Founder life · open source · 2 min read
Next week the phone and laptop come along to the vacation house, but they stay in the cottage during the day. I want real days with my girlfriend and our two daughters without a screen on me. The only thing on my wrist is the Apple Watch, used as a dumb phone. The phone and laptop are there for actual emergencies, not for the reflex of picking them up.
That only works if MCP Pro keeps running without me. So I built a small monitoring layer that lives on the watch. It buzzes when something is actually on fire, and stays quiet otherwise.
Reactive monitoring is not monitoring
A customer pings you on Teams. You open the app to “just check.” A support email shows up on Monday morning. By the time any of those happen, the customer already noticed something was off. That is not monitoring.
I get why small teams don’t fix this. Datadog gets expensive fast. The Grafana plus Prometheus plus Alertmanager stack is a project in itself. Both are built for SRE teams of ten, not for a founder who would rather ship product on Monday. The real constraint for a one-founder ops setup is interruption fatigue, not lack of dashboards. The fix has to be small enough that you actually maintain it.
Our stack is two open-source tools on one small Hetzner VPS:
- Uptime Kuma: open source, self-hosted. HTTP, TCP, keyword and SSL checks.
- ntfy.sh: hosted, open source push notifications. Works on Apple Watch out of the box.
Both run in Docker on their own monitoring VPS, separate from production so a prod incident can’t take down its own alerts. Alerts go out in two tiers:
- Tier 1, production down or SSL expiring within 48 hours: ntfy, wrist buzz.
- Tier 2, staging blip or slow response: email via Brevo SMTP, read when I’m back at the desk.
The full setup is three small Hetzner instances now: production, staging, monitoring. Around €10 a month each. Uptime Kuma runs on the monitoring box and pings production and staging on a schedule. Anything critical fires through ntfy to my Apple Watch within seconds. Anything else lands in my inbox through Brevo. Staging itself runs end-to-end tests via GitHub Actions on every merge, so most regressions never reach production in the first place.
This isn’t a one-off choice. Our website runs on Hetzner. Our CMS is Directus. Analytics and funnel work happen in PostHog. The company itself runs on Odoo. All open source, all hosted in Europe. Monitoring fits the same pattern: pick simple tools you can understand, run them on infrastructure you own, and avoid building a second product just to watch the first one. A founder who is also the on-call engineer doesn’t need more dashboards. Two tools and two tiers cover it.
The vacation test
The phone and laptop stay in the cottage. The watch comes along.
If it stays quiet, MCP Pro is healthy and my daughters get my full attention. If it buzzes, I walk back, deal with it once, and then go back to building sandcastles. The phone and laptop are there if I really need them. In practice, they shouldn’t be. That is the whole point.
If you’re another Odoo partner or technical founder trying to actually take a week off, we’re happy to share the configs. Find us at pantalytics.com.