NVIDIA ovpackage

ovpackage is a command-line tool and async Python API for reproducible asset packaging and publishing. It snapshots a directory of assets into an immutable, named package version, publishes that version to a repository on local disk, Amazon S3, Azure Blob, Google Cloud Storage, an Omniverse Storage API endpoint, installs a chosen version into a workspace, and later proves exactly what was installed — using content-hash catalogs, gitignore-style .wrappignore rules, .wrapp dependency metadata, and commands for diffing, mirroring, patching, freezing, and offline export/import. It is built for asset-heavy build pipelines, shared content libraries, CI/CD promotion, offline delivery, and agents that need deterministic, verifiable package operations across storage backends.

python -m pip install -e .[local]

Optional backends are available through extras: s3, azure, gcs, and storage-api.

python -m pip install -e .[local,s3,azure,gcs,storage-api]

Check the installed command:

ovpackage --help
ovpackage version

• Package: A named collection of files.
• Version: An immutable snapshot of a package. Use a new version string for every changed snapshot.
• Repository: A storage location that contains a .packages directory with package versions.
• Install location: A workspace path where a package version is materialized for a project.
• Package metadata file: A JSON file named .<package>.wrapp that records package metadata and dependencies. The .wrapp extension is part of the package format.
• Ignore file: .wrappignore uses gitignore-style rules and is picked up by create, catalog, and freeze workflows.
• Catalog: A JSON description of a directory tree, including paths, sizes, and hashes.

Create and install a package using only local files:

mkdir -p demo/source demo/repo demo/workspace
printf "hello assets\n" > demo/source/readme.txt

ovpackage create sample_assets 1.0.0 demo/source --repo demo/repo --progress-report off
ovpackage list-repo demo/repo
ovpackage install sample_assets 1.0.0 demo/workspace/sample_assets --repo demo/repo --progress-report off

Track installed package dependencies in a project-level metadata file:

mkdir -p demo/project
ovpackage new demo_project 1.0.0 demo/project
ovpackage install sample_assets 1.0.0 demo/project/assets \
    --repo demo/repo \
    --package demo/project/.demo_project.wrapp \
    --progress-report off
ovpackage status demo/project/.demo_project.wrapp --progress-report off

ovpackage create textures 2026.05.11 ./textures --repo s3://asset-packages --progress-report off
ovpackage list-repo s3://asset-packages

ovpackage install textures 2026.05.11 ./project/assets/textures \
    --repo s3://asset-packages \
    --progress-report off

ovpackage catalog ./assets assets-before.json --progress-report off
ovpackage catalog ./assets assets-after.json --progress-report off
ovpackage diff assets-before.json assets-after.json --show

ovpackage mirror textures 2026.05.11 \
    --source-repo s3://dev-packages \
    --destination-repo s3://release-packages \
    --recursive \
    --progress-report off

ovpackage export textures 2026.05.11 --repo s3://release-packages --dedup --progress-report off
ovpackage import textures.2026.05.11.tar --repo ./offline-repo --progress-report off

Install only the extras you need:

omniverse:// Nucleus URLs are also supported when the required client libraries are available in the environment.

Authentication can be passed with --auth or environment variables. Prefer OVPACKAGE_* variables in new automation, for example OVPACKAGE_REPO, OVPACKAGE_AUTH, OVPACKAGE_LOG_FILE, and OVPACKAGE_PROGRESS_REPORT.

The public import namespace is ovpackage:

import ovpackage

print(ovpackage.__version__)

Most command implementations are asynchronous. A typical application initializes storage backends through ContextManager and then calls package operations with that scheduler. See tests such as tests/test_api.py and tests/test_ovpackage_aliases.py for executable API examples.

This repository includes an agent skill at .agents/skills/ovpackage-cli/. Agents that support repository-local skills can use it directly. Otherwise, copy or symlink that directory into the agent's skill search path.

The skill contains a command reference, storage/auth notes, and multi-step recipes. It is meant to compensate for the intentionally minimal initial documentation set.

• Published package versions are intended to be immutable. Publish a new version instead of changing files under an existing package version.
• Package metadata filenames use .<package>.wrapp; keep that filename format.
• Ignore rules use .wrappignore; keep that filename format.
• Object stores do not preserve empty folders. Add placeholder files if empty folders are meaningful to your pipeline.
• Use --progress-report off in automation and agent runs to avoid interactive progress output.

Install test dependencies:

python -m pip install -e .[local,test]

Run the default public test set:

make test

Run emulator-backed backend tests:

make start-test-services
make test-emulators
make stop-test-services

The GitHub workflow in .github/workflows/tests.yml runs both the default public tests and emulator-backed S3, Azure, and GCS tests.

This project is currently not accepting contributions.

Generated version: 0.2.1.