AIO-grounded imports and maintained, evidence-backed artificial intelligence concept records. AIMech extends the Mech curation workflow to AI; it is not a fork of the upstream ontology or a claim that ontology membership demonstrates a model's performance, safety, or causal mechanism.
- Artificial Intelligence Ontology on GitHub
- AIO on BioPortal
- AIO documentation
- Joachimiak et al., The Artificial Intelligence Ontology: LLM-Assisted Construction of AI Concept Hierarchies, Applied Ontology 19(4), doi:10.1177/15705838241304103.
The bootstrap pins AIO release 2026-07-22, upstream commit
397851c547d1c4d8f06ef184f63c1d40bc86937a. It contains 442 named AIO
classes spanning Bias, Layer, Machine Learning Task, Mathematical Function,
Model, Network, Preprocessing, and Training Strategy. BioPortal's total may
include external classes; AIMech counts only class IRIs in the AIO namespace.
Every generated record is NOT_REVIEWED. This is a deterministic import of
terminology, not an independent scientific review. The paper supplies ontology
context, not blanket evidence for every imported claim. The retained source
snapshot preserves the complete original export; the projection is not an OWL
reasoner and does not replace that export.
Concepts absent from AIO can be curated under data/curated/ with stable
AIMech: IDs, inspected primary evidence, claim-specific citations and explicit
AIO mappings or documented no-match assessments. These maintained records remain
DRAFT; scientific reviews are saved separately. See curated concepts.
uv sync --locked --extra dev
just check
just vendored-check # verify the public, immutable CLAW pin
uv run --locked aimech seed # read-only preview
uv run --locked aimech seed --check # verify byte-for-byte reproduction
uv run --locked aimech validate
uv run --locked aimech validate-curatedjust check is the offline local and CI quality gate. CI additionally runs
just vendored-check, which compares shared artifacts with the public CLAW
revision in scripts/.vendored_canon_ref. No API keys or live research
providers are needed. seed --write creates missing files only; it refuses to
overwrite changed records or delete unexpected files. Source upgrades are
explicit, reviewed changes to the snapshot and manifest, not automatic pulls.
| Path | Purpose |
|---|---|
sources/aio/ |
Pinned upstream JSON, checksum, citation, and license |
src/aimech/schema/aimech.yaml |
Native LinkML contract |
src/aimech/importer.py |
Deterministic source-to-record projection |
data/aio/*.yaml |
Generated per-concept records; do not hand-edit |
data/curated/*.yaml |
Maintained AIMech-native concepts, evidence and AIO mappings |
src/aimech/schema/curated_concept.yaml |
Native curated-record contract |
.claude/skills/ |
Record/category review and repository workflows |
reviews/structured/ |
Timestamped schema-validated review YAML and derived Markdown |
history/ |
Append-only curation-session sidecars created when work occurs |
Review issues belong to the upstream concept, importer, or maintained native record, according to ownership. See the review profile and source policy. No website, benchmark evaluator, or live model-execution capability is claimed by this bootstrap.
Use $aimech-add-record to resolve a
concept, search the web and literature, inspect abstracts and PDFs, check for
duplicates, and add an evidence-backed native record or restore a missing AIO
record. It also supports upstream-term/source-update proposals. The skill is exposed
through the repository's .agents/skills link.
New AIMech code is BSD-3-Clause. AIO source data and derived concept records
remain CC-BY-4.0, attributed to the AIO authors. See LICENSE-DATA
and the unchanged upstream files under sources/aio/. Shared CLAW artifacts
retain their own notices and are governed by the pinned upstream manifest.