KG-Registry kg-registry

active curating

KG-Registry is a registry of knowledge graphs and the resources they are built from: data sources, ontologies, data models and aggregators. One record per resource: each record holds the resource’s description, domains, taxa, license, contacts, publications and activity status, and lists its products (graph dumps, APIs, browsers, mappings, documentation) with the upstream resources each product was derived from. Those provenance links make the registry a graph of where biomedical knowledge comes from and where it goes. Agents expand stub records and refresh stale ones with skills kept in the repository, scripted syncs pull records from the OBO Foundry, FRINK, KG-Bioportal and Translator, and every agent change arrives by pull request. KG-Registry is not strictly a Mech: its records are metadata about other resources rather than evidence-backed claims, and it keeps no append-only curation history beyond the repository’s own. It keeps the Mech practices of one record per entity in a closed LinkML schema, agent curation behind human review, and publication references checked against cached sources. MechRegistry is modeled on it.

One record is

a knowledge resource (knowledge graph, data source, ontology, data model or aggregator) and its products

1137 records as of 2026-10-12

Counted from resource/*.md, leaving out *.*.md, *_eval*.md at 7b84d29. Refreshed weekly.

Links

License

Records: CC0 1.0

Code: BSD 3-Clause License

Domains

biomedical

Curation model

Agent-curated	Yes. AI agents generate or maintain most of the content.
Human review	`pull_request`. Agent changes arrive as pull requests that a person reviews and merges.
Agents and tools	Claude Code, dragon-ai-agent
Evidence policy	Records describe resources rather than make claims, so evidence is the resource's own homepage, repository and publications. Publication titles, years, DOIs and first authors are checked against cached reference metadata, and provenance links must name registry identifiers rather than free text or URLs.
Validation in CI	schema references
Append-only history	No
Record identifiers	Records are keyed by a lowercase slug and stored as resource//.md. Products take the resource id followed by a dot and a short name, and their pages are generated next to the resource page. Many resources also carry their Biolink Information Resource (infores) identifier.

Features

static_browser A static site lets a reader browse every record.

Grounding and sources

Ontologies and vocabularies: NCBITaxon MESH NCIT

Source	Relation	What is taken
OBO Foundry	prov:wasDerivedFrom	The OBO Foundry ontology registry, synced into ontology records.
FRINK OKN registry	prov:wasDerivedFrom	The Proto-OKN knowledge graph registry, synced into records with FRINK endpoint products.
KG-Bioportal	prov:wasDerivedFrom	KGX transforms of BioPortal ontologies, added as graph products.
Translator	prov:wasDerivedFrom	The current Translator KGX release products.
Biolink Information Resource Registry	prov:used	Infores identifiers for resources and stubs for resources not yet curated.

Products

Product	Kind	Format	Description
Resource pages `kg-registry.records`	RecordCorpus	markdown	One Markdown file with a YAML header per resource under resource//, next to the generated pages for its products.
KG-Registry LinkML schema `kg-registry.schema`	DataModelProduct	linkml linkml
KG-Registry site `kg-registry.app`	GraphicalInterface	html	A browser of every resource and product, with SQL search over the Parquet tables, a data quality dashboard and an interactive graph of resources and their sources.
Registry YAML `kg-registry.yaml`	OtherProduct	yaml	Every resource in one YAML file. JSON-LD and Turtle versions sit beside it.
Registry Parquet tables `kg-registry.parquet`	OtherProduct	other	Resources, domains, products and taxa as Parquet tables, which the read-only discovery skills query over HTTP.
KG-Registry documentation `kg-registry.docs`	DocumentationProduct	html

Contacts

Knowledge Graph Hub (Organization) Knowledge-Graph-Hub
J. Harry Caufield (Individual) ORCID 0000-0001-5705-7831 jhc@lbl.gov caufieldjh

Registry entry

Created 2026-10-10. Last modified 2026-10-10. Edit this entry on GitHub.

KG-Registry is in the registry as a Mech-like resource. It shares the Mech curation practices, but a record describes a resource held elsewhere rather than making evidence-backed claims about an entity.

At commit b7fd67f on 2026-10-10, resource/ held 1,137 resource pages. The record count takes one page per resource, resource/<id>/<id>.md. The same folders hold 4,982 product pages generated from each resource’s front matter (<id>.<product>.md) and 125 automated evaluation pages (<id>_eval_automated.md), which are not counted. Resource identifiers contain no dots, so the *.*.md exclusion takes only product pages.

Every agent-co-authored commit on the main branch since 2026-04-01 arrived through a merged pull request. The dragon-ai-agent workflow that answers mentions on issues is switched off and runs only when dispatched by hand.

KG-Registry has no cross-references here: none of the Mechs in this registry has a KG-Registry record yet.