Cast AI shows what each cluster, namespace and workload costs, next to error rate and latency. CastPilot answers in plain English, and your AI agents connect through MCP. One lightweight install, with data in minutes.
Trusted by 2,100+ companies
Running a cost tool yourself becomes your job, cost and health live in separate tools, and the bill arrives as one number once a month, when it is an investigation instead of a recap.
average CPU utilization across Kubernetes clusters.
See what each cluster, namespace and workload costs, and group spend by team, product or environment with allocation groups.
Throughput, error rate and latency (P50, P95, P99) sit beside cost for every workload, so you see what a change did to both.
CastPilot, the built-in AI agent, answers questions about your cluster costs and health in plain English. On paid plans, the MCP Server connects the agents and automation your team is building to the same live data.
Cost anomaly detection flags spikes minutes after they happen, and the savings view shows where requested resources exceed what workloads use. On paid plans, set a budget and get an alert when spend drifts.
See GPU utilization, memory and spend per workload, and find idle capacity.
āI had an aha moment ā an iPhone moment ā with Cast. Literally two minutes into the integration, we saw the cost analytics, and I had an insight into something I had never had before and had tried to get for a very long time.ā
Install Cast AI with castctl, Helm or Terraform. Cast Radar only reports, so nothing in your cluster changes.
Open cost and reliability for every cluster, namespace and workload, and check the numbers against your bill and your own tools.
Use the findings and the savings simulation to decide what to change. Cast Radar reports and never changes your cluster.
Install with castctl, Helm or Terraform/GitOps alongside the monitoring tools you already run. Paid plans add Prometheus API endpoints for Grafana and an MCP Server for your own AI agents.
One lightweight, read-only install
castctl Ā· Helm Ā· Terraform / GitOpsLive context across every
cluster, namespace and workload
Explore, compare and export
Ask questions in plain English
Connect to live data via MCP
MCP Server Ā· Paid plansReports only. No cluster changes. Full CastPilot and Prometheus API access are included on paid plans.
Provisioned, requested and actual CPU, GPU and memory, plus unscheduled pods.
Storage cost sits next to compute, so overprovisioning shows up as dollars. Network cost includes cross-AZ, split by egress and ingress.
One year of history, CSV export, and Prometheus API on paid plans.
Roll up cost across every cluster, with workload insights per team.
Cast Radar is priced by average monitored CPU per month. Clusters are unlimited on every plan, and each plan is one flat monthly fee up to its ceiling.
| Plan | Average vCPUs | Price |
|---|---|---|
| Free | Below 250 | $0 |
| Starter | 250 to 400 | $249 a month |
| Team | 400 to 900 | $499 a month |
| Scale | 900 to 1,500 | $699 a month |
| Above 1,500 | 1,500+ | Talk to us |
| Capability | Free (below 250 CPUs) | Paid (from 250 CPUs) |
|---|---|---|
| Cost reports by cluster, namespace, workload | ā | ā |
| Cluster cost and efficiency reports | ā | ā |
| Allocation groups (cluster level) | ā | ā |
| Org-level rollups and allocation groups | ā | ā |
| Reliability reporting | ā | ā |
| GPU cost and utilization reports | ā | ā |
| Storage cost reports | ā | ā |
| Network cost with cross-AZ attribution | ā | ā |
| Cost comparison | ā | ā |
| Savings simulation | ā | ā |
| CSV export | ā | ā |
| Queryable history | 1 year | Unlimited |
| Deploy with castctl, Helm, Terraform/GitOps | ā | ā |
| Anomaly detection, budgets, alerts | ā | ā |
| Automation preview (Workload Optimization, Autoscaler) | ā | ā |
| Prometheus API endpoints | ā | ā |
| CastPilot, full AI agent | ā | ā |
| MCP Server | ā | ā |
| Support | ā | Standard |
If an account past 250 CPUs does not convert after the 30-day trial, it steps down to a limited tier: 30 days of history at hourly resolution, 7-day cost comparisons, savings simulation without the optimized-configuration view, and CSV export.
Reports first. Automation is a separate decision whenever you are ready for it, on the same platform.
Canāt find what youāre looking for?
Contact usOn paid plans, the built-in MCP Server exposes cost and health data to the agents and automation your team is already building, so nobody checks another dashboard.
A lightweight install in your cluster reports usage, cost and performance data to Cast AI. Install with castctl, Helm or Terraform/GitOps. Data typically shows up 5 to 10 minutes behind your cluster, and 15 at most.
Self-hosted cost tools mean you own the deployment, the metrics storage and the upgrades. Cast Radar is a managed SaaS: one lightweight install, reliability signals next to cost, CastPilot investigation in plain English, and flat monthly pricing. If you are happily running OpenCost on your own platform, that is a legitimate choice. Cast Radar is for teams that would rather read reports than run them.
No. Cast Radar only reports, and it has no automation. It never changes your cluster.
Costs are estimates based on published cloud prices, so they will not match your invoice to the cent, especially if you have discounts. Use them to see which workload drove a change. Cast Radar shows the difference between periods and flags spikes, so you can find what changed.
Yes. It sits next to your existing monitoring, and on paid plans you can pull its data into Grafana or your own tools through the Prometheus API. If you pay for Kubecost today, Cast Radar can replace it.
No. Cast Radar covers Kubernetes cost, and it fits best when Kubernetes is the main part of your cloud bill.
Cast Radar is free below 250 average CPUs, with no credit card. If your account grows past 250 CPUs, a 30-day trial of the paid plan starts automatically. If you donāt upgrade, you step down to a limited tier.
#1 on G2262 reviewsConnect one cluster in minutes. Full clarity for your team, and for your agents. Free below 250 CPUs.
