed, zero failover

▸ audited all 464 channels → 365 had no backup feed

▸ added failover to Telemundo + 7 priority channels

✓ deployed · smoke passed · 22 pages render clean

Built on it · homelab

HomeDash — a live ops dashboard it built and deploys itself.

homedash.manselmo.com

▢ HomeDash — Overview

screenshot drop-in

Real-time view of the whole estate. Commit → deploy → smoke-test is one command; the agent ships its own changes and verifies every page renders.

Built on it · work

ClearDash — a ClearFuze ops tool it built.

cleardash.manselmo.com

▢ ClearDash — Overview

screenshot drop-in

cleardash.manselmo.com/aia

▢ AIA Exhibitors

screenshot drop-in

Sourced and mapped conference leads at AIA — and ran the Pax8 Beyond CTF on its own.

ClearDash · the CTF, autonomous

8th of 57 teams. 46 solved. Zero human input.

It remembers

Persistent memory of the whole estate — across every session.

~/memory/ MEMORY.md

› proxmox-cluster — 3-node Ceph; CT rootfs on PodBayDrives; VMIDs unique cluster-wide

› caddy-reverse-proxy — CT 135; *.manselmo.com via DNS-01 + dynamic_dns

› dispatcharr-failover — 464 channels; added failover to 8; account 1 is the standing gap

› pbs-backup-pipeline — vzdump → PlanB → hourly sync → PlanC archive

› homedash · cleardash — project locations, deploy paths, design system

…it doesn't start from zero. It picks up where it left off.