Move RDF datasets through streaming readers and writers without buffering the whole transfer.
Load and export graph data without first fitting the entire transfer into memory. Streaming lets a constrained host handle datasets larger than its working memory while the consumer controls the pace.
Use standard RDF formats to move data between systems. Keep named graph boundaries when the exchange needs them.
Query languages covers querying the resulting graph. Files as graph values covers document and media content.
Developer reference
Async readers and writers process Turtle, TriG, N-Triples, and N-Quads incrementally. Parser state survives transport chunk boundaries, including split IRIs, prefixes, and triple terms. A network chunk is not a syntactic record boundary.
Turtle and N-Triples describe triples. TriG and N-Quads preserve dataset graph names. Choose the syntax for the dataset semantics, not just its filename.
A streaming consumer requests more data as it has capacity. Bounded buffers and incremental serialization avoid retaining the complete input or result merely to transfer it. Slow downstream consumers must not cause unbounded upstream accumulation.
Parsing a stream does not publish each received fragment as an independent committed fact. Workspace transaction and commit-time validation rules still determine publication. A truncated or invalid stream is an error, not a successful complete import.
Streaming support is format-specific. These guarantees describe the four named syntaxes; they are not a claim that every supported RDF format has the same buffering behavior.