Documentation

Aitra is AI Infrastructure Intelligence, from model to facility. Aitra Meter is the first component shipping: it measures and attributes AI infrastructure efficiency, computing joules per output token across every workload, model, and GPU. The canonical documentation lives in the open source repository. This page is a map into it.

Start here

Guides

• Getting started install and first measurement
• Energy providers NVML, AMD, DCGM, Zeus
• Writing a provider add an inference server or energy backend
• Operations upgrading, scaling, air-gapped install
• Cost budgets per-namespace budgets and alerting
• KEDA integration
• OpenCost integration
• Kepler integration
• Troubleshooting

Install

• Install walkthrough Helm install, labeling GPU nodes, verifying pods
• Helm chart chart source and values
• Air-gapped and offline install

Providers

• Energy providers NVML (default), AMD, DCGM, Zeus
• Compatibility matrix provider × hardware × inference server
• Inference server configs vLLM, TGI, SGLang, Ollama examples

Metrics

• Metrics reference every Prometheus metric, type, and label
• Glossary J/token, CV gating, attribution, and more
• Alerting rules

Configuration

• Configuration reference all Helm values and CRD fields
• HTTP API reference aggregation service API
• Example configurations site configs, measurement policies, KEDA, Grafana

Concepts and design

• Positioning where Aitra Meter fits next to DCGM Exporter, OpenCost, and Kepler
• Technical specification
• Architecture Decision Records
• Roadmap · Changelog

Contributing

• Contributing guide how to add a provider, energy backend, or measurement agent
• Governance
• Code of conduct
• Security policy

Try Aitra Meter