Server-born alerts and emergency response what the platform does when you are not looking

"Will it warn me?" is the question that falls to the bottom of the list when people choose a system — and it is the right one. The answer in HoumeOS now has two parts: alerts born on the server, and protective actions the server takes by itself.

Seven things the server watches

The alerts are not a list of events but checks that run on the server at an interval:

• Silent telemetry — a device that stops reporting is a problem, not a normal state.
• Transport down — if the link to MQTT disappears, you want to know before you feel it in the buttons.
• Devices offline — the same status grouping you see in the Devices screen.
• Failed commands — somebody pressed, nothing happened.
• Failed PIN attempts — on a protected entrance you can see who tried and when.
• Flat batteries — sensors whose reserve is running out.
• The Matter controller — whether the network the paired devices ride on is alive.

Each alert fires on a rising edge (not on every check), has a cooldown so it does not rattle on, and appears in the panel with acknowledgement on the screen. Critical items show on every screen that is on in the house, not only the one you are standing in front of.

On a hazard the server does not wait

More important than the alert is the action. The protective rules are stored on the server and run with no cloud and no human:

• Leak — closing the water solenoid, but only on four confirmed conditions; the main valve is a critical command with confirmation, because the mistake in the other direction costs more.
• Gas or CO — stopping the burner.
• Smoke — stopping the ventilation so it does not spread what it should not.
• Frost — heating toward 21°C so a pipe does not burst.
• Night flow — if the meter measures a flow when nobody uses water, a warning arrives before it becomes a leak.

The same goes for motion: gates and garage doors report travel time, doors report an unlock history, and cameras report last motion and a last frame. This is not a "clever alarm" — it is a journal you can check a claim against.

Why local matters here

If the rules live in the browser, they stop when you close the tablet. Here the logic, the rights and the journal sit on the server in the LAN — which is why both the alert and the protection hold at 3 a.m. with the internet down. How that is built is described in Local-first smart home without cloud, and the rights and the PIN in Access security: roles, PIN and audit.

An honest note

Today: the alert engine, the panel with acknowledgement, the protective actions and the journal all work in the platform and in simulation. Not yet: delivery to the phone (the email, push and webhook channels await a decision), video recording to an NVR, doorbells and temporary codes for guests. The current order is in The roadmap; the general view of hazards in A clever alarm and emergency scenarios.