Skip to main content

AI-BOM Generator

PyPI CI Release License

AI-BOM Generator is a small Python 3.12-3.14 CLI and GitHub Action for producing an AI/ML bill of materials from a local model project directory.

It records declared model metadata, in-root MODEL_CARD.md paths, model or checkpoint digests, training-code references, dependency files and parsed Python packages, dataset references, prompt templates, and eval artifact references, then exports CycloneDX JSON 1.7 or an SPDX 3.0.1 AI Profile preview.

It is an evidence reporter. It is not a model registry, vulnerability scanner, legal compliance engine, dataset auditor, or AI governance platform.

Install

py -3.12 -m pip install ai-bom-generator
ai-bom --version
ai-bom --help

With uv, you can run the published package without installing it into the current environment:

uv run --python 3.12 --with ai-bom-generator ai-bom --version
uv run --python 3.12 --with ai-bom-generator ai-bom --help

Quickstart

Clone the repository so the bundled minimal model project is available:

git clone https://github.com/0disoft/ai-bom-generator.git
cd ai-bom-generator
uv sync --locked
$out = Join-Path ([System.IO.Path]::GetTempPath()) ("ai-bom-example-" + [System.Guid]::NewGuid().ToString("N"))
New-Item -ItemType Directory -Path $out | Out-Null
uv run --python 3.12 ai-bom generate examples/minimal-model-project --format cyclonedx-json-1.7 --output (Join-Path $out "bom.cdx.json") --warning-report (Join-Path $out "warnings.json") --summary (Join-Path $out "summary.json")
Get-ChildItem -LiteralPath $out

Expected files:

bom.cdx.json
summary.json
summary.json.manifest.json
warnings.json

The output directory also retains one dot-prefixed *.ai-bom.lock coordination file per generated destination. These files contain no project evidence; their stable paths let overlapping writers coordinate across later invocations.

For the bundled minimal project, summary.json reports "status": "success", "completeness_status": "complete", "artifact_count": 1, and "warning_count": 0.

The command writes:

  • bom.cdx.json: CycloneDX JSON 1.7 BOM.
  • warnings.json: machine-readable warning report.
  • summary.json: generation status and output summary.
  • summary.json.manifest.json: output-set manifest when the default manifest path is used.

Generated output paths must resolve outside the target model project directory.

Minimal Config

Example aibom.toml:

schema_version = "1"

[output]
format = "cyclonedx-json-1.7"

[model]
name = "minimal-example-model"
version = "0.1.0"
model_card = "MODEL_CARD.md"
license_declared = "NOASSERTION"

[artifacts]
include = ["models/*.safetensors"]

[[dependencies]]
path = "requirements.lock"
type = "pip"

[[datasets]]
name = "minimal-example-dataset"
license_declared = "NOASSERTION"

Explicit uv.lock, poetry.lock, Pipfile.lock, requirements-file, and unified conda-lock v1 YAML references are parsed locally into package components by default. Use type = "conda-lock", type = "poetry", or type = "pipenv" when a supported lockfile uses a non-standard filename. Set parse = false on a dependency reference to keep file-level evidence only. Includes, constraints, editable requirement directives, environment solving, resolution, downloads, and automatic lockfile discovery are not performed.

Parsed package components preserve directly evidenced source locators, indexes, channels, platforms, revisions, and artifact hashes. Missing provenance stays absent, and malformed source fields produce warnings instead of invented values.

When --config is omitted, the CLI reads aibom.toml from the target model directory if that file exists. It does not search parent directories or alternate filenames. Artifact discovery defaults are opt-in with [artifacts].discovery = true. The spdx-ai export format is available as a partial SPDX 3.0.1 AI Profile preview; it explicitly marks unavailable SPDX AI fields instead of inventing missing metadata. GitHub Marketplace registration is deferred.

CLI

ai-bom --version
ai-bom generate <model-directory> [--config <path>] --format <cyclonedx-json-1.7|spdx-ai> --output <bom.json> --warning-report <warnings.json> --summary <summary.json> [--manifest <manifest.json>] [--error-report <error.json>]

Missing optional metadata is reported as warnings without pretending the BOM is complete. Unreadable required files, invalid config, unsupported exporters, unsafe paths, and invalid generated BOM output fail with non-zero exit codes. Add --error-report <path> when CI needs a strict-redacted, versioned JSON failure envelope instead of parsing terminal output. Successful runs remove a stale report at that requested path.

GitHub Action

- uses: actions/checkout@v7

- id: ai-bom
  uses: 0disoft/ai-bom-generator@v0.6.1
  with:
    model-directory: .
    warnings: allow

- name: Verify AI-BOM outputs
  shell: bash
  run: |
    set -euo pipefail
    test -s "${{ steps.ai-bom.outputs.bom-path }}"
    test -s "${{ steps.ai-bom.outputs.warning-report-path }}"
    test -s "${{ steps.ai-bom.outputs.summary-path }}"
    test "${{ steps.ai-bom.outputs.status }}" = "success"

The action prepares Python 3.12 and pinned uv 0.11.28, disables the setup-uv GitHub cache, and invokes the packaged CLI with uv run --project --locked. Its virtual environment and uv download cache stay under RUNNER_TEMP; the caller repository is not used for action runtime state. When format or warnings inputs are omitted, the action lets the CLI use the discovered or explicit config values and CLI defaults. Generated files are written to explicit paths when provided, or to a run-unique directory under RUNNER_TEMP. Explicit relative output paths resolve from GITHUB_WORKSPACE. redaction: off is an intentional local-debugging escape hatch that can place secret-shaped values in generated artifacts; strict redaction is still best effort, so do not put credentials in declared evidence or lockfile locators.

Summary-derived action outputs are published only when the generation manifest matches the BOM, warning report, and summary files from the current run.

Use @v0 for compatible 0.x updates, or pin the exact @v0.6.1 tag when a workflow needs release reproducibility. GitHub-enforced immutable releases apply to versions published after v0.2.0.

Validation

Local validation source of truth is VALIDATION.md.

uv sync --locked
uv run --python 3.12 python -m compileall -q src tests scripts
uv run --python 3.12 ruff check src tests scripts
uv run --python 3.12 python -m unittest discover -s tests -v
uv build
uv run --python 3.12 python scripts/verify_wheel.py dist
uv run --python 3.12 python scripts/verify_github_action.py

Post-release verification:

$env:RELEASE_VERSION = "0.6.1"
$env:PUBLISH_RUN_ID = "<successful-publish-run-id>"
$env:SMOKE_RUN_ID = "<successful-exact-version-action-smoke-run-id>"
uv run --python 3.12 python scripts/verify_release.py --version $env:RELEASE_VERSION --publish-run-id $env:PUBLISH_RUN_ID --smoke-run-id $env:SMOKE_RUN_ID

Project Docs

Security

Report suspected vulnerabilities privately through the repository Security tab. Do not include real credentials, model weights, prompts, or private datasets. See SECURITY.md for supported versions and reporting guidance.

License

Apache-2.0.

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

ai_bom_generator-0.6.1.tar.gz (148.9 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

ai_bom_generator-0.6.1-py3-none-any.whl (125.4 kB view details)

Uploaded Python 3

File details

Details for the file ai_bom_generator-0.6.1.tar.gz.

File metadata

  • Download URL: ai_bom_generator-0.6.1.tar.gz
  • Upload date:
  • Size: 148.9 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.13

File hashes

Hashes for ai_bom_generator-0.6.1.tar.gz
Algorithm Hash digest
SHA256 3a8ac77425848223a8a4cfb003c22bb7d847bdb844c8f2bfd378152edc75c8a1
MD5 9fa3aec1f7aa17a1c1286f17552b4061
BLAKE2b-256 ab33b965426e0e47925e8b7f7918153f8417e49692a4813aeb1db40ee63f180a

See more details on using hashes here.

Provenance

The following attestation bundles were made for ai_bom_generator-0.6.1.tar.gz:

Publisher: publish-pypi.yml on 0disoft/ai-bom-generator

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file ai_bom_generator-0.6.1-py3-none-any.whl.

File metadata

File hashes

Hashes for ai_bom_generator-0.6.1-py3-none-any.whl
Algorithm Hash digest
SHA256 3b5751c04eb54deebd57a6347f7ae5f79da46030dc9744148251e0c2f6154d7b
MD5 3275ca0d6cbb54f2f1c01dba7040b405
BLAKE2b-256 d50ab1a23340598647680778d3e4fba042ffd22e764c39a3598bce21c398c871

See more details on using hashes here.

Provenance

The following attestation bundles were made for ai_bom_generator-0.6.1-py3-none-any.whl:

Publisher: publish-pypi.yml on 0disoft/ai-bom-generator

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Release history Release notifications | RSS feed

This release

0.6.1 This release

2 files

0.6.0

2 files

0.5.0

2 files

0.4.0

2 files

0.3.1

2 files

0.3.0

2 files

0.2.2

2 files

0.2.1

2 files

0.2.0

2 files

0.1.4

2 files

0.1.2

2 files

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page