User Documentation

July 21, 2026 · View on GitHub

Documentation for platform operators deploying and operating GPU-accelerated Kubernetes clusters using AI Cluster Runtime (AICR).

Audience

This section is for users who:

  • Install and use the aicr CLI tool
  • Deploy the AICR agent to capture cluster snapshots
  • Generate recipes and bundles for their environments
  • Use the API for programmatic configuration generation

Documents

DocumentDescription
InstallationInstall the aicr CLI (automated script, manual, or build from source)
End-to-End TutorialLearning path: install → recipe → bundle → deploy → validate, start to finish
CLI ReferenceComplete command reference with examples for all CLI operations
Generating BundlesTask-oriented how-to: deployers, value overrides, node scheduling, offline/vendored charts, readiness gates
Artifact VerificationVerify bundles and recipe-evidence: trust levels, public-trust, KMS/PEM keys, offline, and CI gating
API ReferenceREST API quick start and endpoint documentation
Agent DeploymentDeploy the Kubernetes agent for automated cluster snapshots
ValidationTask-oriented walkthrough: validate a cluster against a recipe, both training and inference performance phases
Component CatalogEvery component that can appear in a recipe
Slurm AccountingConfigure Slinky Slurm accounting with a customer-operated MariaDB database
Container ImagesContainer image inventory across all components (BOM)
Recipe HealthPer-recipe health and validation status tracking
Air-Gap MirrorMirror images and charts for air-gapped deployment

Quick Start

# Install aicr CLI (Homebrew)
brew tap NVIDIA/aicr
brew install aicr

# Or use the install script
curl -sfL https://get.aicr.run | bash -s --

# Generate a recipe for your environment
aicr recipe --service eks --accelerator h100 --intent training -o recipe.yaml

# Create deployment bundles
aicr bundle --recipe recipe.yaml -o ./bundles

# Deploy to your cluster
cd bundles && chmod +x deploy.sh && ./deploy.sh

Validating a cluster instead of deploying? See Validation for a task-oriented walkthrough covering deployment, performance (training and inference), and conformance phases.