Integrations
Where alert delivery and cross-cutting platform behavior are configured: which channels get which
severities, whether infra spans correlate with agent traces, and your analytics consent. Reached
via the user menu; /settings/notifications redirects here.
What you see
Four tabs: Notification Routing (default), Observability, Privacy, and Redaction.
Weekly brief email
A toggle for the Monday email for your current organization. The org owner is opted in by default; other members are opted out until they switch it on. The email lists what needs attention, in the same order as the dashboard’s worklist, with a link to the evidence for each item. A finding that was already in last week’s email is marked still open with how many weeks it has been open, not announced again. Spend, runs and production success appear as three separate lines. A week with nothing to flag still sends a short note, so no email means it was not sent, never “all clear”.
Channel routing
A grid of 3 severities (Info, Warning, Critical) by 5 channels (Email, In-app, Slack, PagerDuty, Webhook) — one toggle per cell. Email and in-app are on by default for every account; Slack, PagerDuty, and webhook need their own configuration below the grid before they fire.
Slack connects via a workspace-connect button, PagerDuty needs an integration key, and webhook needs a URL and a signing secret — outbound webhook payloads are signed with HMAC-SHA256.
You can also set an email override address so critical alerts route somewhere other than your account email.
When the Investigator finishes looking into an alert, the alert’s owner gets one message on the email and in-app channels enabled for that severity: what changed, the hypothesis, how strong the evidence is (confidence plus how many evidence items and traces were checked against your data), the suggested next step, and links. Every confidence level sends; low confidence is labelled as such. The message is a hypothesis, not a confirmed root cause.
Observability
Two toggles — HTTP infra spans and Database infra spans — that control whether HTTP and database spans from your own OTel instrumentation (any framework, any database or driver) correlate with your AI agent traces on the same trace. Both default on; turning one off drops those spans at ingest going forward (existing traces are unaffected).
Privacy
An Analytics toggle for anonymized product-usage analytics consent — the same setting the in-app consent notification’s Accept/Decline actions update.
Redaction
Organization owners and admins can add organization-specific regex patterns on the Redaction tab and preview their effect on sample text. These rules extend the mandatory server-side redaction set; they do not disable it. Select the organization whose patterns you want to manage.
How to use it
- Toggle a cell in the channel grid to mute or enable that severity on that channel; changes take effect on the next alert.
- Configure Slack, PagerDuty, or webhook before relying on them — an enabled-but-unconfigured channel won’t deliver.
- Turn off HTTP or DB infra spans if you already have separate infra/APM tooling and only want AI traces here.
- Set analytics consent once — it stays in sync whether you change it here or from the consent notification.
Before you have data
The grid isn’t hidden until you configure a channel — it always renders with Email and In-app already on. Enabling Slack, PagerDuty, or webhook before you’ve connected them is safe: nothing is sent on that channel until it’s configured.