Why most K8s cost tools need Prometheus (and nable doesn't)
The usual path, Kubecost and similar, runs an agent in your cluster that scrapes Prometheus for pod-level usage, then joins it to cloud prices. That is deep, but it means deploying and maintaining infrastructure in every cluster before you see a number.
nable takes the read-only shortcut: it connects to any cluster reachable via your kubeconfig, reads the running workloads and node types through the Kubernetes API, and attributes cost from the node/instance prices. No agent, no Prometheus, no write access. You trade some per-second granularity for a cost picture in minutes with nothing installed.
How it works
1. Point it at your cluster
nable uses KUBECONFIG (or ~/.kube/config) and is multi-cluster aware, pass a context to switch clusters.
2. Ask in your editor
"What is my Kubernetes cost by namespace?" nable reads the workloads and returns cost per namespace, workload, and label, ranked.
3. See it with the rest of the bill
K8s is one line in your whole bill, next to AWS/Azure/GCP and AI/LLM spend, not a separate silo.
# in Claude or Cursor, after `uvx nable`
You: Kubernetes cost by namespace this month
nable: $8,120 total. checkout $2,940 · search $1,810 · batch $1,240 ...
batch is 3x its request/limit, likely over-provisioned.See your Kubernetes cost in minutes, with nothing to install in the cluster. nable reads your kubeconfig locally, shows cost per namespace and workload, and puts it next to your cloud and AI bill, answered in Claude or Cursor. Free for solo use, runs on your machine.
Common questions
Can you monitor Kubernetes cost without Prometheus?
Yes. nable reads Kubernetes cost from your kubeconfig through the Kubernetes API and attributes it from node prices, with no Prometheus and no in-cluster agent. You get per-namespace and per-workload cost in minutes. A Prometheus-based tool like Kubecost goes deeper on per-pod, per-second allocation if you need it.
Does nable install anything in my cluster?
No. nable is read-only and runs on your machine. It connects via your existing kubeconfig, reads workloads through the Kubernetes API, and deploys nothing into the cluster. Your kubeconfig never leaves your machine.
Does it work with multiple clusters?
Yes. nable is multi-cluster aware, it uses your kubeconfig contexts, so you can ask about a specific cluster or compare them.
How is this different from Kubecost?
Kubecost runs an in-cluster agent (usually with Prometheus) for deep per-pod real-time allocation. nable reads kubeconfig with nothing installed and covers Kubernetes alongside your cloud and AI spend in one tool. See the full comparison at /nable-vs-kubecost.