1. Turn it on for a project
On /errors/settings, each project has an "AI issue summary" toggle, off by default. Turning it on is a real decision, not a display preference: it opts that project into sending an issue's data to Anthropic's API server-side, so only turn it on for a project you are comfortable summarizing that way.
2. What gets sent, and what does not
Only the issue's most recent occurrence: stack trace, breadcrumbs, tags, release metadata, and occurrence pattern, always AFTER your project's scrub rules (default and any custom ones you have added) have run. Nothing is ever sent for a project with the toggle off.
3. Read the summary on the issue page
Open any issue in a project with the toggle on. A summary appears labeled "AI-generated" with the release and model it was generated against — verify it against the raw occurrences below before acting on it. It is cached per issue and release, so opening the same issue again does not call the API a second time.
4. Regenerate after a fix attempt
The "Regenerate" button forces a fresh summary, bypassing the cache. Regeneration still counts against the account's daily generation cap, which stops silently once spent for the day — no error, just no new summary until the cap resets.
5. When you see nothing
The summary section is simply absent, not an error, when the toggle is off, the account's daily cap is spent, the provider is slow or unreachable, or (staff-only, logged server-side) the feature is not configured for this deployment. The rest of the issue page is unaffected either way.
Go deeper
The full reference lives in the docs: Getting started documentation. Error codes named above are each explained in the error-code reference.