Docs▸Getting started

Error groups and alerts

Vigilon groups repeated errors into one error group per app, and emails your team the first time a new group appears and when a resolved one comes back. This page explains what makes two errors the same group, what resolve and mute do, and who gets the email.

How errors are grouped

Two errors land in the same group when they come from the same app and environment, have the same exception type, and have the same message once variable parts are removed. Vigilon lowercases the message, replaces every run of digits and every long hex string (IDs, hashes) with ?, and collapses whitespace. So Order 18233 not found and Order 90412 not found are one group.

For errors from your backend, stack frames and the app version are not part of a group's identity. A deploy does not split a group, and the same error thrown from a different line stays in the same group as long as its type and message match. Groups are kept per project, environment (deployment.environment) and app (service.name).

A request fails when its root span has an error status or responds with a 5xx status, or when any span in its trace records an exception. If the trace has an exception, the earliest one decides the group. A failed request with no exception is grouped by route, status code and app instead, so a burst of 503s on /checkout is one group.

A background job run that fails is grouped by app, job name, exception type (or the span name when there is none) and normalised message. Jobs report runs only when you wrap them; see Background jobs in the Node and Python guides.

Browser errors are grouped by where they were thrown, not by their message, because each browser words the same error differently. A browser error's group is its app, its exception type, the first one or two named functions from your own code on its stack (file and function, read through your source maps), and the text of the first one's source line when your source maps include source text. The line's text counts, not its number, so adding code above it or reindenting it does not split the group. An error with no named function from your code on its stack is grouped by type and message instead, with each browser's wording of common errors, such as reading a property of undefined, rewritten to one form.

Caught errors

An exception you catch and handle yourself never reaches a group unless you record it. Call recordException() in Node or record_error() in Python wherever you handle errors.

Resolve, mute and regressions

Every group is Unresolved, Resolved or Regressed. The Errors page shows unresolved and regressed groups by default; resolved groups are one filter away.

Resolve marks the group fixed as of its latest occurrence. If any new occurrence arrives after that, the group becomes Regressed, shows up in the list again, and sends a regression email. Regression is not tied to a release: any later occurrence counts. Resolve it again once the fix is out.

Mute silences a group's emails, either until you unmute it or until a date you pick. A muted group stays in the list and keeps counting occurrences. Mute is independent of resolve: a muted group can still be resolved and regress, it just does not email anyone. Occurrences that arrive while a group is muted never send an email later.

Any workspace member can resolve, reopen, mute and unmute. Coding agents connected over MCP can read error groups but cannot change their state.

Email alerts for new and regressed errors

Vigilon sends two kinds of error email: New error in <project> the first time a group appears, and Error regressed in <project> the first time a resolved group occurs again. Each group sends at most one new-error email, and at most one regression email each time it is resolved.

The email names the project, environment, app, origin (request or job, plus the job name), and starts with the error message. Its button opens the group in the dashboard on the hour before the last occurrence.

Alerts go out within about a minute of the error reaching Vigilon. There is no digest. To keep a bad deploy from flooding your inbox, Vigilon sends at most 5 error emails per project and 15 per workspace in any 10 minutes; the rest wait, regressions first, and any still waiting after 24 hours are dropped. An alert that no longer applies by the time it would send, because the group was resolved or muted in the meantime, is cancelled.

Who gets the email

By default every active member of the workspace gets error emails for every project. Admins and owners change this under Alerts in the dashboard: turn error alerts off for the whole workspace, or pick the people who receive them.

Each project can follow the workspace setting or override it: turn its alerts on or off, choose its own recipients, or turn alerts off for single environments such as staging. New environments are alerted on by default. Turning alerts off for the workspace stops them for every project.

There is no per-person opt-out yet: a member who should not receive alerts is removed from the recipient list by an admin. Serious findings from automatic insights, on the Team and Pro plans, are sent separately and have their own switch.

How long error data is kept

Individual occurrences and their traces are kept for your plan's retention period, 30 days by default. A group's first seen and last seen dates and its resolve and mute state are kept after that, so a group that was quiet for months still regresses, and still emails, when it comes back.