OntoDB
A database management system for ontologies. Data is RDF and the query language is SPARQL. The shell answers a SPARQL subset over B+ tree indexes with a cost-based optimizer, runs statements under strict two-phase locking with MVCC beside it, so snapshot readers never wait for writers, keeps every committed version of an ontology by name, and recovers pages with ARIES after a crash. It also does semantic search: a class hierarchy becomes integer intervals in a B+ tree, and vectors are searched by exact, IVF, or HNSW indexes, optionally limited to the terms under a class.
Build
# Ubuntu
sudo apt-get install -y clang cmake ninja-build pkg-config libraptor2-dev
# macOS
brew install pkgconf raptor cmake ninja
cmake --preset dev && cmake --build --preset dev
Presets are dev (AddressSanitizer and UndefinedBehaviorSanitizer), tsan, and release. ./build/dev/ontodb_test runs the tests. Every push runs the suite under ASan/UBSan and ThreadSanitizer, then checks formatting and clang-tidy.
Run
$ ./build/dev/ontodb mydb
OntoDB> \load data/tiny.owl
loaded 14 triples in 5.2 ms
OntoDB> SELECT ?kind WHERE { ?kind <http://www.w3.org/2000/01/rdf-schema#subClassOf> <http://example.org/materials#Material> . }
?kind
<http://example.org/materials#Metal>
<http://example.org/materials#Polymer>
(2 rows)
OntoDB> \verify
verify: ok
Pass a directory (ontodb mydb) to keep the database on disk: every update statement and every \load is a transaction, forced to a checksummed write-ahead log before the prompt returns, and reopening the directory replays the committed ones. Without a directory the shell is an in-memory database. \replace file makes the database exactly a file's triples as one commit of what differs, \tag 0.2.0 names that commit for good, and \at 0.1.0 reads an earlier one without locks; versions share the triples they have in common. One process writes a directory; ontodb --read-only mydb reads beside it. A query may span lines and runs when its braces close; \help lists the commands and \explain prints the plan.
Documentation
The book, OntoDB: Database Internals and Vector Search, by Example (PDF), and two papers — a fast and persistent query engine and hierarchy-filtered vector search — are in docs/.
License
BSD 3-Clause. Copyright (c) 2026 Lam Nguyen.
