Small details.
A clearer everyday picture.

Refresh the data, not your whole workspace.

The agent dashboard now refreshes its existing charts and live values without rebuilding the page, helping you keep your place and selected timeframes.

• CPU, memory, network and disk charts share one history request when they use the same timeframe.
• Older responses cannot overwrite a newer timeframe or draw into another view after navigation.
• Failed timeframe loads retain the previous chart and selection, with an inline message explaining how to retry.
• The dashboard uses the hosted chart library instead of a separate external CDN request, and library-load failures can be retried.
• Chart instances and their listeners are released when leaving a view. Updates skip animation, and manual refreshes share an in-flight request.

This updates the dashboard, not the agent or probe scripts installed on your hosts. Auto-refresh remains every 30 seconds while visible, subject to API caching. Historical data and monitoring behavior are unchanged.

Your whole workspace, not just the first page.

The dashboard now uses a compact workspace-wide summary instead of downloading full monitor and incident lists to calculate its overview.

• Attention Needed covers down, degraded and unknown monitors across every page, while respecting paused and maintenance states.
• Agent-update totals cover the whole workspace, with one collapsible notice instead of duplicate panels. Expanded details stay open during refresh.
• Preview up to five attention items and five agents, with shortcuts to filtered monitor lists. Recent incidents show the latest three with an accurate total.
• Dashboard loading uses three API requests instead of five; summary refresh uses one instead of two. Overlapping refreshes share a request.
• Counts, notices and recent incidents refresh without replacing the whole page. Failed requests keep the last summary visible and offer a retry.
• The setup checklist starts collapsed for existing workspaces, and attention items appear before charts and activity.

The summary polls every 15 seconds while visible and may use a server cache up to 60 seconds old. Maintenance is not counted as an active incident. Response-time averages use active monitors' last recorded positive response samples, not a historical percentile. Monitoring records and uptime history are unchanged.

A failed refresh should not erase your charts.

Public monitor pages now keep the last successful view when a history request fails, rather than replacing working charts with empty datasets.

• See clear loading, synced and error states, with a Refresh or Retry button.
• The last-synced timestamp advances only after a successful load. A failed timeframe change identifies the timeframe still on screen.
• Stalled requests have a deadline, invalid responses cannot poison the cache, and retries bypass browser validators.
• Requests are shared while in flight, the browser cache is bounded, and only enabled historical charts trigger history requests.
• Refreshes preserve your scroll position and timeframe-button focus. Returning online can retry a failed view.

Last synced means the page fetched its data successfully, not that a new monitoring check just ran. Existing server-side cache intervals still apply. No historical monitoring data was changed.

Plan around maintenance, in your own calendar.

Keep scheduled work alongside your team's plans. Public maintenance now has a calendar feed and downloadable events, without another account or email signup.

• Copy a public iCalendar subscription URL from Subscribe to Updates, or download the calendar from incident history.
• Add a single scheduled maintenance window from its banner or incident timeline.
• Calendar times adapt to your calendar app's timezone; incident history also labels the timezone used on the page.
• The subscription dialog now supports Escape, keyboard focus containment, focus restoration and scrolling on smaller screens.

Subscribe by URL for updates; a downloaded file is a snapshot. Calendar apps choose when to refresh, so use email for timely alerts. The feed includes up to 200 public windows overlapping the last 90 days or starting within the next year. Unscheduled maintenance is excluded. No reminders are added automatically.

Follow updates your way. Share the whole timeline.

Your visitors can now follow public incident and maintenance updates in their own feed reader, without handing over an email address.

• RSS and Atom feeds are available from Subscribe to Updates and the incident-history page.
• Each published update has a stable identifier and links directly to its incident timeline, including active incidents and planned maintenance.
• Filter history by active incidents, maintenance or resolved incidents, or see all updates together.
• A failed history request keeps the last loaded view available, with a retry button and clearer error messages.

Feeds contain public incidents only, with up to 100 updates from the 30 most recently updated incidents. Delivery timing depends on your feed reader's polling interval; this is not a full-history export.

Less friction in everyday workspace changes.

Workspace settings and group management now make it easier to see what is included, what changed and what still needs saving.

• Plan entitlements use readable, responsive tiles rather than an unformatted block of text.
• Brand-only changes no longer trigger custom-domain DNS validation. Domain errors appear beside the settings form instead of the delete-workspace area.
• Group edits preserve unsaved work, explain partial save failures and let you retry without losing your changes.
• Dashboard loading and error states are more consistent, with checks across desktop and small mobile screens.

See what happened after an alert.

Notification delivery should not disappear into a black box. The dashboard now brings destinations and recorded delivery outcomes together on the Notifications page.

• View recent sent, failed and pending outcomes for channels that still exist in your workspace.
• Filter by channel or outcome, move through older history, and see a last-24-hour summary.
• Get understandable failure explanations without displaying webhook credentials.
• Configure an email recipient alongside Discord, Slack, Microsoft Teams and generic webhooks.

Sent means the destination accepted the attempt, not a read receipt. Pending means no final result was recorded; it is not a promise of an automatic retry.

Help visitors find the service that matters.

A busy status page can still be easy to understand. New discovery controls let visitors focus on the services they need without changing the workspace's overall health.

• Search service names and groups, or filter by status and service group.
• Share a filtered page URL and keep the same selection after reloading.
• See whether the page has synced, is delayed, or is offline. A failed refresh keeps the last successful view and offers a retry.
• Stay at your place in the service list during automatic refreshes, with better mobile layouts and keyboard navigation.

A monitor list that works your way.

Keep the broad overview when you need it, then narrow the list down to the services that need attention.

• Switch between cards and a table, with a saved 25, 50 or 100-monitor page size.
• Use Healthy, Needs attention, Maintenance and Paused quick filters. Counts cover the matching list, not only the current page.
• Combine search, monitor type, group and status filters; clear filters without bringing back an old bulk selection.
• Select multiple monitors for maintenance changes and expand compact agent-update notices when needed.

Make the planned work stay planned.

Scheduling a future window should not make a service look as though the work has already started.

• Future maintenance edits and updates respect the scheduled start instead of activating early.
• Scheduled end times complete the window and add a completion entry to its timeline.
• Finishing one window preserves another active window or a monitor's manual maintenance setting.
• Released monitors wait for fresh monitoring data rather than being forced to an operational state.

Automatic completion runs in the background after the end time. Recovery still depends on fresh checks or agent reports; a maintenance schedule is not evidence of availability.