Configuring alerts
Agent alerts are tenant-scoped records for noteworthy agent activity. The settings page controls in-app delivery, email delivery, daily digest delivery, and minimum severity.
What this is
The settings page lives in Accounting at /agent-alerts-settings. It reads agentAlertPreferences and saves changes through updateAgentAlertPreferences.
Four controls are available:
- In-App Notifications - show the alert bell and unread badge in the header.
- Email Notifications - send real-time email alerts to the configured address.
- Daily Digest - send a daily summary of recent agent activity.
- Minimum Severity - choose
info,warning, orcritical.
The settings page syncs from server preferences when they load and shows a Save Preferences button only after local changes are dirty.
When you’d use it
- Set the alert email during onboarding.
- Route alerts to a shared operations inbox.
- Turn off in-app delivery if your team relies on email.
- Raise the minimum severity when info-level alerts are too noisy.
- Keep the daily digest on for a slower review of recent agent activity.
Walkthrough
In-App Notifications. Toggle whether in-app alerts are enabled. The broader Accounting shell uses the alert queries for unread counts and alert dropdown behavior.
Email Notifications. Toggle email alerts and set the destination email address. The saved preference shape is emailEnabled plus emailAddress.
Daily Digest. Toggle the daily digest. The backend digest handler is POST /internal/agent-alerts/daily-digest, requires scheduler authentication, and sends a “Daily Agent Activity Digest” email for tenants with recent activity and digest-enabled preferences. The UI describes this as 07:00 UTC daily.
Minimum Severity. Pick one:
| Option | Value | Effect |
|---|---|---|
| Info & above | info | All alerts |
| Warning & above | warning | Skip info-level alerts |
| Critical only | critical | Only urgent alerts |
Alert records
The public GraphQL alert record has these fields:
idseveritytitlemessagechannelemailSentreadAtdismissedAtcreatedAt
Alerts can be listed, counted, marked read, dismissed, and filtered to unread records. The source does not expose a single customer-facing taxonomy of alert categories in this page, so this guide avoids naming a fixed trigger list.
Programmatic access
- GraphQL (accounting-api):
agentAlerts(unreadOnly, limit),agentAlertUnreadCount, andagentAlertPreferences. - GraphQL mutations:
markAgentAlertRead(id),dismissAgentAlert(id), andupdateAgentAlertPreferences(input). - Digest: the daily digest endpoint is internal and scheduler-authenticated.
The current AI COO MCP catalog does not expose alert-preference management tools, and this page does not document a public alert webhook subscription.
Related
- Agent activity timeline - operational history behind many alert investigations
- Pending actions: review, approve, reject - the confirmation queue
- The AI COO and agentic workflows - concept
- The audit log and the change ledger - provenance for agent events