v3.7 · gunicorn · Prometheus · Redis SSE · cosign · SBOM
The ontology mesh
for GraphRAG.
Generate production-ready OWL 2 ontologies + SHACL shapes
from your relational schema or your logs — with materialised
inference, full prov:wasDerivedFrom lineage, and
a wizard that puts every candidate in front of you for review.
Currently source-install only.
pip install ontomesh from PyPI lands in v3.6.
Built on open standards
OWL 2
SHACL
SPARQL
JSON-LD
SKOS
Datalog (Rulewerk)
SQLite / Postgres / MySQL
What you can build
Three things you'd otherwise hand-craft for months.
Generate a full ontology from a database schema.
Point Ontomesh at a relational database. The headless
pipeline introspects the schema, lifts it through OWL 2 /
SHACL / JSON-LD / SKOS, runs the reasoner, and produces a
materialised artifact bundle with full
prov:wasDerivedFrom lineage.
{% if bench %}
{{ bench.db_size_human }} DB ·
{{ bench.elapsed_human }} ·
{{ bench.owl_classes }} classes,
{{ bench.owl_properties_human }} properties
{% else %}
OWL 2 · SHACL ·
JSON-LD · SKOS
{% endif %}
Validate with SHACL. Reason with Datalog.
Shapes catch data-quality issues before they hit
production. Datalog rules derive new facts with full
prov:wasDerivedFrom lineage. Every reasoner run
emits a materialisation report you can diff against the
last release.
{% if bench %}{{ bench.shacl_shapes }} SHACL shapes
{% else %}SHACL 1.1{% endif %} ·
SWRL ·
Datalog (Rulewerk)
Mine candidates from raw logs.
Log Discovery (phases L4–L13) mines templates with
online clustering, surfaces candidate entities / events /
relationships from co-occurrence (PMI), fits per-service
HMMs for anomalies, and gates causal edges with Granger /
transfer-entropy. You review one candidate at a time in
the wizard.
L4 templates · L8 regimes ·
L11 causal DAG · L13 rate anomalies
How it works
From raw signal to graph-grounded RAG, in three steps.
-
1
Discover
Point Ontomesh at logs or a database schema. The
discovery pipeline (phases L1–L13) surfaces entity /
event / relationship candidates with confidence,
provenance, and a one-click accept.
- →
-
2
Model & validate
Add SHACL shapes and SWRL / Datalog rules. The
reasoner materialises derived triples with lineage; the
competency-question runner gives you regression coverage
as you iterate.
- →
-
3
Generate & ship
One click emits an artifact bundle — OWL Turtle, SHACL,
JSON-LD, pydantic / TypeScript bindings — plus a hybrid
retriever ready to drop into your RAG stack.
Get started
One command, a full ontology bundle.
# 1. Install (PyPI publish lands in v3.6 — for now, source install)
git clone https://github.com/synaptixs/ontomesh.git
cd ontology && pip install -e .
# 2. Generate the full OWL / SHACL / JSON-LD bundle from a SQLite DB
ontomesh --db db/demo.db --out ./output
# 3. Or launch the interactive wizard at http://localhost:5051
ontomesh-wizard --port 5051
# 4. Mine candidates from raw logs (phases L4–L13)
ontomesh --phase log --db db/demo.db --out ./output
Prefer to point and click?
Open the wizard →
Stop hand-crafting ontologies for your RAG stack.
Open the wizard, load a sample domain (telecom, healthcare,
e-commerce), and watch a real ontology + retriever fall out the
other end in under a minute.
Open the wizard →