Aitra Meter

Try Aitra Meter

Aitra is AI Infrastructure Intelligence, from model to facility, and Aitra Meter is the first component shipping. It measures joules per output token across every workload, model, and GPU in your Kubernetes cluster, then derives cost and carbon per token. It deploys entirely inside a single cluster: one Helm install, no infrastructure changes, no changes to inference application code. The commands below mirror the repository README.

Prerequisites

A Kubernetes cluster with at least one GPU node.
kubectl configured for that cluster.
helm (v3).
NVIDIA GPUs work out of the box via NVML (pure Go, no sidecar). AMD, DCGM, and Zeus are also supported, see the energy providers guide.

Step 1: Label your GPU nodes

kubectl label node <your-gpu-node> aitra-ai.github.io/gpu=true

Step 2: Install with Helm

helm repo add aitra https://aitra-ai.github.io/helm-charts
helm repo update

helm install aitra-meter aitra/aitra-meter \
  --namespace aitra-system --create-namespace \
  --set cluster.name=my-cluster \
  --set siteConfig.electricityCostPerKwh=0.12 \
  --set siteConfig.carbonIntensityFallback=400

All pods reach Running within about 60 seconds.

Step 3: Verify the pods are running

kubectl get pods -n aitra-system

Step 4: Open the dashboard

kubectl port-forward -n aitra-system svc/aitra-meter-dashboard 3000:3000
# then open http://localhost:3000

Once inference traffic flows through a supported server (vLLM, TGI, SGLang, or Ollama), aitra_j_per_token and the derived cost and carbon metrics appear per workload, model, and hardware. For a complete walkthrough, follow the getting started guide.

Next steps

Full getting started guide All documentation GitHub repository