How-to Guides
Connect a cluster
On-demand and spot pricing works everywhere out of the box. Cloud account connections only matter for commitments like RIs, Savings Plans, and CUDs.
You ran the Helm install and the cluster shows up in the dashboard. For most environments, you're done — Kubeadapt fetches on-demand and spot prices from each cloud's public pricing API and attributes node cost from the first snapshot. The only reason to take further setup steps is to reflect your commitment discounts (Reserved Instances, Savings Plans, Committed Use Discounts) in node-level pricing, and that's a per-provider story.
Pick your provider
| Provider | On-demand & spot | Discount attribution | Setup needed |
|---|---|---|---|
| AWS | Out of the box | RIs + Savings Plans (live) | IAM role via Cloud Accounts → AWS |
| GCP | Out of the box | CUDs (planned) | None today; on-demand and spot already work |
| Azure | Out of the box | RIs + Savings Plans (planned) | None today; on-demand and spot already work |
| On-premises | Via custom pricing | n/a — set your own rates | Helm values block for CPU, RAM, storage, GPU |
Which one do I need?
Three short paths to the answer:
- You run on AWS and use Savings Plans or Reserved Instances. Connect the account via the AWS integration. Per-node prices then reflect your blended rate, and Cost Explorer shows what you actually pay rather than the on-demand sticker.
- You run on GCP or Azure. No integration step exists today. On-demand and spot pricing already work; CUD attribution (GCP) and Reservation attribution (Azure) are on the roadmap. The agent install from the Quick start is everything you need.
- You run on bare metal, in a colo, or self-managed in any environment without a public pricing API. Follow the On-premises pricing guide to drop your real rates into Helm values.
If you're not sure which case applies, the agent install works the same way for all of them — connect first, then come back and add discount attribution when you need it.
What the agent ships, regardless of provider
The same agent and the same chart cover every provider above. The cluster sends a compressed snapshot to Kubeadapt every 60 seconds over HTTPS — see How Kubeadapt works for the data path. The differences across providers are entirely about price resolution: which pricing source Kubeadapt reads, and whether commitment discounts get applied.
For the full Helm value reference, see Configuration overview. For provider-specific gotchas around node labels, machine-type naming, and spot/preemptible terminology, follow the link in the table above.