sustainable-computing-io/kepler

Kepler is CNCF Sandbox project to attribute the energy use of Kubernetes workloads. Let's make every watt matter. ⚡

What it solves

Kepler (Kubernetes-based Efficient Power Level Exporter) solves the challenge of measuring and attributing energy consumption in cloud-native environments. It provides granular visibility into the power usage of individual containers, pods, and nodes within a Kubernetes cluster, which is essential for monitoring the environmental impact of compute-intensive workloads like AI training and inference.

How it works

Kepler operates as a Prometheus exporter that collects energy metrics from the underlying hardware. It utilizes the RAPL (Running Average Power Limit) interface and other hardware monitoring frameworks to track power consumption. It then maps these hardware-level metrics to specific Kubernetes abstractions (pods and containers) using active resource usage data, allowing users to see exactly how much power their specific applications are consuming.

Who it’s for

It is designed for platform engineers, SREs, and developers working in cloud-native environments who need to monitor, optimize, or report on the energy efficiency of their infrastructure and workloads.

Highlights

  • Provides granular energy consumption metrics at the container, pod, and node level.
  • Uses a service-oriented architecture with improved performance and reduced resource overhead.
  • Features enhanced security by requiring only read-only access to host system files, removing the need for high-privilege capabilities.
  • Supports various hardware interfaces including RAPL, HWMon, NVIDIA GPUs, and Redfish BMC.
  • Integrates seamlessly with standard Kubernetes monitoring stacks via Prometheus.

Related

  • Project
    google-ai-edge/model-explorer

    Model Explorer is an open‑source visualizer for AI model graphs (TensorFlow, TFLite, TF‑JS, MLIR, PyTorch Exported Program, etc.). It shows hierarchical, interactive diagrams, highlights inputs/outputs, supports search, GPU‑accelerated rendering, and can be extended with custom adapters. Install via `pip install ai-edge-model-explorer` and run `model-explorer <model>`. A web component is also available as an npm package.

  • Project
    utkuozdemir/nvidia_gpu_exporter

    A versatile Prometheus exporter that collects NVIDIA GPU metrics via nvidia-smi or NVML, providing a simple monitoring solution for consumer hardware, homelabs, and mixed-environment clusters.

  • Project
    ottosulin/awesome-ai-security

    A curated directory of open-source tools, frameworks, and educational resources designed to help developers and security professionals secure AI systems against emerging threats.

  • Project
    google/uncertainty-baselines

    A collection of high-quality, standardized baselines and templates for researchers to benchmark uncertainty and robustness in deep learning models.