Money
Allocated cost from OpenCost, a bundled price catalog (AWS, Azure, OVHcloud, Scaleway) or your own rates. The share expected to reach the bill is shown separately, with its realization factor.
VPA, Goldilocks, KRR, cloud consoles and in-house scripts all produce lists of things to optimize. The hard part comes next:
GreenKube closes that loop.
GreenKube’s own measurements, corroborated by VPA and Goldilocks, Robusta KRR and any tool through a push API, in one model with their provenance.
Sources →GreenKube’s measured value decides each size; the other sources raise or lower the confidence, and contested advice is labelled. Each recommendation carries evidence, a projected impact, a risk level, a confidence score and a risk-adjusted expected value.
Recommendations →GreenKube finds the workload’s Git source through Argo CD or Flux, renders the exact diff, and opens a pull request on GitHub, GitLab or Gitea once you approve that diff.
GitOps delivery →After the merge, GreenKube waits for Argo CD or Flux to sync the merged revision and checks that the new configuration is really live in the cluster.
Sync and apply →Each change is measured against a counterfactual (what the workload would have cost without it), with a 90% confidence interval and confounder checks, and compared with its projection.
Savings measurement →Outcomes go into an append-only, hash-chained ledger linked to the pull request, merge commit, synced revision and measurement. CSV and FOCUS-aligned exports.
The ledger →Every change is reported in three dimensions, each with its own units and its own label (projected or measured):
Money
Allocated cost from OpenCost, a bundled price catalog (AWS, Azure, OVHcloud, Scaleway) or your own rates. The share expected to reach the bill is shown separately, with its realization factor.
CO₂e
Operational emissions (energy times grid intensity from Electricity Maps, Wattnet or regional defaults) and embodied emissions, side by side with cost on every recommendation.
Released capacity
CPU cores and memory freed, projected and measured. Never converted to money, and the honest result on a fixed-capacity or on-premises cluster where the bill only moves when a node is removed.
No admission webhooks, no in-place mutation, no ignoreDifferences drift. Every change is a pull request your team reviews, and a rollback is a pull request too.
Projected, measured allocated and realized (billed) savings are separate methods, labelled everywhere, and never added together.
When a change cannot be automated safely, GreenKube refuses and says why, with the fix. A signal that cannot be read is reported as unavailable, never counted as a pass.
Every source is a plugin of the same Source SDK, and the evidence travels with the change into the pull request. A source alone never creates a recommendation: GreenKube’s own measurement is required.
GreenKube runs in the cluster it optimizes, on-premises and on clouds: AWS, GCP, Azure, OVHcloud, Scaleway, kubeadm, k3s or bare metal. One Helm install deploys the API, the controller, an isolated renderer and PostgreSQL. There is no profile and no feature flag: discovery, previews, delivery, sync tracking and measurement activate themselves when their prerequisites appear, and the Setup page lists what is active and the fix for anything missing.
# The demo: the full dashboard with sample data, no cluster neededdocker run --rm -p 9000:9000 greenkube/greenkube demo --no-browser --port 9000# Open http://localhost:9000# Install in your clusterhelm repo add greenkube https://GreenKubeCloud.github.io/GreenKube && helm repo updatehelm install greenkube greenkube/greenkube -n greenkube --create-namespaceYou can also explore the hosted demo at demo.greenkube.cloud.
GreenKube is open source under the Apache 2.0 license. Contributions are welcome, especially new recommendation sources and support for more GitOps layouts.

Creator & Lead Developer