Alert Instructions

When you enable AI Agent Analysis, you can add instructions that act as a step-by-step playbook. The agent reads them when the alert fires and uses the instructions to run the right checks and suggest next steps.

Add instructions

In the alert editor, enable AI Agent Analysis, select + instructions, enter your text, choose Save & Confirm, then save the alert.

How it works

Instructions are tied to the alert rule and run on every trigger. Use them to specify what to check, where to look, and what to try next.

When creating metric alerts, you'll need to link a dashboard and panel for additional context.

Next, you can use the instructions to reference the dashboards, panels, queries, links, or any other element in your environment that can help the analysis. Instructions support plain text, and up to 2,000 characters.

Writing guidelines

For best results, keep it short, specific, and action-oriented. Use the exact names of dashboards, panels, services, etc.

For example:

Check Explore for `service.name:<SERVICE>` over last 15m; focus on `level:error` and exception keywords.
Look for a recent deploy in releases table; if found, diff config for `<SERVICE>`.
If a single operation drives >40% of errors, capture the last message.