Skip to content
NewKubeOn AI: a managed LLM gateway with cost per 1M tokens for every model
KubeOn docs
API referenceReleases

Enterprise evaluation guide

A practical acceptance plan for platform engineering, FinOps and security, covering both Kubernetes and AI.

On this page

1. Agree on scope and ownership

Choose a representative cluster and billing account for Kubernetes, or one provider and model deployment for AI. Name an engineering owner, a FinOps reviewer and a security contact. Agree on the billing window, cost basis, deployment location and success criteria before connecting.

2. Review access and data flow

  • Review the permissions reference and deployment architecture.
  • Choose KubeOn Cloud or a self-hosted hub. Record who operates storage, backups, keys and upgrades.
  • For AI, separate read-only provider access from gateway invocation credentials. Confirm route logging and region settings.
  • Agree on access roles, retention, support expectations and the exit procedure in the evaluation scope.

3. Validate the baseline

AreaEvidence to reviewAcceptance decision
Kubernetes allocationInvoice window, matched resources, shared and idle policies, unallocated spendFinOps can explain the total and any reconciliation difference
OwnershipKnown team labels and namespace mappingsTeam-level totals match the agreed ownership policy
AI tokenomicsBilling buckets, token counts, provisioned and reservation chargesEffective cost uses the same window and denominator across models
Data healthHeartbeats, billing freshness and missing metricsCoverage is sufficient for the decision being evaluated

4. Validate a controlled action

For Kubernetes, review one proposal and apply it through your pipeline in a suitable environment. For AI, send a controlled set of requests through one gateway route and test fallback and budget behavior. Compare service health and subsequent costs with the baseline. Savings estimates become realized savings only after measurement.

5. Decide and expand

  • Record acceptance results and unresolved gaps with an owner for each.
  • Agree on the next clusters, teams, providers or applications and their change windows.
  • Confirm the edition, measured billing units, deployment responsibilities and support terms in the quote.
  • Retain a rollback and offboarding plan for agents, integrations and gateway routes.

Bring the evaluation team

Book a walkthrough to review this plan together. The initial demonstration can use sample data; connecting production data is a separate evaluation step.
Need help with this guide?Contact support