Skip to main content

​
Every agent, on one screen

On Call → Agents (/ops/oncall) is the on-call area’s landing page: a card grid titled On-Call Agents, one card per agent in the current project. A project usually runs several agents with different objectives — one watching payments, one watching the batch jobs — and the grid shows them all side by side, each with its own status and wiring. The header counts them (“4 agents in this project”) and carries the two entry points: Escalation Teams and New Oncall.

​
What a card tells you

Each card is a complete health check of one agent, without clicking into it:
A card showing “No escalation configured — nobody gets paged when this agent finds an incident.” deserves attention before the incident happens. Wire a default policy or a rotation from the agent’s settings — though note the organization’s default policy, if one exists, still applies as a fallback.
Two controls live directly on the card, no menu needed:
  • Pause agent / Start agent — the play-pause button in the corner stops and resumes the agent in place.
  • The ⋮ Agent actions menu — View Details, Edit Settings, and Delete (with a confirmation; deleting is irreversible).
Clicking anywhere else on the card opens the agent: the live agent-team view when the AI Agents badge is present, the detail page (tabs Overview, Incidents, Executions, Team) otherwise. The sidebar stays with you either way.

​
Creating an agent

Press New Oncall. Two prerequisites apply:
  • The project needs at least one connected log provider — without one you are taken to the Integrations page first.
  • Creation requires an active subscription or trial.
The five-step wizard walks Foundation (provider and monitoring mode), Alert Rules, Response Team, Notifications, and Review & Launch. The new agent appears as a card when you finish. See Set up on-call for the full walkthrough.

​
Editing an agent

⋮ → Edit Settings opens Configure Oncall Agent, the agent’s full-width editor:
  • Basic Settings — Agent Name, Agent Status (Active / Paused / Inactive), Focus Description (what the card shows as the objective — be specific about the systems, logs or metrics to watch), and the Integration the agent reads from. An agent with no integration assigned stays inactive, and the editor says so.
  • Monitoring Mode — Autonomous Mode (AI-driven cadence, with a Conservative / Balanced / Aggressive Autonomous Level) or Fixed Interval (Check Frequency from every 5 minutes to hourly).
  • Alert Rules — the severity thresholds that turn findings into alerts.
  • On-Call Rotations — the rotations and the Escalation Policy select. Leaving the policy on “Derive from the rotations below” makes each rotation one escalation level, 15 minutes apart.
  • Cooldown & Anti-Spam Control — how long the agent holds fire after alerting.
  • AI Agent Team — the Enable AI Agent Team switch: a team of specialized DevOps/SRE agents that investigates alerts, queries logs, analyzes infrastructure and populates the knowledge base. With the team on, the agent’s cadence becomes coordinator-driven — checking more often during active incidents and backing off when the system is healthy.
  • Notification Channels — Discord, Slack and Microsoft Teams, side by side. The Slack app and Teams app connect here.
A sticky save bar appears the moment anything changes; Save Changes applies it all at once.
Pausing from the card sets the agent Inactive. The distinct Paused state is set from the editor’s Agent Status field — both stop the agent; they differ only in what the badge communicates to your team.

​
Who can do what

Everyone in the project sees the agent cards. Creating, editing and configuring agents requires the on-call configure capability — the Admin and Owner roles. The backend enforces the same rule on every action.

Agentic on-call

What the AI agent team actually does when it investigates.

Escalation policies

The chains behind the card’s Escalation panel.

Set up on-call

The guided setup, end to end.

Notifications

Every channel a page can reach.