Skip to content

Triage in the inbox

The Inbox shows everything that still needs a decision, grouped by what you do next:

  • Needs you — open incidents nobody has taken yet.
  • Acknowledged — someone has it; repeats don’t alert.
  • Snoozed — quiet until the snooze ends.
  • Recently resolved — folded away; Show opens the last 20, and Show more goes further back. Each group’s count is how many incidents it is listing.

A new workspace, with no source yet, says so: the list reads “No sources yet”, and the right-hand side shows how an alert travels (source → Notifications → destination) and three setup steps, each with a button: Add a source, Choose where you’re told and Your first alert was delivered. Until the workspace has a destination, the second step opens Destinations, where you add a chat, a webhook or pair a phone — a paired phone counts on its own once it’s actually paired; after that it opens Routing. The third step goes to a source’s Connect tab, where a real alert can actually be sent from. Once a destination is on Default, a line under that step says what a plain event would actually do right now — for example “Critical and Error → Telegram” — read live from the same check Routing’s coverage grid runs, so it can’t promise more than your rules actually deliver. It reads as a warning, not the ordinary muted line, whenever a severity above Debug or Info would reach nobody.

Adding a source isn’t the end of the first step: until it actually receives its first event, the step reads Finish connecting <name> instead and still links to its Connect tab. Once a source exists and until every step is done, a Finish setting up card sits above the list with the same steps, including that same outcome line once it applies. Each ticks itself off from what has really happened (a source is actually receiving; a real destination — a paired phone included — is on Default; a real alert has actually been delivered, never a receiving source alone or a destination’s “Send test”), and the card goes away once all three are done. Members don’t see the third step or the outcome line, since checking coverage needs an owner or admin.

If the list can’t be loaded (the server is busy or unreachable), the Inbox says so where the rows would be, with Retry, rather than showing an empty, calm list. When the server asked to wait, it retries by itself.

Each incident is one row of two lines: its severity (a shape and a word: ◆ critical, ■ error, ▲ warning, ● info) and title with how many events it has gathered (×14), then its source, what state it is in (“flapping”, “acknowledged 5m ago”, “snoozed until 12:00”) and, once a notice has actually reached somewhere, who was told (“Told: Telegram, once”). Within a group the loudest come first, then the newest to open.

The Inbox lists what needs a person; everything that arrived, whatever became of it, is in History › Events received instead.

The box at the top narrows every group at once. Plain words match the title, the source and the resource. Two prefixes do more:

Type Finds
sev:crit Incidents of that severity (the first letters are enough)
source:proxmox Incidents from sources whose name contains it (spaces ignored)

Beside the box, two pills narrow it further, the same way every list’s filters do: Severity (Everything, Critical only, Error and worse, Warning and worse, Info and worse) and Source. The box and the pills appear once there is an incident to filter; the key legend joins them too, but only with a mouse or trackpad — a touch-only screen never shows keyboard hints it can’t use.

The filter and the pills are part of the address, so a filtered Inbox is a link you can share or bookmark. Press / to jump to the box, and Esc to leave it.

The Inbox updates itself as events arrive. A row’s count, time and state change where it stands, but rows don’t move under you: when an incident resolves, comes back, is acknowledged elsewhere or is new, a bar at the top says so, for example ““Backup failed on pve2” is resolved”. Press Update list to put everything where it now belongs. Your own actions move their row straight away. If the pill at the top right says Reconnecting… instead of Live, the Inbox is not hearing about new events; it catches up once the pill turns green again.

On a wide screen the chosen incident sits beside the list; the Inbox opens on the first one. On a phone the list and the incident take turns, and the Inbox breadcrumb goes back. Open on its own page shows the same view at full width, at an address you can share; History links there too, and its breadcrumb leads back to where you came from. An incident of another workspace is never shown under the one you’re in: the view says which workspace it is in and offers to switch.

It starts with one sentence: how bad it is (and how bad it got, if that was worse), what state it’s in and since when, how many events, and who was told — naming any destination a notice failed to reach, even when others heard. An incident whose source was deleted says so rather than reading as a manual resolve. When the source measured something — the watcher’s disk, memory and clock checks do — a second line gives the numbers: “Memory stall reached 14% at 12:03, over its 10% limit. Highest: 22%.”, or “was back to 3% …, under 5%” once it recovered. They come from the incident’s own events; there is no graph. Below the actions, What happened is its whole story, newest first, in one list:

  • the events that make it up, as the source sent them, with any recovery marked — click one to read it in a panel on the right, with its body, its metadata and Preview routing for this event, without leaving the incident;
  • every state change, who made it and from where (the web, the app, Telegram, the API);
  • every notice, to which destination, and whether it was sent, skipped or failed.

Above the story, Who was told, and why shows the source, what decided (a rule, built-in rules, a rule for some sources or Default) and who was told, then one row per destination: sent, held by quiet hours, or skipped with its reason. A failed one shows its error and, for an owner or admin, Retry right there; N notices under a row lists every notice it got (the alert, a repeat, the all-clear), each with its status. When nobody was told, one line says why — for example “Default, the rule for everything else, has no destination for this severity” — with a link to change it. Each part links to what decided it, and back. See Who was told, and why. Every notice from every incident is also in History › Notices sent.

From the buttons beside the incident or a keyboard shortcut: Acknowledge, Snooze (for an hour, four hours, until 08:00, or a time you pick), Resolve (asked first). See How incidents work for exactly what each one does. A heartbeat-owned incident whose heartbeat is still down offers Pause heartbeat in place of Resolve. A resolved incident offers none of them: a new alert for the same problem within 30 minutes reopens it; after that, it opens a new incident. Mute source… holds back notices from every incident of the incident’s source for a while — not just this one; its incidents still open.

After an action a short message says what happened; for acknowledge, unacknowledge and snooze it offers Undo for a few seconds, and leaving the inbox takes it away. Resolving has no undo.

With the Inbox open and nothing else capturing keys (no dialog open, not typing into a field):

Key Does
j Choose the next incident
k Choose the previous incident
e Acknowledge it
s Snooze it for an hour
x Resolve it (asks first)
w Unsnooze it
u Undo the last action, while its message shows
/ Filter

Moving stops at either end rather than wrapping around.

⌘K (macOS/iOS) or Ctrl K opens a search box over areas (Watchers included, as Sources › Watchers, and, once you type, every Settings screen and area tab), sources and your open incidents, so you can jump to one without paging through the list. On a phone, the search button in the top bar opens it. It stands aside while a dialog is already open or you’re typing into a field.

Acknowledge, snooze and resolve all work the same way from a paired phone as from the web dashboard — they’re the same actions, over the same API, so a change made on one shows up on the other within moments. See iPhone app and push.