# Kubernetes Cost Optimization | LevelFour

> Kubernetes cost optimization on EKS, GKE, AKS and any cluster: each pod's cost from your own bill, workloads and nodes rightsized in place or by pull request.

Source: https://levelfour.ai/kubernetes

Updated: 2026-11-11

Install it once. Your clusters stay rightsized.

LevelFour reads what each pod costs from your own bill and rightsizes it, in place or by pull request, inside limits you set once. It runs on EKS, GKE, AKS and any cluster on Kubernetes 1.23 or later, and measures each saving on your bill.

[Read the docs](https://docs.levelfour.ai/connectors/kubernetes)

## Kubernetes savings that keep shipping

Teams set requests at deploy time and move on, and your cloud bill charges for every core those requests reserve. The LevelFour agent reports what each container uses, and LevelFour turns the gap into savings that ship in place with the automation, or as pull requests your pipeline applies.

- **Each pod's cost, from your own bill**: LevelFour reads each node's cost from your Cost and Usage Report or billing export, with your commitments applied, and splits it across the containers that ran on it. You see cost per cluster, namespace, workload and node, the forecast for the month, and what a vCPU or a GiB costs you a day.
- **What you reserve next to what you use**: For CPU, memory and storage, LevelFour shows what the nodes provide, what pods request and what they use, hour by hour, with what each level costs. Recommendations size each request from 14 days of usage, and LevelFour works out the cheapest set of nodes each NodePool's pods fit on.
- **Reliability beside the bill**: OOM kills, restarts, CPU throttling and evictions sit next to cost, cluster by cluster, and the workloads at risk come first. LevelFour sizes memory from each container's peak and holds back a cut on a suspected memory leak. After OOM kills, it raises memory.
- **Resized in place, inside your guardrails**: Turn the automation on for a namespace and it cuts oversized requests on running pods, at most 20% a step. It also fixes OOM kills and CPU throttling without a restart, and gives new pods the values it reached. Maintenance windows, minimums, maximums and a ceiling in your Helm install set how far it goes. [How the automation works](https://docs.levelfour.ai/connectors/kubernetes/automation)
- **A pull request when you want one**: Without the automation, each change arrives as a pull request on the repository that deploys the workload, and your pipeline applies it. NodePool and node group changes come the same way, and LevelFour merges the classes of change you allow. A week after a change holds, LevelFour measures the saving on your bill.
- **Kubernetes in your terminal and your AI agent**: The l4 CLI lists Kubernetes costs and savings with --provider k8s, as tables for you or JSON for your scripts. Over the LevelFour MCP server, Claude, Cursor or any other MCP client reads the same clusters. [See the CLI](https://docs.levelfour.ai/cli)

## Kubernetes documentation

The docs cover the install with Helm or GitOps, what the agent reads, the automation, recommendations and node optimization. The API, the CLI and the SDKs read the same costs and savings.

[Read the docs](https://docs.levelfour.ai/connectors/kubernetes)

## Kubernetes FAQ

### Which Kubernetes platforms does LevelFour support?

Amazon EKS, Google Kubernetes Engine, Azure Kubernetes Service and any other cluster on Kubernetes 1.23 or later, including clusters on your own hardware, priced from node prices you set. An EKS or GKE cluster without the agent still shows cost per namespace from the bill.

### What does the LevelFour agent collect?

The agent runs as one pod. It reads how each workload is configured, samples each container's CPU and memory every 30 seconds, and sends percentiles once an hour. Changes to your cluster come from pull requests, or from the optional automation component in the namespaces you allow.

### How does LevelFour size container requests?

From 14 days of hourly usage by default: CPU from its P99 and memory from its peak, each with headroom set by how spiky the workload is, and never below what the container needs to start. A JVM keeps its heap plus the memory it uses outside it.

### Does LevelFour resize pods on its own?

Only where you turn on the optional automation component, in the namespaces your Helm install lists. It resizes running pods in place on Kubernetes 1.33 or later, a cut takes at most 20% of a request per step, and Stop and revert puts every pod back on its own spec. [How install and forget works](https://levelfour.ai/blog/automated-kubernetes-cost-optimization.md).

### How does LevelFour work with HPA, KEDA and VPA?

When an HPA or KEDA scales on CPU or memory utilization, LevelFour changes its target together with the request, so it keeps scaling at the same real usage. Requests a VPA writes stay with the VPA.

### Can I connect many clusters at once?

Yes. One install key connects as many clusters as you run, through Argo CD, Flux or Terraform, and each cluster registers itself under its own name on its first report.

### How is Kubernetes priced?

Kubernetes Savings comes with every plan: up to 50 active pods on Pro, 500 on Scale, and unlimited on Enterprise. Pods above a plan's limit cost from $1.40 per pod a month.

## About LevelFour

The FinOps platform that actually cuts costs. One system of record for Cloud, SaaS, and AI spend. LevelFour applies each saving, or your engineers merge it as code.

- **Cloud Savings**: 30% success fee on savings found and applied. No hidden charges, only pay on what you actually save.
- **Pro**: $30 per month · Up to $2,500 spend/mo
- **Scale**: $200 per month · Up to $7,500 spend/mo
- **Enterprise**: Custom · Unlimited spend

[Pricing](https://levelfour.ai/pricing.md) · [Guide for agents (llms.txt)](https://levelfour.ai/llms.txt) · [Documentation](https://docs.levelfour.ai) · MCP server: `https://mcp.levelfour.ai/mcp`
