IAKG v0.1.0
Release Notes — v0.1.0
Released: 2026-08-03
IAKG is the Audel toolchain generalised. What began as a set of scripts for
one specific collection is now a package with a scope that matches what it
actually does: download books from the Internet Archive, clean the OCR text into
Markdown, and ingest the result into a DocKG knowledge graph. The package is
ia_kg, the command is iakg, and nothing about it assumes a particular
collection any more.
What changed
The Audel assumption is gone, in the name and in the code. src/audel_kg/
became src/ia_kg/, and the hardcoded ALL_GENRES = ["audel-electric"] that sat
at the top of both download_ia.py and ingest.py was replaced with
_discover_genres(), which scans the corpus/ subdirectories at runtime.
--genre accepts free-form strings rather than a fixed choice list, and is
inferred from the catalog filename stem when omitted — so adding a genre means
creating a directory, not editing source.
A real CLI. The scripts are now a Click application with two subcommands:
download, covering single books, catalogs, search, and survey; and ingest,
which builds the knowledge graph. Both are reachable as iakg <subcommand>.
The text pipeline is under test. Nine tests cover the parts most likely to
silently corrupt a corpus: slug generation, ligature normalisation, smart-quote
cleaning, hyphen-joining across line breaks, page-number removal, index
stripping, running-header removal, heading detection, and the final Markdown
conversion. These are the transformations that make OCR output usable, and they
were previously unverified.
Curating a catalog before you download it. iakg download search --export-catalog FILE writes every search result to a draft catalog with each
line commented out, so nothing downloads by accident. You uncomment the entries
you actually want and feed the file to iakg download catalog — which turns an
all-or-nothing search into a reviewable list. docs/catalog-workflow.md walks
the whole path: search, export, curate, test a single book, bulk download,
survey, ingest.
Dependency floors now point at software that exists. kgmodule-utils was
pinned at >=0.2.0 and doc-kg at >=0.12.3, both long dead; they are now
>=0.10.0 and >=0.21.0, with pycode-kg>=0.21.4. One behavioural consequence
worth knowing: from doc-kg 0.20.0 the vector backend defaults to "sqlite-vec"
outright instead of being inferred per-store from whatever happens to be on
disk, and lancedb is no longer a core dependency — it moved to an opt-in
extra needed only to read a pre-0.20.0 store.
Repository hygiene brought in line with the rest of the fleet. Pre-commit
hooks (whitespace, EOF, YAML/TOML validation, merge-conflict and large-file
guards, ruff lint and format), a GitHub Actions CI workflow with separate lint
and test jobs, a README following the doc_kg conventions, and a .gitignore that
actually excludes the KG artifacts (.pycodekg/, .dockg/) and wires up the
shared KGRAG model cache. Packaging moved from [tool.poetry] metadata to the
PEP 621 [project] table, and doc-kg moved from a git source to PyPI. The CI
lint and test jobs were also using poetry install --only dev, which targets
dependency groups — dev here is a PEP 621 extra, so both jobs were
erroring; they now use --extras dev. A tag-triggered release workflow rounds
this out.
Upgrading
This is the first release under the new name, so for most people it is an
install rather than an upgrade:
pip install ia-kg
iakg --helpIf you were running the old audel_kg package, two things change. The import
path is ia_kg, not audel_kg — the old package is deleted, not aliased. And
the genre list is no longer compiled in: corpus/ is the source of truth, so any
genre you want available needs a directory there. Existing corpus/ layouts are
discovered as-is and need no migration.
Full changelog: CHANGELOG.md