Routing: who is told, and when
Open Notifications → Routing. It starts with what you change most, top to bottom:
- Default: what happens to an event when no custom or built-in rule decides first, one row per severity — Severity, In InfraInbox (Opens an incident, Saved in History or Hidden), Notify (who is told, by name: Phones is the built-in push, a recipient told without sound says (silent), and nobody when no one is) and Edit. Success reads Ends its incident: it only closes the incident it matches, and whoever was told gets the Resolved notice. Edit on a row, or Edit default behavior, opens Default in a panel on the right (a member sees View and reads it). If Default sends Debug to Phones, a line under the table offers Change Debug’s recipients, which opens Default too: fewer messages, with the events still kept.
- By source: the sources that tell other people than Default does (rules for some sources), one row each (“source is homeassistant, jellyfin → like Default, but tell Telegram”), with how many sources go by it. Change recipients adds one. With none, it says Every source uses Default’s recipients.
- Custom rules: your own rules, each read as a sentence (“When type is backup_failed → Incident · Telegram”), with an On / Paused switch. Open one to edit it; Create a custom rule adds one. A filter box appears once you have five rules. With none, it says No custom rules. — the built-in rules and Default already handle every event.
- Quiet hours & mutes: one row saying where things stand (“Quiet hours off · nothing muted”, “Quiet hours 23:00–07:00 · 1 mute or silence active”); it opens Muting.
- Built-in rules: the built-in rules each integration ships, for the integrations you have a source for, one line each (“Proxmox · 6 rules”), saying which sources they apply to, whether they are as shipped or edited, and Update available when a newer version ships. Click a line to read its rules as rows (When · Outcome · Notify · Closes, the exact event types under Details); Edit rules opens their own page. The integrations you have no source for are folded under Other integrations (N).
How rules are checked, folded under the list, gives the order every event is checked in: custom rules top to bottom, then the built-in rules of the event’s integration, then Default (with the source’s own recipients if it has them). The first rule that decides wins. Under it, Coverage check shows the coverage grid (see below). A rule or Default opens in a panel on the right, over the list; close it with ✕ or Escape (it asks first if you have unsaved changes). Save closes it too and takes you back to where you opened it from — a source’s Notifications tab, say — with the list already showing the change; a new rule gets the focus in the list. Each has its own address, so you can link to it.
Suggestions
Section titled “Suggestions”Under your custom rules, Suggestions lists a few common ones, one line each: Hide Debug events, Quiet low severities at weekends, Always tell me about backups (only when a Proxmox or backup source exists) and Send one host somewhere else. It stays folded until you open it. + Add saves the rule switched off and opens it, so you can check it and turn it on when it looks right. The list hides once you have three rules; Suggestions brings it back. A paused rule reads dimmed, with its switch saying Paused.
How rules are checked
Section titled “How rules are checked”Your custom rules run first, top to bottom, then the built-in rules, then Default. The first rule that matches decides the event’s route — dropped, kept in History, or opened as an incident — and its notification policy. A rule can also be marked continue, which doesn’t decide anything by itself: it only adds its destinations, then evaluation keeps going to the rules below it. A rule routed drop is always terminal; nothing below it runs for that event.
Drag rules to reorder them (or use the arrows beside each one), pause one with its switch without deleting it, or remove it entirely.
Building a rule
Section titled “Building a rule”A custom rule opens as Create a custom rule (or Edit custom rule). Three parts stay open — 1 When, 2 Do this with it and 3 Who is told — and the rest fold under one line each that says what is set, such as “Repeat: every 5m, up to 6 times” or “If nobody reacts: off”. An untouched new rule has nothing unsaved, so closing it doesn’t ask.
1. When an event looks like this — the condition. It starts with Severities: All (the default: every severity, Debug included) or the ones you pick, such as Critical and Error. Below that, leave the conditions empty and the rule matches everything that reaches it. A condition you add starts empty: pick a field, then the operator and the value. Build one from these fields:
| Field | Operators | Matches against |
|---|---|---|
source |
eq, ne, in | Which source sent it |
source.integration |
eq, ne, in, glob | Which integration it speaks for |
severity |
eq, ne, in, gte, lte | The event’s severity, by rank for gte/lte |
action |
eq, ne | trigger or resolve |
type, resource, title, body |
eq, ne, in, contains, glob, exists | Free text or the event’s own dotted type |
tags |
contains, in, exists | Tags on the event |
metadata.<path> |
eq, ne, in, contains, glob, exists, gte, lte | Anything in the event’s own metadata |
time |
within | A day/time window, in the workspace’s time zone |
The Field box is a search: fields are grouped (Source, Event, When), each with a line on what it holds, and metadata keys already seen in your events are listed with a few of their values. A key nobody has sent yet can be typed as metadata. and then the key.
Metadata is matched by type, the way the event sent it: "443" (text) and 443 (a number) are different values, and most senders send text. A value you type is text; when it reads as a number or as true/false, tick Match as a number (or Match as true or false) to match it as one. A saved value keeps its type, contains and glob compare text, and gte/lte compare numbers.
With two or more conditions you can choose All of these or Any of these, and add groups to nest them.
2. Do this with it — one of:
| Route | Effect |
|---|---|
| Drop | The event is hidden from the inbox and never opens an incident. When a Drop could hide Error or Critical events, the editor warns that nobody will be told about failures, says how many of the last 7 days’ Error and Critical events it would have hidden, and offers Limit to Debug and Info. Saving it anyway takes a second click on Save anyway. |
| Keep in History | Saved in History, but opens no incident. If you attach destinations, this becomes an informational notice — a one-off message about the event, not an alert about an ongoing incident. |
| Incident | Opens or updates an incident. |
| Only change recipients… | Only on a new rule: opens Change recipients instead, which keeps Default’s handling and changes only who is told for some sources. |
A rule can also have no route — most rules in an integration’s built-in rules only set how long an incident stays open. Such a rule decides nothing: it changes the event on its way through, and the rules below still decide where it goes. Its editor says so, with no route picked; picking one makes the rule decide.
3. Who is told — one row per destination (including Default’s recipients, meaning Phones plus whatever Default attaches): tick it to notify it, then choose Sound or Silent beside it (silent is still delivered, just quietly where the channel supports it — passive delivery on the phone, or without a notification tone on Telegram). An incident route with no destinations at all is a silent incident: visible everywhere in the dashboard, nothing pushed anywhere. Add a delay holds the first message back. On an incident route it is useful for “don’t tell me unless this is still open in five minutes”: nothing is sent if the incident is acknowledged or resolved by then. On Keep in History there is no incident: the one message waits that long, then is sent. No delay removes it.
Repeat (folded) — repeat the message while the incident stays open and nobody has acknowledged it: how often (1 minute to 720 hours), how many times at most, and from which severity up (Critical unless you lower it), all on one line. Acknowledging, snoozing or resolving stops the repeats. If a severity this rule (or this Default card) covers is quieter than that floor, a note says so and offers to lower it in one click, such as Repeat Error too.
If nobody reacts (folded) — notify one more destination, once, when nobody has acknowledged after a delay, or when every delivery in step 3 failed outright; from which severity up (Critical unless you lower it) sits beside it too.
When it recovers (folded) — Send a Resolved message toggles whether this route hears about it at all when things clear.
Advanced, closed until you open it, holds the one rarely changed setting: At most one repeat every (10 minutes if you don’t set one), which limits how often a repeat occurrence of the same open incident messages a destination.
Change the event (folded) — optionally Change the event’s severity to… (this rewrites the event; it is not a filter — a built-in rule never does this, only your own rules can) and add tags. Repeat, If nobody reacts and When it recovers only show for an incident; a Drop has no “who is told”. A folded part with something to fix opens by itself.
At the foot of the panel, In plain words says what the rule does (two lines at most on a phone), Test this draft runs your unsaved rule through Preview routing, and a rule you have saved has the same On / Paused switch as the list, and Delete, at the top.
Default
Section titled “Default”Default (Notifications → Routing → Default) is the table at the top of Routing and the rule every event that nothing above caught falls through to. It’s always last, always enabled, and it’s the one place a rule is defined per severity rather than by condition:
| Severity | What you set |
|---|---|
| Critical, Error, Warning | Route (drop, inbox or incident) and a full policy, same as any other rule |
| Info, Debug | Same, though these usually stay in the inbox |
| SUCCESS | Fixed: a success never opens anything on its own — it only resolves the incident it matches (by dedupKey or fingerprint), and whichever destinations were already alerted get a Resolved notice |
It opens as Edit default behavior, one card per severity, each saying in words what happens (“Opens an incident, tells phones, Telegram”); click a card to change its route and who is told, one row per destination with its Sound/Silent choice. A link to one severity (from Routing’s table, say) opens that card. Inside Default, Phones is the built-in push, and its row says so: every member’s paired phone, on top of anything else you attach.
Changes here are saved from the save bar at the foot of the panel, which says how many severities changed; a save confirms with Saved and Undo and closes the panel, and leaving with unsaved changes asks first.
Rules for some sources
Section titled “Rules for some sources”A rule for some sources handles them like everything else, but tells someone else: “Quiet: Telegram only, no push” for the projects that don’t matter at 3 a.m., say. It is listed under By source, reading “source is … → like Default, but tell …”, and one such rule can serve many sources. A source can be in only one of these rules.
- It changes one thing. For its sources, it replaces your default destinations. Everything else still follows Default: which severities open an incident, delays and re-notify. A severity Default keeps silent stays silent, and Default’s fallback step isn’t used for these sources.
- It runs last. Your custom rules and the built-in rules still decide first; it only stands in for Default. A built-in rule that tells “Default’s recipients” still tells them, unless you switch on built-in alerts too.
- Phones is a choice. Leave it out and what it decides for those sources never reaches a phone; the editor warns “No push” when you do.
- Nobody at all is allowed. One with no destinations makes those incidents silent: they stay visible in the dashboard, and no one is notified.
- One per source. Ticking a source that another such rule has moves it: each source in the list says where it is now (“kuma (now: Default)”), and the panel lists the moves a save will make.
Change recipients opens its panel: the sources (when it wasn’t opened for one), who is told, the built-in alerts switch, and a name. Opened for one source, it is titled Recipients for that source, with no picker. It offers no Drop / Keep in History / Incident choice: handling follows Default. One line says what the switch means (“Built-in alerts keep their own recipients; custom rules can still tell others”). A new custom rule can switch to it with Only change recipients… (its source conditions come along). A row opens it to edit or delete. Under the form, What changes for its sources compares it with Default, severity by severity, and lists the sources a save moves out of another such rule. Nothing counts as unsaved until you change something.
On a source’s own page, its Notifications tab starts with Recipients: which row handles it, and lets you switch (Handled by: Default or one of these rules), says so with Undo, links to Edit it, and Change recipients adds one for that source in a side panel. Default behavior under it is Default’s table, with the source’s own recipients in Notify where it has them. A source on one of these rules shows its name by its own. The tab’s Preview shows who a chosen event from the source — a plain one, the integration’s example, or one of its own recent events — would reach. Members see all of this; owners and admins change it. See Who hears about a source.
Built-in alerts too
Section titled “Built-in alerts too”Also use these recipients for built-in alerts, under Who is told and off by default, hands the rule’s recipients its sources’ built-in alerts as well: each built-in rule of theirs — its own copy, else the workspace’s, else what ships — that tells “Default’s recipients” in a step or its fallback tells this rule’s recipients instead. A destination a built-in rule names itself stays, and your own rules keep their meaning, those that also apply the rules below included. Repeats, the fallback step, the coverage grid and Preview routing follow the switch too, and with nobody told those built-in incidents are silent as well.
Turned on, the panel reads each selected source’s own effective built-in rules and folds a before and after for each rule it changes behind one line: “Changes 5 Proxmox built-in rules”, then “Backup and PBS job failures: phones, Telegram → Telegram”; sources whose result matches fold into one line, named “(3 sources)”. Leave Phones out and it says plainly that Critical built-in alerts (HA fencing, disk faults) will reach no phone. Under By source, the row adds “· built-in alerts too”. The panel’s No push and Nobody is told notes hold only for what Default decides; the line under the switch says whether built-in alerts follow, and that your custom rules can still tell others.
Seeing what actually happens
Section titled “Seeing what actually happens”Under the list, Coverage check is folded to one line by default — “Coverage check · test-proxmox”, with the number of warnings when there are any — and opens itself when something needs a look: a destination blocked right now, one no rule reaches, or one that can’t deliver yet. Your choice to open or close it is remembered. Opened, it shows coverage: one plain event per severity from the source you pick, run down the list live (nothing sent, nothing stored), as a grid of destinations × severities. A filled mark means notified (a one-off message from a severity kept in History counts), a hollow one means notified silently, a warning mark means blocked right now (by a silence or quiet hours, say), and a dash means nothing is sent. Point at or tab to a mark to see what decided it: for example ‘Dropped by your rule “Drop debug noise”’, Proxmox’s built-in rules or Default. When one of your rules decides every severity, the note under the grid names it and links to it. It also calls out destinations no rule reaches at all for the source you picked, and destinations your rules do name but that can’t deliver yet (an unpaired phone, for instance).
For a specific event rather than the coverage grid, use Preview routing — the same button is one click away from both screens.
- Built-in rules — the rules an integration ships, which run under your own
- Muting: silences, mutes and quiet hours — hold a notice back without touching routing; a silence matches the same fields as a rule, metadata included, but not
action,title,bodyortime - Preview routing — run a real or sample event through everything above, live