This guide describes how to install the AI Cluster Runtime (AICR) CLI tool (aicr) on Linux, macOS, or Windows.
What is AICR: AICR generates validated configurations for GPU-accelerated Kubernetes deployments. See README for project overview.
- Operating System: Linux, macOS, or Windows (via WSL)
- Kubernetes Cluster (optional): For agent deployment or bundle generation testing
- GPU Hardware (optional): NVIDIA GPUs for full system snapshot capabilities
- kubectl (optional): For Kubernetes agent deployment
brew tap NVIDIA/aicr
brew install aicrInstall the latest version using the installation script:
curl -sfL https://raw.githubusercontent.com/NVIDIA/aicr/main/install | bash -s --To install to a custom directory instead of the default /usr/local/bin:
curl -sfL https://raw.githubusercontent.com/NVIDIA/aicr/main/install | bash -s -- -d ~/binOptional: if you hit GitHub API rate limits, set GITHUB_TOKEN before running the install command. No special repository scope is required for public releases.
This script:
- Detects your OS and architecture automatically
- Downloads the appropriate binary from GitHub releases
- Installs to
/usr/local/bin/aicrby default (use-d <dir>for a custom location) - Verifies the installation
- Uses
GITHUB_TOKENenvironment variable for authenticated API calls (avoids rate limits)
Supply Chain Security: AICR includes SLSA Build Level 3 compliance with signed SBOMs and verifiable attestations. See SECURITY for verification instructions.
- Download the latest release
Visit the releases page and download the appropriate binary for your platform:
- macOS ARM64 (M1/M2/M3):
aicr_<version>_darwin_arm64.tar.gz - macOS Intel:
aicr_<version>_darwin_amd64.tar.gz - Linux ARM64:
aicr_<version>_linux_arm64.tar.gz - Linux x86_64:
aicr_<version>_linux_amd64.tar.gz
- Extract and install
# Example for Linux x86_64
tar -xzf aicr_linux_amd64.tar.gz
sudo mv aicr /usr/local/bin/
sudo chmod +x /usr/local/bin/aicr- Verify installation
aicr --versionRequirements:
- Go 1.26 or higher
go install github.com/NVIDIA/aicr/cmd/aicr@latestCheck that aicr is correctly installed:
# Check version
aicr --version
# View available commands
aicr --help
# Test snapshot (requires GPU)
aicr snapshot --format json | jq '.measurements | length'Expected output shows version information and available commands.
Enable shell auto-completion for command and flag names:
Bash:
# Add to ~/.bashrc
source <(aicr completion bash)Zsh:
# Add to ~/.zshrc
source <(aicr completion zsh)Fish:
# Add to ~/.config/fish/config.fish
aicr completion fish | sourceAICR is also available as container images for integration into automated pipelines:
docker pull ghcr.io/nvidia/aicr:latest
docker run ghcr.io/nvidia/aicr:latest --versiondocker pull ghcr.io/nvidia/aicrd:latest
docker run -p 8080:8080 ghcr.io/nvidia/aicrd:latestSee CLI Reference for command usage
If aicr is not found after installation:
# Check if binary is in PATH
echo $PATH | grep -q /usr/local/bin && echo "OK" || echo "Add /usr/local/bin to PATH"
# Add to PATH (bash)
echo 'export PATH="/usr/local/bin:$PATH"' >> ~/.bashrc
source ~/.bashrc# Make binary executable
sudo chmod +x /usr/local/bin/aicrSnapshot GPU measurements require nvidia-smi in PATH:
# Verify NVIDIA drivers
nvidia-smi
# If missing, install NVIDIA drivers for your platform# Remove binary
sudo rm /usr/local/bin/aicr
# Remove shell completion (if configured)
# Remove the source line from your shell RC file- Documentation: User Documentation
- Issues: GitHub Issues
- API Server: See Kubernetes Deployment