Notification bell
Notification bell
The bell in the top bar is lazyit’s in-app nudge surface: a small, curated set of “someone should glance at this” events. It is not the audit record — that lives in the activity history and the ledgers. The bell is allowed to forget; the history is not.
Who sees the bell
The bell shows for every signed-in person, but what each person sees is scoped:
- Broadcast notifications — the estate-wide operational nudges (a critical-app grant, an admin elevation, low stock, a workflow that needs a human or failed, a sensitive permission grant, an offline agent, an expiring asset warranty or access grant) are visible only to people whose role holds the notification permission. By default that is administrators only.
- Targeted notifications — a notification addressed to one specific person lands in their own bell, even if they are not an administrator and hold no notification permission. These include the vault-setup nudge and the access-request decision that tells a requester their request was approved or denied (see below).
So a non-administrator with nothing addressed to them simply sees a clean bell with no badge; an administrator sees the broadcast feed plus anything targeted to them.
What triggers a notification
The set of triggers is fixed and deliberately small — the bell is a curated nudge, not a firehose:
| Notification | Fires when |
|---|---|
| Critical-app access | A grant opened access to an application flagged as critical. |
| Admin granted | A grant or role change raised someone to the Administrator role. |
| Low stock | A consumable crossed from above its minimum stock to at or below it. |
| Manual task | A workflow run paused and is waiting for a person to act. |
| Run failed | A workflow run failed or escalated and stopped. |
| Sensitive permission granted | A role-permission edit gave the Member or Viewer role a high-risk capability — instance settings, user management, access-grant control, or any delete permission. Links to the role-permission editor. |
| Agent offline | A reporting agent stopped sending reports and its node was flipped offline. One notification per outage — not once per check. Links to the topology map. |
| Duplicate machine ID | Two servers reported the same machine ID but different hardware (a different serial and different network cards) — almost always a VM template or golden image cloned with one baked in. Both are listed separately (nothing is merged) and the notification names the fix. One notification per colliding host, not one per check-in. Links to the topology map. |
| Vault setup | (Targeted, one-time) A person who can read secrets but has never set a vault passphrase is nudged at sign-in to set one up. |
| Access request decided | (Targeted) Your own access request was approved or denied. Lands in your bell — and inbox, unless you opt out — so you don’t have to poll your request list. Links to the application. |
| Warranty expiring | An asset’s warranty is within its look-ahead window (the last 90 days before it ends) and has not yet lapsed. One heads-up per asset, so you can renew or replace before the deadline instead of finding out after. Links to the asset. |
| Access expiring | An active access grant is within two weeks of its expiry date and will be auto-revoked when it passes. One heads-up per grant so you can re-grant before a contractor or temp silently loses access. Links to the application. |
| Receipt acknowledged | (Targeted) Someone confirmed receipt of an asset you assigned to them. Lands in your bell — and inbox, unless you opt out — so you know the handover is complete without asking. Links to the asset. |
| New AI agent connected | (Targeted, security) An external AI agent — Claude Code, Cursor or another MCP client, or a personal MCP token — was used on your account for the first time. One per connection. Lands in your bell — and inbox, unless you opt out. If it wasn’t you, revoke the connection in your AI connections and change your password. |
Notifications are emitted after the originating action completes and are best-effort: a notification that fails to send never blocks or undoes the underlying change. Repeated triggers for the same event collapse into one notification, so a consumable hovering around its threshold will not spam the bell.
Reading notifications
Open the bell to see your notifications, newest first. Each row carries an icon, a short title, an optional one-line summary, and a relative time.
- Click a row to open what it is about — the application, the consumable, or the workflow task inbox — and the row is marked read.
- Mark all read clears the unread badge in one click.
- The red badge counts unread notifications; it shows
99+past ninety-nine. The exact number is still read out to screen readers.
Grouped by type and day
A bulk operation — an import, a large batch of grants, an offboarding — can raise hundreds of notifications of one kind in a day. So the bell groups notifications by type and day: everything of one kind that happened on one day is a single row, such as “312 grants on critical applications · Today”.
- A group with one notification shows as that notification, exactly as above.
- A larger group shows how many notifications it holds, the day (Today, Yesterday, or the date), how many are still unread, and a preview of the newest few.
- Select the group to expand it in place and see each notification in it; select it again to collapse it. Long groups load more as you ask for them with Load more.
- Mark group as read marks every notification in that group read at once, including the ones not loaded yet. Other groups are untouched.
Days follow the instance’s display time zone (see Time zone & formats), so “Today” means today where your team works. When there are more groups than fit, Load more at the bottom of the bell brings the next ones; the bell itself stays a fixed height and scrolls.
Read state is per person: marking a broadcast read clears it for you only, not for other administrators.
Dismissing notifications
Once you have dealt with a notification, you can remove it from your bell instead of waiting for it to age out:
- Dismiss one — hover a notification and select the × at its right edge. Inside an expanded group, each notification has its own ×. On a phone or tablet the × is always shown; with a keyboard, tab to it. Dismissing does not open the notification.
- Clear all — in the bell’s header, removes every notification up to the newest one shown, all at once — including every notification inside each group, and older ones further down the list than the bell loads. Anything newer stays: a notification that arrived after you opened the bell is not cleared, and it stays unread. When the bell is empty it reads “You’re all caught up.”
Dismissing also marks the notification read, so the badge drops with it. A notification that arrives after you select Clear all shows up normally.
Dismissing is per person, like read state. It hides the notification from your bell only: the event itself is not deleted, and other administrators who see the same broadcast still have it in their own bells. The durable record in the activity history is untouched.
[!IMPORTANT] Neither action asks for confirmation, there is no undo, and there is no view of dismissed notifications. Once dismissed, a notification stays out of your bell until it is pruned with the rest after 90 days. Dismiss what you have handled; leave in place anything you still need to act on.
What a notification carries
Notification text is short and redacted by design — it carries names and identifiers only, never record bodies, secrets, or sensitive personal data. The vault-setup nudge in particular carries no key material: it only says a vault passphrase has not been set and links to the Secret Manager.
Retention
The bell keeps notifications for 90 days, then prunes the old ones automatically. This is intentional: the bell is an operational nudge surface, so it is allowed to forget. The durable record of who did what lives in the activity history and reports, which is never pruned this way.
Delivery
In the current version the bell polls for new notifications, so there is a short delay before a new event appears. Live push is a planned upgrade behind the same surface; nothing about how you use the bell changes when it lands.
Granting the bell to other roles
The broadcast feed is administrator-only by default because it surfaces sensitive operational state (who was given access to a critical app, who was made an administrator). An administrator can grant the notification permission to the Member or Viewer role from the role-permission settings if they want those roles to see the broadcast feed. See Permissions for how to tune what each role may do.