KubeSpend DocsKubernetes cost, measured honestly.

How KubeSpend prices your clusters, where the numbers come from, and what it will not pretend to know. Install in about a minute; no cloud billing access required.

QuickstartHow optimization worksWhat is eBPF?

What is KubeSpend

A metrics agent in your cluster, a server that prices it against real AWS rates, and a console. Start here.

Introduction →

The cost model

How an instance price becomes a per-pod number — the 9:1 vCPU-to-GiB split, node-hours, and root volumes.

Cost model →

Cost attribution

Per-workload and per-namespace cost from requests, and why idle capacity is reported separately instead of spread around.

Attribution →

Rightsizing

The thresholds behind every recommendation — 25% headroom over observed peak, a 30% waste floor, and how savings are computed.

Recommendations →

eBPF, explained

What eBPF is, why it is the right tool for network measurement, and what it costs you to run.

eBPF primer →

AWS network costs

Cross-AZ, NAT, inter-region and internet egress — the charges that are invisible in a Kubernetes dashboard.

Network costs →

What we measure

An honest inventory of what is priced today, what is approximate, and what returns "unavailable" rather than zero.

Coverage →

FAQ

Agent overhead, self-hosting, kernel requirements, security posture, and why a number sometimes reads N/A.

Questions →

Start here

If you are installing KubeSpend for the first time, read Quickstart — it gets a cluster reporting in about a minute and needs nothing but an API key.

If you are evaluating whether the numbers can be trusted, read The cost model and then What we measure. The second page is deliberately blunt about the gaps.

If you are here for the network story, start with What is eBPF and follow it into AWS network costs.

The one rule worth knowing up front

Every dollar figure in KubeSpend traces to a real published cloud price. When a price cannot be resolved — an unsupported region, a missing instance type, no credentials on the server — the value renders as unavailable, never as $0 and never as an estimate.

This matters more than it sounds. A cost tool that quietly substitutes a fallback number is worse than one that admits a gap, because you cannot tell the two apart from the dashboard. What we measure is the full inventory of which dimensions are live, which are approximate, and which are not implemented yet.