Protocols & Standards

Superset philosophy SutraDB is a superset of W3C standards — both in data model (RDF-star ⊃ RDF 1.2) and query language (SPARQL + VECTOR_SIMILAR ⊃ SPARQL 1.1). Any valid standard query or dataset works unchanged. Extensions add what the standards can't express.

Standards We Follow

Standard	Status	Notes
RDF 1.1 / RDF-star	Superset	Full RDF-star support (triples in any position). All RDF 1.2 data is valid. Details →
SPARQL 1.1	Superset	Standard queries work as-is. Extensions: VECTOR_SIMILAR, VECTOR_SCORE. Details →
SPARQL Protocol	Implemented	HTTP GET/POST to `/sparql` endpoint. JSON results format.
N-Triples	Implemented	Full N-Triples parsing for bulk data import via POST `/triples`.
OWL 2	Planned	Opt-in query-time reasoning layer. Will not modify stored data. Details →
Turtle / N-Quads	Planned	Parser planned for data ingestion. Considering Oxigraph's oxttl crate.
SPARQL Graph Store Protocol	Planned	PUT/POST/DELETE for named graph management.

ACID Compliance

SutraDB targets full ACID compliance:

Property	Status	Implementation
Atomicity	Yes	Triple insertions are atomic — a triple is either fully inserted into all three indexes (SPO/POS/OSP) or not at all. Vector insertions atomically create both the triple and the HNSW entry.
Consistency	Yes	Dimension mismatches are hard errors. Duplicate triples are rejected. All three indexes are always consistent with each other.
Isolation	Yes	Mutex-based serialization in server mode. Single-writer guarantee ensures no dirty reads. Sled's persistent store provides snapshot isolation.
Durability	Partial	In-memory store is volatile (by design — serverless mode). Sled-backed persistent store (`.sdb` files) provides full durability with WAL. Planned: crash recovery for HNSW indexes.

Deployment Modes

Serverless (Default)

Like SQLite. Open a .sdb file directly from your application. No daemon, no port, no configuration. Single-process access.

sutra query "SELECT * WHERE { ?s ?p ?o }" --data my.sdb

Server Mode

Like PostgreSQL. Start a SPARQL HTTP endpoint for multi-client access, remote queries, and REST API operations.

sutra serve --port 3030 --data my.sdb

What We Deliberately Skip

Technology	Status	Rationale
SQL	Never	Wrong model for graph data. Use PostgreSQL.
MongoDB Query Language	Never	Wrong model for graph data. Use MongoDB.
GraphQL	Never	Application-layer concern. Build a GraphQL wrapper over SPARQL if needed.
Cypher	Planned	Translation layer over SPARQL, not a native engine. For Neo4j migration convenience.
RDFS Inference	Never	Too implicit. OWL reasoning (opt-in, explicit) covers the useful subset.

File Format

SutraDB uses the .sdb file extension. This is deliberately not a standard RDF serialization format — it's a binary storage format containing:

SPO/POS/OSP triple indexes (sorted u64 keys)
Term dictionary (IRI interning)
HNSW vector indexes (one per declared vector predicate)
Metadata (predicate declarations, index parameters)

The .sdb extension signals that this file contains a superset of RDF — standard triples plus native vector indexes that no RDF serialization format can represent.