Cost spike
Spend compared with its normal range for that hour, per scope. Sudden or sustained.
Anomaly vs baseline band
+312% spike
7-day median · per hour-of-day
Smart AlertingBudgets, spikes and new workloads
Kubernetes cost alerts that reach the team that owns the spend.
Budgets per team or namespace, spike detection against each hour’s normal range, and a flag when a new expensive workload appears. Sent to Slack, email or a webhook.
- Budgets
- Cost spikes
- New expensive workloads
Example · payments-prod · 7 days
Firing+312% spike
Sat · 12:48 UTC
Baseline
$24 /hr
Observed
$92 /hr
Excess (12h)
+$816
Alert
Cost spike
Routing · alerts → policies → channels
Route alerts by team and severity.
Policies set daily caps, quiet hours, and which channel gets each severity. Channels: Slack, email and webhook.
- Reusable policiesOne policy, many rules. Share across teams.
- Delivery historyDelivery history per alert: sent, deferred, or suppressed.
Alerts
Policy
Channels
Cost spike · ml-prod
+$1,8402m
Budget burn · data-platform
78% of $12K18m
New workload · checkout-api-v2
+$420 / mo1h
Policy
payments-prod
- modeimmediate
- cap10 / day
- quiet22:00–06:00 IST
- severityP1 · P2 · P3
Slack
#finops-alerts
finops@example.com
Webhook
jira.example.com/hook
Example · last 24h · 14 / 14 deliveredSlack · email · webhook
What changed
Spike alerts list the workloads that drove the increase.
Each cost spike alert names the workloads behind the increase and the Kubernetes changes around it: image, replicas, requests and HPA.
- Kubeadapt AI summaryA short plain-English summary of each incident and a suggested next step.
- Unexplained shareEach alert shows how much of the increase the Kubernetes changes explain, and how much they don’t.
- Typed causesReplica, image, request and HPA changes, each tied to the workload it happened on.
Example incident
ml-prod · Cost spikeCost spike on ml-prod · checkout-worker driving 35% of the increase
Recommended actionReview checkout-worker resource changes
+$620/day+312% vs 14d baselineFirst time in 30 days
Top contributors
checkout-worker
CPU request changed
+$220/day
35%
ml-inference
Container image upgraded
+$180/day
29%
feature-store
Replica count changed
+$80/day
13%
$480 of the $620 increase is linked to Kubernetes changes.
Three alert types
Spikes, budgets and new workloads.
Each rule watches one scope. Spike rules learn that scope’s normal range; the alert floor and thresholds can be changed per rule.
Scopes a rule can watch
- Organization
- Cluster
- Namespace
- Team
- Department
Budget threshold
Monthly or quarterly budgets, with warning and limit thresholds.
Monthly burn
78%
day 23 / 31
50%80% warn100% capNew expensive workload
Flagged on first stable appearance.
First-seen stable
flagged
checkout-cache
namespace · payments
$1,240
/mo · NEW
Auto-excludes Argo · Spark · Ray · Kubeflow
Install in 5 minutes
Stop paying for unused capacity.
Free up to 100 vCPU. After that, $1.99 per vCPU per month, dropping to $0.60 at volume.
Agent runs in your clusterNo credit card