nemotron-customizer-airgap

What problem does it solve?

Teams running Nemotron Customizer in disconnected environments need a reliable way to package the right launcher and execution container images, along with all required offline Python dependencies and baked-in git repo overlays, without downloading models or code from the internet at runtime.

Core Features & Use Cases

• Plans airgap bundles from a single YAML: validates selected step targets, expands declared dependencies, groups steps into minimal execution-image sets, and produces an airgap manifest.
• Generates offline-ready container artifacts: builds a launcher image plus one or more execution-image tarballs suitable for loading on airgapped clusters.
• Bakes repo overlays for auto_mount inputs: treats ${auto_mount:git+...} as a connected-machine input and bakes pinned repo overlays into derivative execution images so airgapped jobs do not clone from GitHub.
• Supports resumable builds: uses airgap-build-state.yaml to resume long-running probe/build phases after failures.
• Applies runtime overlay configs: provides guidance and example overlays that clear runtime git auto-mounts while preserving persistent storage mounts.

Quick Start

Use the nemotron-customizer-airgap skill to first plan your bundle by running the airgap runner in plan mode and reviewing deploy/nemotron-customizer/airgap/out/airgap-manifest.yaml before executing any Docker build.