Kerf compared with other rightsizing tools¶
Kerf takes ideas from the tools below. They are good tools: pick the one that fits.
| License | Your data stays in your infrastructure | Applies changes | In-place resize | Rollback on bad health | Savings in $ | HPA and JVM aware | |
|---|---|---|---|---|---|---|---|
| Kerf | Apache-2.0 | ✓ | ✓ policies, maintenance windows | ✓ | ✓ | ✓ | ✓ both |
| Vertical Pod Autoscaler | Apache-2.0 | ✓ | ✓ evict and webhook | ✓ InPlaceOrRecreate |
– | – | not with an HPA on the same metric |
| Goldilocks | Apache-2.0 | ✓ | through VPA, opt-in | through VPA | – | – | – |
| Robusta KRR | MIT | ✓ CLI; the UI is SaaS | optional webhook, with a Robusta account | – | – | – | --allow-hpa |
| IBM Kubecost | commercial, free tier | ✓ self-hosted option | Kubescaler (alpha) | – | – | ✓ | – |
| StormForge | commercial | metrics go to SaaS | ✓ | ✓ | ✓ | ✓ | ✓ both |
| PerfectScale | commercial, free tier | ✓ self-hosted option | ✓ maintenance windows | ✓ | – | ✓ | JVM; HPA advice |
| CAST AI | commercial | SaaS console | ✓ | ✓ | – | ✓ | ✓ both |
✓ means documented. – means we did not find it in the tool's public docs on 2026-11-05, which is not proof that it is missing. Products change: check the source before you rely on a cell. The sources for every cell are listed on their own page, with notes on each tool. Corrections are welcome: open an issue or a pull request.
Which one fits¶
- One small cluster that you watch closely. A Vertical Pod Autoscaler in recommendation mode, or Goldilocks on top of it, is enough. Kerf is more than you need.
- A quick report from Prometheus. Robusta KRR gives recommendations from the command line.
- Automation as a commercial product with vendor support. StormForge, PerfectScale and CAST AI automate most of what Kerf does.
- That automation, open source, inside your own infrastructure. This is what Kerf is for: hub, agents and UI run in your clusters, and usage data stays in your Prometheus and your PostgreSQL.
What Kerf adds to a VPA¶
A Vertical Pod Autoscaler recommends and can apply requests in one cluster. Kerf adds:
- savings in dollars per workload, namespace and cluster, priced per node type;
- one UI for a fleet of clusters;
- a mode per resource: CPU and memory are
off,auditorauto-patchseparately (modes); - maintenance and freeze windows, and guardrails that cap each decrease;
- rollback when a change is followed by OOM kills, restarts or CPU throttling;
- HPA coordination and JVM-aware memory.
Not a fit (yet)¶
- Node-level autoscaling or bin packing: use Karpenter or Cluster Autoscaler next to Kerf.
- Teams that want a managed SaaS with vendor support.