Documentation · Alerts

AI explanations

On accounts where it is turned on, Claude reads the probe evidence of an outage and writes what happened and what to try.

When a monitor is confirmed down, PingWise can ask Claude, Anthropic's model, to read the evidence and write a short explanation: a headline, what happened, the likely cause, up to three things to try, the scope, and how confident it is. It appears on the Incidents page and, if the account chooses, in the down alert itself, which then waits up to 45 seconds for it.

Who has it

It is off by default. While we finish the paperwork that lets us list Anthropic among the providers in our Privacy notice and the Data Processing Agreement, it is on only for the account of our own hosting service, Hostbox, which built PingWise; we intend to offer it to every account, as a switch the account's owner controls, once that is done. Sub-accounts of a reseller do not inherit it.

What is sent to Anthropic

The monitor's type, target and check settings; each probe's city, continent, provider, result and diagnosis; and the monitor's incidents from the previous seven days. Never sent: names, email addresses, request headers or bodies, credentials, or the monitor's notes.

What it is not

The explanation is a reading of evidence, not a verdict; the alert, the incident and uptime are computed by the quorum rules without it. The explanation can be wrong, and says so when the evidence is thin.

Last change: 31 October 2026.