Concepts glossary

A quick-reference glossary. For the fuller treatment, see Core concepts.

Data model

Entity — a named node in the graph (e.g. auth-service, user-db).

Relation — a typed edge predicate (e.g. WRITES_TO, STORES).

Assertion — a relationship event connecting two entities, carrying confidence and evidence.

Observation — evidence text or source metadata supporting an entity or assertion. Provenance stays attached and searchable.

Triplet — the write unit: a source entity, a relation, and a target entity (plus optional confidence/evidence).

Attribute — a typed scalar property on an entity, read back flat under attributes (there is no nested metadata.attributes blob).

Tenant — an isolation boundary; on the SaaS a stack maps to a tenant.

Stack — a hosted, isolated tenant with its own graphs, data, and API keys.

Graph — a named graph within a tenant (default main); its ontology is fixed at creation.

Branch — a copy-on-write snapshot of a graph (default main).

GraphKey — the (tenant, graph, branch) triple that scopes every operation.

WAL — the append-only write-ahead log of immutable graph events.

Snapshot — a consistent view identified by a commit sequence; all query surfaces agree on the snapshot they serve.

commit_seq / indexed_seq — the graph head commit vs. the sequence covered by the loaded snapshot / index run.

Hybrid search — fused lexical + BM25 + vector + ontology + graph-neighborhood retrieval over one snapshot.

BM25 — full-text lexical scoring.

ANN / vector — approximate nearest-neighbor search over embeddings; persisted as TurboQuant-quantized blocks.

RRF — reciprocal-rank fusion, used by multi-search to combine subquery rankings.

Filter — a structured JSON predicate applied across search paths.

Facet — a post-filter count of a metadata bucket.

Index run — an immutable, object-backed acceleration structure derived from the graph; disposable and rebuildable.

Consistency — strong (overlay the WAL tail on the index run) or eventual (index run only).

source — persisted (durable index runs) or ephemeral (built on the fly).

Ontology — the entity/relation/concept/term vocabulary; supports import and additive evolution.

Structured query — a typed SPARQL-subset SELECT/aggregate built as JSON.

SPARQL — the conformant SPARQL 1.1 text engine, also served at /sparql.

SHACL — shape validation, selection, property paths, and (with SHACL-AF) rule inference.

Reasoning / inference — deriving facts you didn’t write, via type closure and rule inference; folded into queries with entailment/reason.

Type closure — RDFS subClassOf entailment: a query for a type also matches its subtypes. On by default; disable with entailment: "none".

Rule inference (SHACL-AF) — rules that entail new edges to a bounded fixpoint; folded into a query with reason: true.

Provenance / why — the evidence chain that justifies an assertion.

ObjectBackend — the storage contract; LocalFsObjectBackend or S3ObjectBackend.

CAS — content-addressed conditional writes (If-None-Match / If-Match) that back durable, idempotent updates.

ETag / checksum — content-hash object identity and integrity validation.