Documentation

Introduction

Oxid-DB is a single-node, embedded database that fuses an ontology reasoner and an approximate-nearest-neighbour index into one engine. Instead of stitching a graph store to a vector store with glue code, every node lives in a single identity space, so the ontology can filter candidates during the search rather than after it.

These docs cover the core concepts behind the engine, the OxQL query language, and how the system behaves in production. Oxid-DB is an early research preview; each page tells you what is shipped today and what is still on the roadmap.

Architecture

Traditional stacks keep symbols and vectors in separate systems and pay a brittle sync tax to keep them aligned. Oxid-DB collapses that into one store: a node's logical type, its properties and its embedding are all attributes of the same identity.

Ontologies & reasoning

Load a standard OWL ontology and Oxid-DB compiles it into an EL++ classification. As you insert data, the reasoner assigns each node its inferred classes so that subsumption (IS-A) queries resolve against materialised types instantly.

Want the algorithm itself: EL normal form, the five completion rules, and exactly where the engine is complete? Read the reasoner deep-dive.

Vectors & embeddings

Insert text and Oxid-DB embeds it locally with a bundled ONNX model: vectors never leave your machine. Embeddings are indexed for approximate nearest-neighbour search and attached to the node's identity, so they are filterable by the very same ontology.

Bring your own vectors, or let the engine create them from plain text via FastEmbed, OpenAI, or Ollama, see how the database makes embeddings.

OxQL basics

OxQL is Oxid-DB's query language. A query binds a variable to a set of nodes with FIND, constrains it with WHERE, and bounds the result with LIMIT.

-- every OxQL query starts from a set of nodesFIND ?xWHERE ?x IS-A BeerLIMIT 10

Hybrid queries

The point of Oxid-DB is the NEAR clause: rank nodes by semantic similarity while symbolic constraints prune the candidate set in the same pass. Logic and meaning, one query, one traversal.

-- filter by the ontology, rank by the vector index,-- in one single pass over one identity spaceFIND ?xWHERE ?x IS-A PaleLager  AND ?x.abv < 5.5NEAR ?x TO "crisp, dry, noble hops"LIMIT 5

Here the ontology restricts the search to sub-5.5% pale lagers before a single vector is scored, so the ANN index only ever ranks nodes that already satisfy the logic.

Durability & ACID

Writes are ACID. A write-ahead log makes every committed transaction crash-safe, and MVCC snapshots give readers a consistent view without blocking writers.

Limitations

Oxid-DB is a research preview. It is a genuinely novel engine that is also honestly early: we would rather tell you exactly where the edges are than overclaim.