The problem you find out about too late
A boiler can stop on Friday evening — and you find out on Sunday, arriving at a cold house. A ventilation filter clogs up over months — and you only notice when the air turns stale. An ordinary home keeps its problems to itself until the last moment.
In our country house project we built a monitoring system that works like a dispatch centre: the home continuously watches its own engineering systems and messages the owner on Telegram when something goes wrong.
What's being watched
- Boiler room — two boilers: supply and return temperature, burner status, boiler connectivity, alarm signals. Plus eight heating circuits, each with its own sensor
- Ventilation — the air handling unit reports not just "on/off" but decodes 24 distinct alarm and warning types: from a clogged filter to a frost risk on the heat exchanger
- Climate and humidity — per-zone readings and deviations from targets
- Internet and connectivity — the home checks its own network access; if the provider goes down, you're the first to know
- The automation server itself — the system watches its own health too: load, temperature, free disk space
What an alert looks like
A Telegram message arrives with a severity level:
🔴 ALARM — Boiler 1: fault
🟡 Warning — AHU filter: clogging
⚠️ No data — Humidity sensor: connection lost
And when the problem is resolved — a confirmation follows:
🟢 Normal — Boiler 1: connection restored
One detail worth noting: the system notices not only bad readings but silence. If a sensor simply stops reporting — that's an event worth knowing about too. A broken sensor that "says nothing" is more dangerous than a sensor reporting a problem.
Why it doesn't turn into spam
A naive monitoring system floods you with notifications: a temperature flickers at the threshold — and dozens of messages pour into the chat. Here, a transition into an alarm state must hold for a set confirmation time before the system accepts it as real. Short spikes and jitter get filtered out — only what genuinely needs attention reaches the owner.
History — not just "now" but "how it was"
All readings and events are written to a local database and kept for 90 days. That answers questions that would otherwise stay guesses: how did the supply temperature behave all week? When exactly did the filter start clogging? How many times did the internet drop this month?
The technical side
Monitoring runs locally, on the automation server inside the home — no cloud subscriptions. Events and time series live in SQLite on site; notifications go through Telegram. A failure in one data collector never takes down the others: every subsystem is isolated.