On-call

On-call answers one question: when a critical alert fires at 3 a.m., who gets paged, and who is paged next if that person does not react? It is built in, free, and needs no extra service. You describe who (contacts), when (schedules) and what happens if nobody answers (escalation policies), then plug the result into the usual notification routing.

alert → routing (labels, severity, rule channels)
      → "DumbMonit on-call" channel → escalation policy
            step 1: whoever is on call on schedule X   → wait 10 min
            step 2: Alice + the #ops channel           → wait 15 min
            … repeated K times, until someone acknowledges

1. Your contacts and quiet hours

Every user declares how to reach them, in Settings → My account → On-call contacts. A contact is one of two things:

Through a team channel. Reuse a channel an admin already configured (email SMTP, SMS Twilio, Signal, Telegram) and give only your own address: your email, your phone number, your Telegram chat id. The server and its credentials stay the team’s.
Your own service. ntfy, Pushover, Telegram, Gotify, Bark, Pushbullet, Discord, Slack, Matrix… with your own settings and secrets, exactly as in a notification channel. Secrets are never shown again.

Each contact has a minimum severity (for example SMS only for critical, ntfy for everything) and a Test button. A contact that fails shows its last error.

Quiet hours (same screen, with your time zone) keep you from being woken for small things: outside an on-call shift, only critical alerts get through during those hours. While you are on call you are always paged, quiet hours or not.

Calendar link. Subscribe to my shifts generates a private iCal URL for your calendar app (30 days back to 90 days ahead). Regenerating it revokes the previous link, and Remove kills it.

2. Schedules

Alerts → On-call → Schedules → New schedule.

Time zone: an IANA name such as Europe/Paris. Handoffs happen at the same local wall-clock time all year, across daylight-saving changes (a rotation counted in hours is the exception: it counts real elapsed time).
Layers: a schedule is a stack of layers. Where several layers are active, the one lower in the list wins. Put the general coverage first and the exceptions below it.
Rotation per layer: daily, weekly, or custom (a length in hours, days or weeks). The start is the first handoff: its time of day is the handoff time, and for a weekly rotation its weekday is the handoff day.
Participants: an ordered list of users; the rotation walks through it.
Restrictions: time windows per day of the week during which the layer applies. A window whose end is at or before its start crosses midnight (18:00 to 09:00). Without restrictions the layer is always active.
Notify on handoff: optionally tells people when their shift starts and ends.

The page shows a week or month calendar and a Who is on call now panel.

Example: a team of three

Alice, Bob and Chloé, time zone Europe/Paris.

Layer	Rotation	Restrictions	Participants
Business days	weekly, starting Monday 09:00	Monday to Friday, 09:00 to 18:00	Alice, Bob, Chloé
Nights and weekends	weekly, starting Monday 18:00	Monday to Friday 18:00 to 09:00, Saturday and Sunday all day	Bob, Chloé, Alice

The second layer is lower, so it wins whenever it is active. Weekday working hours go to the first layer, evenings and weekends to the second, and nobody carries both in the same week.

Replacements

On a schedule, Add a replacement reads “X replaces Y from … to …”. If you leave Y empty, X takes over whoever would be on call during that period. Admins and operators can create and delete replacements, so a swap does not need an admin to edit the rotation. A replacement shows on the calendar and in the iCal links.

3. Escalation policies

Alerts → On-call → Escalation policies.

A policy is an ordered list of steps. Each step has one or more targets:

On-call of a schedule: whoever is on call at that moment;
a person: paged on all their contacts that match the alert’s severity;
a channel: any notification channel (a chat room, for instance).

Each step has a delay in minutes before the next one. Repeat replays the whole chain K more times when it ends without an answer.

Example: two steps, repeated twice

Step	Targets	Then wait
1	On-call of Team schedule	10 min
2	Alice, and the #ops channel	15 min

Repeat: 2. The alert pages the person on call, 10 minutes later Alice and the channel, 15 minutes later the on-call person again, and so on: three passes in total, then the escalation is marked exhausted.

What stops an escalation

Acknowledging, by anyone: in the UI, with the signed link in the notification, or by snoozing, silencing or ignoring the alert. Escalation stops and nobody else is paged.
Resolution of the alert. Escalation stops and the people who were already paged are told it cleared.

Each alert keeps an escalation log: who was paged, when, through which contact, and who acknowledged. It is on the alert’s details.

4. Plugging it in

Create a notification channel of type DumbMonit on-call (Settings → Notifications → Channels) and pick its escalation policy. It then behaves like any channel: routing filters by label, a minimum severity, and the channels you choose per rule all apply. A typical setup is a rule for critical production alerts that sends to this channel, while the rest still goes to chat.

The existing global escalation (“if nobody acknowledges, notify another channel”) is untouched and still sends to a single channel.

Use Test on the channel to page step 1 with a test message.

Roles

Everyone signed in can read schedules, policies and who is on call. Only admins create or edit schedules and policies; admins and operators manage replacements. Each user manages their own contacts, quiet hours and calendar link.

Backup and restore

Backups include schedules, replacements, policies, contacts (with their secrets, since the archive is encrypted) and preferences. Personal iCal links are not restored: generate them again.