Correlates related alerts into a single incident instead of flooding the team with separate alarms. It's opt-in per routing rule: switch a rule's Alert grouping setting to
Auto (AI)
, and the engine takes over from there, learning which alerts tend to fire together over time. No grouping keys to set.
When multiple alerts fire close together, the engine works down a signal ladder to decide whether, and why, they belong together, checked in order:
  • Same alarm rule
    firing on multiple targets, always merges. A rule firing across many hosts becomes one group with many instances, not a separate ticket per host
  • A shared identity label;
    both key and value must match
    (host_name=fedora
    and
    host_name=ubuntu
    do not match)
  • Co-fire history
    , alerts that have repeatedly fired together before
  • A known cause-effect pattern
    , e.g. a DynamoDB throttle causing Lambda errors
  • Topology
    , the entity graph connects the two resources
  • An LLM pass
    , for novel combinations nothing above can explain
Each group shows its members and a
"Why grouped"
reason. You can thumbs-up or thumbs-down a grouping decision; a single thumbs-down is a one-off override, but a second person flagging the same pairing stops it from grouping that way again. Groups link directly into
Run RCA
for investigation.
Screenshot 2026-07-31 at 11