Open Ontologies Lite (Python bridge)

A lightweight, pip-installable Python bridge to the same Oxigraph RDF/OWL engine that powers Open Ontologies. No Rust toolchain, no compilation, no multi-gigabyte build directory — pyoxigraph ships the engine as a prebuilt wheel, so everything here is pure-Python glue installed from PyPI.

It exposes the core ontology lifecycle as both a Python library and an MCP server.

Why this exists

The full Rust engine compiles a large dependency tree from source (5+ GB of build artifacts, heavy SSD churn). This bridge is the opposite trade: install in seconds, run anywhere Python runs, keep the Oxigraph SPARQL engine underneath. It covers the core surface (validate, load, query, diff, lint, convert, stats, save), not the full 100-tool engine.

Install

pip install open-ontologies-lite        # one universal wheel, no compiler

Use as a Python library

from open_ontologies_lite import OntologyEngine

engine = OntologyEngine()
engine.load(open("ontology.ttl").read())          # load Turtle
print(engine.stats())                              # {'triples':..,'classes':..,..}

rows = engine.query(
    "SELECT ?c WHERE { ?c a <http://www.w3.org/2002/07/owl#Class> }"
)
print([r["c"] for r in rows["rows"]])

print(engine.lint())                               # missing labels/domains/ranges
print(OntologyEngine.convert(ttl, "turtle", "ntriples"))

See examples/python_usage.py for a runnable end-to-end script.

Version governance with KGCL

from open_ontologies_lite import kgcl_diff

cs = kgcl_diff(open("v1.ttl").read(), open("v2.ttl").read())
print(cs.counts())     # {'node_creation': 1, 'node_rename': 1, ...}
print(cs.to_kgcl())    # KGCL change records, one per line

kgcl_diff classifies the change between two ontology versions into KGCL records (node created/deleted, renamed, annotation changed, edge created/deleted). Pure structural comparison, no model. Also exposed as the onto_kgcl_diff MCP tool.

Alignment candidate generation with HNSW (optional [align] extra)

pip install "open-ontologies-lite[align]"

from open_ontologies_lite import AlignmentIndex

idx = AlignmentIndex(dim=384)
idx.add("flw:PC-BAK", vec_bakery)       # vectors come from YOUR embedder
idx.add("FOODON:00001626", vec_foodon)
idx.build()
idx.query(vec_query, k=5)               # -> [Candidate(id, score), ...]

MCP-native by design: the package owns the HNSW index, you supply the vectors. Lite never calls an embedding model; bring vectors from your orchestrator and let it adjudicate the candidates.

Use as an MCP server

open-ontologies-lite          # stdio MCP server
# or: python -m open_ontologies_lite

Register it with any MCP client (e.g. Claude):

{
  "mcpServers": {
    "open-ontologies-lite": { "command": "open-ontologies-lite" }
  }
}

Tools

Relationship to the Rust engine

This is the Python layer of the project. For the full engine (three-layer Dynamics/Causal/Planner architecture, HNSW semantic search, OWL2-DL tableaux reasoning, PDDL planning, governance, 100 tools), use the Rust build. Same Oxigraph core; pick the weight class you need.

License

MIT