Skip to content

Alerts ​

The Alerts page is where Overdesk tells you something needs a person to look: a drafting run that failed, a report that could not reach Slack, a problem an analysis just found in your conversations. Most alerts also say what to do next, so the page doubles as a to-do list for whoever looks after Overdesk.

Where it lives ​

Open Alerts in the sidebar. The page lists the most recent 100 alerts, newest first. Each one shows its severity, a title, the message and how long ago it was raised. An alert about a finding also has a View finding link that opens it.

When you have dealt with an alert, click Acknowledge. That hides it from the list for the whole team. Tick Show acknowledged to see the ones already handled. Agents, admins and the owner can acknowledge alerts, and viewers can read them.

Severities ​

SeverityMeaning
criticalSomething stopped working and needs attention.
warningSomething went partly wrong, or will sort itself out but you should know.
infoA record of a change, with nothing to fix.

What raises an alert ​

Findings that need a decision ​

When an analysis finds a problem, it becomes a finding (see Intelligence). Some findings also raise an alert, so the team hears about them without opening Analytics:

AlertWhen it is raised
New findingAn analysis found a new issue rated High or Critical in conversations it had mostly read on its previous run, and it is not one of the issues it had already named. With the default schedules, that means Post-release regressions, Frustration signals and Churn signals, which run every day over three days. A Regression, Recurring bug or Unusual pattern raises it from a weekly run too, such as Recurring bugs.
Finding came backA finding someone had resolved or dismissed has opened again, rated High or Critical, other than a dismissed finding that is back, which has the alert below. This can come from any analysis.
Dismissed finding is backA dismissed Regression, Recurring bug or Unusual pattern, rated High or Critical, has opened again because a customer wrote about it after it was dismissed. The alert says in how many conversations, and the day it was dismissed.
Confirmed defectAn investigation confirmed a defect, found the version where it started and read its cause, and Overdesk opened a finding for it (see Confirmed defects). The alert comes once Overdesk has also checked recent conversations for the same defect, so it can say in how many conversations the defect is confirmed and in how many more it is only suspected, and name the first of them with their level. It gives the versions and the cause, and says the finding holds the workaround, the suggested patch, or both. When that check could not run, the alert says so and that Overdesk will check again.
Defect came backA confirmed defect someone had resolved has opened again, because a matching conversation started after it was resolved. It carries the same counts, versions and cause as the first alert.
Dismissed defect is backA dismissed confirmed defect has opened again, because a customer wrote about it after it was dismissed. The alert names the conversations where they wrote, each marked confirmed or suspected, and the day it was dismissed, then gives the counts, the versions and the cause.
Findings need a decisionOne run opened more than 5 findings that would each have raised an alert. They arrive as this one alert, which links to the Intelligence page.

A Critical finding raises a critical alert, and a High one a warning. A confirmed defect always raises a warning. A finding raises one alert each time it opens: until someone acknowledges it, it raises no second one, and when it comes back after being resolved or dismissed it raises a new one. A finding getting worse does not raise an alert; the What needs a decision panel on Analytics and the weekly reports show it instead.

An issue raises one alert, however many analyses report it. When another report of the same issue, as one entry on the Intelligence page, raised an alert in the last 14 days, a new report of it raises none. Nor does one when that other report was dismissed, unless the new report rates the issue at least two levels worse, such as Critical against a dismissed Medium, or is a dismissed finding that is back.

An analysis's first run raises no alert for a new finding, since everything it finds was already there. A run on a weekly schedule raises one only for a Regression, Recurring bug or Unusual pattern. An issue seen a week apart is often recorded as new, so when a weekly analysis describes a problem in different words from the week before, it can alert on it again. The other findings of these runs appear on the panel and in the weekly reports.

The three confirmed defect alerts follow none of the rules in the last two paragraphs. They are not held by another report of the same issue, and they do not depend on a run being the first or on an analysis's schedule: they are raised when the defect is confirmed, or when it opens again. A confirmed defect that comes back raises one of them, not Finding came back or Dismissed finding is back.

Critical ​

AlertWhat it means
Drafting failedA drafting run could not deliver drafts, for example because a help desk could not be reached or Anthropic refused your AI key. The message names the help desk, or says what is wrong with the key on Settings → AI. When either stays broken, Overdesk raises this once rather than on every cycle, until you acknowledge it.
Report failed, or Analysis failedA report or analysis could not be produced. The title names which one, and the message gives the reason.
Knowledge base search is degradedOverdesk stopped indexing new knowledge for search. Until it is resolved, new content is found by keyword only. If you see this one, write to hello@overdesk.io.
A help desk event could not be queuedAn update sent by your help desk could not be taken in, so Overdesk asked the help desk to send it again. If this keeps happening, contact hello@overdesk.io, because repeated failures can make the help desk switch the connection off.

Warning ​

AlertWhat it means
A destination could not receive a report or analysisThe title names the destination and the report. See Notifications for the usual causes.
Conversation import is still catching up, or Conversation sync cap reachedA large history is arriving over several runs. Overdesk saves what it read and carries on from the same place next time.
Some help desks were not imported this runOverdesk ran out of time or import allowance before it reached every help desk. Each help desk it did not reach says so on its page under Integrations and on the Knowledge base page, and goes ahead of the others on the next run. Those conversations import on the following runs.
Draft review scored every conversation the sameEvery conversation in a run of Score drafts against replies got the same result, which usually means the review paired the wrong messages. Treat those scores as unreliable until it has been checked.
Needs a human, or Handed to a named personA guardrail rule flagged a conversation for a person, or handed it to someone in your help desk. The title carries the conversation number. See Escalation.
Login link postedThe customer posted a link that logs in to their site. Overdesk hid it from its records and will not send a reply on its own. The message names the person who logs in to customer sites and has the conversation, if one does, and the customer should revoke the link once your team is done with it. It is raised whether or not sending is on, and once until it is acknowledged. The title carries the help desk and the conversation number. See Escalation.
A drafted reply has waited over a dayThe customer is still waiting, and a reply Overdesk drafted for their latest message has gone unsent for more than a day. Overdesk raises it once for that message, while the draft is between one and three days old, and acknowledging it does not bring it back. The title carries the conversation number. Overdesk checks the conversations its automatic drafting reaches on each run.
Weekly insights email was not sent, or failed for some recipientsThe weekly report was generated, but its email went out to only some recipients, or to none. See Weekly insights email.

Info ​

AlertWhat it means
Email destination changedSomeone added, changed or removed the email address that receives notifications. The message says where reports and alerts now go.

Alerts in Slack, ClickUp and email ​

Alerts about findings, and guardrail escalations (the Needs a human and Handed to alerts), are also sent to every destination that has Alerts switched on in Settings → Notifications, so the right person hears about them without opening Overdesk. A finding's alert carries a link to that finding. Every other alert stays on this page. See Notifications.

Emails to owners and admins ​

The workspace's owner and admins are also emailed when an analysis raises alerts about findings: one email per run, listing each alert with a link to it. Each email has a link to stop receiving them. It stops these emails for that person only, and leaves the weekly insights email and the rest of the team's emails as they were. To start them again, email hello@overdesk.io.