Learn
What is agent on-call?
Agent on-call is the human + automation rotation when agent workloads need live response — incident triage, Ask queue backlog, connector outages, or runaway spend — with playbooks, not ad-hoc panic in chat.
In depth
Heartbeats and autonomous roles run 24/7; someone must answer when the system pages a human. Agent on-call covers:
- Ask queue SLA — approvals waiting block external work - Connector failures — OAuth expired, API down - Quality regressions — eval alert or spike in human reject rate - Cost anomalies — budget threshold breached - Security signals — unusual tool access pattern
Runbooks: pause role, rollback skill version, switch to read-only connectors, notify customers if needed.
On-call is not 'watch the agent chat all day.' It is pager-driven response to structured alerts from Neuro OS governance and monitoring.
Examples
- Pager — Ask queue > 20 items; on-call approves or reassigns
- Stripe connector 401 — on-call refreshes token; finance role resumes
- Reject rate spike — rollback skill; open incident doc
- Weekend — only critical heartbeats enabled; rest paused
Related terms
FAQ
Do agents replace human on-call?
No — they reduce noise. Humans handle exceptions and approvals.
What belongs in a runbook?
Pause role, rollback skill, contact owner, customer comms if external impact.
Run governed agent roles on a company OS — not only definitions in a glossary.