Aitra is an open source project to measure, understand, and optimize how AI uses infrastructure, from the model down to the facility. It starts with measurement and attribution you can trust, then builds the shared standards and tools the industry needs.
Aitra's scope spans the whole stack, from the model to the facility. Some of this ships today in Aitra Meter. The rest is being defined in the open with the working group.
Aitra Meter is the first component of Aitra to ship. It measures and attributes AI infrastructure efficiency from model to facility, computing joules per output token continuously across every workload, model, and GPU, then deriving cost and carbon per token. It installs with a single Helm chart and exposes everything as Prometheus metrics. This is the one product shipping today. Everything else in Aitra's scope is being defined in the open with the working group.
Everything Aitra Meter measures and ships, with the v0.3.0 additions woven in: the DCGM energy provider, the model-level AI-efficiency metric family, cost-budget and TTFT alerting, per-GPU and per-model attribution, and Grafana as the default dashboard surface.
aitra_model_energy_per_1m_tokens and aitra_model_tokens_total.EnergyProvider interface, no hardware lock-in.Aitra Meter reads GPU power directly via NVML (default), DCGM, or the Zeus community sidecar, correlates it with token output from your inference server, and computes J/token continuously, per workload, per model, per hardware tier.
Aitra Meter connects GPU hardware energy to AI output volume at the token level. It runs entirely inside a single Kubernetes cluster, a DaemonSet agent, an aggregation service, a SQLite store, and a dashboard. One Helm install, no changes to your inference server code.
Read the architecture docsaitra_j_per_token
Joules per output token, workload × model × hardware. The energy cost of a single generated token.
aitra_co2_per_token_grams
gCO₂ per token, J/token × grid carbon intensity. Track and report your inference carbon footprint.
aitra_cost_per_million_tokens_usd
$/M tokens, J/token × electricity cost. The per-token cost visible to finance and platform teams.
aitra_idle_time_ratio
Fraction of the last hour spent idle per node. Surface over-provisioned GPU capacity instantly.
aitra_model_energy_per_1m_tokens
Joules per one million output tokens, by model, the headline primitive of the v0.3.0 model-level AI-efficiency family.
Designed to drop into your existing cloud native stack, not replace it. Works alongside the tools you already run on Kubernetes.
aitra_* metrics available instantly in PromQL.gen_ai.infrastructure.energy.* metrics to any OTel collector. Opt-in via a single Helm value.aitra_j_per_token and aitra_idle_time_ratio. Reference ScaledObjects included in examples/keda/.examples/grafana/./metrics endpoints. Extensible InferenceMetricsProvider interface for custom servers.Aitra Meter is community-driven. All components are available under the Apache 2.0 License on GitHub. There are three high-value contribution paths, a new inference server, a new energy backend, or a new measurement agent in any language.
Aitra is an open source SODA Foundation project for AI infrastructure intelligence, from model to facility. The working group shapes where it goes next: measurement, energy and hardware providers, integrations, dashboards, and docs. Join to get working group updates and an invite to the monthly call.
Takes a minute · your details go to a private roster, not published.
Aitra Meter is a SODA Foundation project with public governance, a published roadmap, and open maintainership. See the project leadership and maintainers.
