PagerDuty Integration
Connect PagerDuty to page your on-call team about AI agent issues. ANTS Platform creates one PagerDuty service per agent and sends events to that service through the PagerDuty Events API v2.
What triggers an incident
| Trigger | How |
|---|---|
| PII, PCI or PHI, ethics or bias, and safety, jailbreak or prompt injection violations found in a trace | Automatic, when PagerDuty alerting is on for the agent (see Per-agent alerting) |
| AI Agent Center > agent > Risk tab > Unapproved Models | On demand: the PagerDuty icon in Send Compliance Notification |
| Drift alerts | On demand: the notification button on a drift alert (see Drift alerts) |
| Prompt automations | Automatic: a PagerDuty action on a prompt version event, sent to the agent service, incident type and severity you choose (see Automations) |
Requirements
- A PagerDuty REST API key with read and write access to Services and Incidents, and read access to Escalation Policies and Users. See PagerDuty's API access keys.
- At least one escalation policy in PagerDuty.
- In ANTS Platform, the Owner role, or a custom role with access to Integrations (notification integrations) and admin access to Governance. See Members & Roles.
Connect PagerDuty
Open the integration
Go to Project Settings > Integrations and open the PagerDuty card.
Enter the API key
Select Connect PagerDuty and paste the key into PagerDuty API Key. ANTS Platform checks the key before saving and encrypts it at rest.
Choose the defaults
Under PagerDuty Settings:
| Setting | Notes |
|---|---|
| Default Escalation Policy | Used for new agent services. Required before services can be created. |
| Default Urgency | High (immediate attention) or Low (can wait for business hours) |
| Auto-Create Services | Lets Sync All Agents create a service for each agent |
| Incident Creation | When off, no incidents are sent |
| Event Routing | Allows ANTS Platform to send resolve events |
| Alert Grouping | Turns on alert grouping for new services |
Create agent services
Under Agent Services, select Sync All Agents. ANTS Platform creates a PagerDuty service with an Events API v2 integration for each agent that does not have one yet. You can change a service's escalation policy or delete it from the table.
Send a test event
Under Test Integration, send a test event for an agent and resolve it to confirm the routing.
Incidents are listed under Recent Incidents on the PagerDuty page, and on the PagerDuty Incidents page with type, severity, status and agent.
Per-agent alerting
Automatic incidents for detected violations follow each agent's policy in Governance > AI Guardrails. In the agent's policy, the PagerDuty Alerting section has:
| Setting | Notes |
|---|---|
| Enable toggle | Turns automatic PagerDuty alerts for this agent on or off |
| PagerDuty Service | A specific service, or Auto-select (use any synced service) |
| Alert Severity | CRITICAL, HIGH, LOW or INFO |
| Auto-resolve Alerts | Saved with the policy. ANTS Platform does not send resolve events for remediated violations today, so resolve incidents in PagerDuty. |
When no service is set, ANTS Platform uses the agent's own service, and falls back to the most recent synced service in the project. If the project has no synced service, no incident is sent.
Deployment notes
Automatic incidents are queued by the data store service and sent by the
worker. On self-hosted deployments, the data store service setting
PAGERDUTY_INCIDENT_PUBLISHING must not be false, and the worker setting
QUEUE_CONSUMER_PAGERDUTY_QUEUE_IS_ENABLED (default true) must stay on.
Troubleshooting
- "Sync All Agents" is disabled. Set a Default Escalation Policy first.
- No incident for a violation. Check that Incident Creation is on, that the project has a synced agent service, and that PagerDuty alerting is on in the agent's AI Guardrails policy.
- Key rejected. The key needs the permissions listed under Requirements.