An RDF toolchain for the Odin programming language, written from scratch in Odin, with no external dependencies at all.
Four independent libraries — a strict stack from the parser up to the query and validation engines. Each depends only downward, and each is usable on its own.
odin-rdf-parser formats, data model, and parser
|
+-- odin-rdf-record tamper-evident system of record: the log,
| the resident projection, epoch-pinned snapshots
|
+-- odin-rdf-sparql query engine
|
+-- odin-rdf-shacl shape validation
| Project | What it does | Status |
|---|---|---|
| odin-rdf-parser | Streaming parsers and emitters for N-Triples, N-Quads, Turtle, and TriG, plus the shared term/triple/quad model | All 1045 W3C conformance tests pass, RDF 1.2 / RDF-star included |
| odin-rdf-record | The family's store. An append-only, hash-chained, segmented log is the only durable form, replayed on every start into a memory-resident projection serving epoch-pinned snapshots — third-party verifiable from the format specification alone | v0.10.0. Log, resident store and write path complete (format v2); cross-checked verdict for verdict against an independent Python verifier on every test run |
| odin-rdf-sparql | SPARQL 1.1 Query with the 1.2 surface: text → algebra → solutions over the record's read API, plus SPARQL results JSON and XML writers | v0.3.0. 352 syntax tests and 546 of the corpus's 556 evaluable entries across 39 directories |
| odin-rdf-shacl | SHACL Core validation of data graphs against shapes graphs | v0.3.0. All 98 entries of the W3C core/ suite pass, with no skip list — SHACL-SPARQL is a later phase |
| odin-rdf-store | Retired. The LMDB-backed quad store both engines were originally written against, and the match() interface they queried through |
Retired 2026-09-07, final tag v0.7.0, no consumers since both engines moved to odin-rdf-record. Read-only, kept for its ADRs |
Libraries, not applications. Each layer supplies primitives and leaves policy to its consumers — there are no servers or protocol layers anywhere in the stack, by design.
Correctness is defined by executable suites: the official W3C test suites, vendored into each repository so runs are hermetic and reproducible offline — and in odin-rdf-record, where the format is its own rather than a W3C one, an independent verifier written from the format specification alone that must agree with the Odin implementation on every test run. Everything is idiomatic Odin: explicit memory management, allocator awareness, streaming over materialization, and zero-copy parsing that leaves the document in a caller-owned buffer.
Start with odin-rdf-parser — it carries the quick-start examples and the data model the rest of the family speaks.