kpm v0.6.0 · early development · organic launch
Stop guessing requests
from a single kubectl top
Tiny Go CLI that samples pod CPU and memory into local SQLite, recommends rightsizing from history, and opens charts with kpm serve — no Prometheus tax, no cluster writes.
Live traffic on gghstats · Cosign signed releases · MIT · 100% organic
Sample. Remember. Recommend. Analyze on your laptop.
Built for operators who oversize for safety
History beats a snapshot
kubectl top is a moment. kpm keeps a SQLite trail so p95 vs requests is honest — oversize and undersize show up as patterns, not vibes.
Analyze-local charts
Copy the store to a laptop, run kpm serve, open workload cards and time windows. Collection can stay off. Optional live Metrics when kubeconfig works.
Advisory recommend — GitOps safe
Recommendations never apply. Copy or download Kustomize-style patches; you own the merge. Production changes stay in your pipeline.
client-go only · bastion-first
No kubectl subprocess. Distroless GHCR image, Cosign signatures, SBOM. Operator manifests live in kpm-selfhosted.
How it works
- 01
Install
Verified install script, go install, release tarball/deb/rpm, or the GHCR image.
- 02
Sample into SQLite
Print a sample profile, point kube context, run one-shot or daemon sampling into a local store.
- 03
Recommend / serve
Rightsizing from history, optional notify and MCP, and kpm serve for analyze-local charts.
Install
Primary path: the Cosign-verified install script. Go install until the Homebrew tap ships. ghcr.io/hrodrig/kpm:v0.6.0
Go install
go install github.com/hrodrig/kpm/cmd/kpm@latestOne liner (Linux / macOS)
curl -fsSL https://get.kpm.hermesrodriguez.com/install.sh | shOperator manifests · Contract: kpm-selfhosted · SPECIFICATIONS
FAQ
Is kpm ready as my only rightsizing tool in production?
Not yet. Until v1.0.0 treat it as early development. Recommendations are advisory — you own production resource changes. Use at your own risk under the MIT License.
Do I need Prometheus, VPA, or Goldilocks?
No. kpm is bastion-first: Metrics API in, SQLite out. Heavy stacks stay optional. Pair with them later if you want; kpm does not require them.
Does it change my workloads?
Never. Read-only against the cluster. Recommend patches are copy/download only — kpm does not apply CPU or memory patches live.
Does it need kubectl installed?
No. kpm talks to the API with client-go. You still need a valid kubeconfig or in-cluster credentials for sampling and optional live enrichment.
Where do I run it?
Bastion cron or systemd, docker run, or in-cluster CronJob and Deployment with a PVC for the store. See kpm-selfhosted for operator notes.
How do I verify a release?
GitHub Releases ship Cosign signatures and SBOMs. The get install script verifies with cosign when available.
Get the binary. Keep the history.
Install, sample, recommend — then open charts with kpm serve. Pre 1.0: evaluate before you trust it alone in production.
Install kpm