OverviewFinOps & Budget

Alerts

View as Markdown

Alerts help teams respond before AI usage becomes expensive, unreliable, or hard to explain.

Alephant alerts should point back to the relevant workspace, department, member, agent, Virtual Key, policy, request, or run trace so operators can investigate quickly.

Alert Types

Alert typeTypical triggerCommon action
Budget alertWorkspace, department, agent, member, or Virtual Key budget passes a thresholdReview spend, throttle traffic, or raise budget
Usage anomalyRequest count, token usage, or cost spikes compared with recent historyInspect run traces and recent workflow changes
Quota alertMonthly recorded-log or usage quota approaches plan limitsUpgrade, export, or reduce noisy traffic
Operational alertError rate, latency, fallback rate, or cache behavior changesInvestigate provider, routing, or policy state
Positive alertCache savings or routing savings reaches a notable valueReview optimization impact

Severity Levels

SeverityMeaningExample
HighNeeds immediate attentionBudget usage above hard-stop threshold or request failure spike
MediumNeeds planned reviewBudget usage above warning threshold or unusual request growth
LowInformationalCache savings, policy change, or successful routing optimization

Budget Alerts

Budget alerts can apply to:

  • Workspace budgets
  • Department budgets
  • Agent budgets
  • Member budgets
  • Virtual Key budgets
  • Paid endpoint or workflow spend limits

Common thresholds are warning, critical, and hard stop. A warning should notify operators; a hard stop should block or pause the relevant traffic according to policy.

Anomaly Alerts

Usage anomaly alerts are useful when an agent or workflow changes behavior unexpectedly:

  • Request volume increases sharply.
  • Token usage per request grows.
  • A workflow starts retrying repeatedly.
  • A fallback route becomes the primary route.
  • A paid endpoint begins losing margin per call.

The investigation path should start from the alert and move into logs, run traces, analytics, and policy history.