arrow-graph-core

Arrow-native graph store with configurable namespace partitioning.

Store RDF-like triples in Apache Arrow RecordBatches with:

• Configurable namespaces — partition your graph by any string key
• Optional layer column — sub-partition within namespaces (e.g., for knowledge layers)
• Provenance tracking — causal chains, source documents, confidence scores
• Zero-copy operations — queries filter Arrow RecordBatches without materialization
• Schema versioning — read-path migration when schemas evolve

Quick Start

use arrow_graph_core::store::{ArrowGraphStore, Triple, QuerySpec};

// Create a store with your own namespace partitions
let mut store = ArrowGraphStore::new(&["world", "code", "self"]);

let triple = Triple {
    subject: "Alice".to_string(),
    predicate: "knows".to_string(),
    object: "Bob".to_string(),
    graph: None,
    confidence: Some(0.9),
    source_document: None,
    source_chunk_id: None,
    extracted_by: Some("agent-1".to_string()),
    caused_by: None,
    derived_from: None,
    consolidated_at: None,
};

let id = store.add_triple(&triple, "world", Some(1)).unwrap();
assert_eq!(store.len(), 1);

// Query by subject
let results = store.query(&QuerySpec {
    subject: Some("Alice".to_string()),
    ..Default::default()
}).unwrap();

Three API Levels

SimpleTripleStore

use arrow_graph_core::triple_store::SimpleTripleStore;

let mut store = SimpleTripleStore::new();
let id = store.add("Alice", "knows", "Bob", 0.9, "source").unwrap();
let results = store.query(Some("Alice"), None, None).unwrap();
store.remove(&id).unwrap();

KgStore (Knowledge Graph)

use arrow_graph_core::kg_store::KgStore;

let mut store = KgStore::new();
store.bind_prefix("ex", "http://example.org/");
store.add_triple("ex:Alice", "rdf:type", "ex:Person", None, 1.0).unwrap();

// Keyword search (case-insensitive)
let results = store.search_by_keywords(&["Alice"]);

Causal Chains

Track provenance with caused_by and derived_from links:

use arrow_graph_core::store::{ArrowGraphStore, Triple};

let mut store = ArrowGraphStore::new(&["default"]);

let t0 = Triple {
    subject: "observation".into(), predicate: "saw".into(), object: "rain".into(),
    caused_by: None, derived_from: None,
    graph: None, confidence: Some(1.0), source_document: None,
    source_chunk_id: None, extracted_by: None, consolidated_at: None,
};
let id0 = store.add_triple(&t0, "default", None).unwrap();

let t1 = Triple {
    subject: "inference".into(), predicate: "implies".into(), object: "wet_ground".into(),
    caused_by: Some(id0.clone()), derived_from: None,
    graph: None, confidence: Some(0.8), source_document: None,
    source_chunk_id: None, extracted_by: None, consolidated_at: None,
};
let id1 = store.add_triple(&t1, "default", None).unwrap();

// Traverse the causal chain
let chain = store.causal_chain(&id1);
assert_eq!(chain.len(), 2); // t1 → t0

Schema

The triples table has 16 columns (v1.1.0):

Access columns by name constant: col::SUBJECT, col::PREDICATE, etc.

License

MIT