Ubuntu 24.04 LTS, everything in Docker, managed via Portainer
Power Consumption
12.72 kWh/month
No GPU. No NVMe. No ECC. A CPU older than some of the people in this sub. It just refuses to die.
It physically lives at my parents' place. I run and maintain the whole thing remotely from another city over a Cloudflare Tunnel, with Tailscale as a fully independent backup path. Zero ports exposed.
Fully automated, hands-off
This is the part I'm proudest of. The box runs itself:
Container updates are automated via Watchtower (rolling tags auto-pull, pinned versions stay put), with Telegram notifications on every update.
Backups run nightly with zero config: auto-discovery of every named volume and database, DBs dumped before snapshot, AES-256 encrypted, pushed off-site to Cloudflare R2, 4-day retention.
Security: weekly Trivy CVE scan across every image.
Add a new stack and it gets backed up automatically, no changes needed.
I haven't physically touched the box in months and rarely have to log in at all.
What it's actually running (20+ compose stacks, 46 containers)
Infra & ops
Portainer, Pi-hole, Cloudflared tunnel, Tailscale, Watchtower, Trivy CVE scanning (0 CRITICAL/HIGH across every image)
Automated nightly backups of every volume and DB, AES-256 encrypted, pushed to Cloudflare R2
Monitoring
Grafana, InfluxDB, Telegraf
Speedtest Tracker (hourly)
Smart home
Home Assistant, Zigbee2MQTT, Mosquitto. The entire house is run through Home Assistant: cameras, Zigbee devices, automations, the lot.
Finance
Firefly III (plus MariaDB, custom importer, auto-classifier, recurring cron) with rule-based merchant normalization and 14-category auto-tagging
Ghostfolio for the investment portfolio
Automation & AI
n8n
Evolution API (WhatsApp gateway)
Self-hosted AI email triage agent: pulls my Proton Mail over the Proton Mail Bridge (IMAP) into n8n, then through a 5-tier LLM fallback chain (Groq Qwen3/Llama, then Claude Haiku). It classifies and files mail, sends WhatsApp digests, and learns from manual corrections.
AI Garmin analysis: family fitness data syncs into InfluxDB, then n8n runs an LLM post-activity analysis after every workout plus a weekly AI Garmin report, delivered over WhatsApp.
Langfuse self-hosted (web, worker, Postgres, ClickHouse, Redis, MinIO) for full LLM observability. Yes, on the Haswell i5.
Hindsight (AI notes), Outline wiki, n8n MCP
Plus pgAdmin, IT Tools, Webtop (browser desktop), Homepage dashboard.
The flex
46 containers on a 12-year-old quad-core: 6 Postgres instances, MariaDB, ClickHouse, InfluxDB, multiple Redis, and a headless Chromium renderer all running at once, still 15% disk with RAM to spare. Fully automated updates, backups and scanning, monthly cloud cost for the AI pipeline ~$0.30 (Groq free tier does the heavy lifting), and I manage the whole thing remotely.
Old enterprise mini-PCs are the most criminally underrated homelab hardware. AMA.