Vector Search¶

Uni provides native vector search over embedding properties with ANN indexes (HNSW, IVF_PQ, Flat). Use it for semantic search, RAG, and similarity-based retrieval.

What It Provides¶

Vector properties stored alongside graph data.
ANN indexes with cosine, L2, or dot distance.
CALL uni.vector.query(...) for KNN retrieval.

Example¶

RustPython

use uni_db::{DataType, IndexType, Uni, VectorAlgo, VectorIndexCfg, VectorMetric};

# async fn demo() -> Result<(), uni_db::UniError> {
let db = Uni::open("./my_db").build().await?;

db.schema()
    .label("Document")
        .property("title", DataType::String)
        .property("embedding", DataType::Vector { dimensions: 384 })
        .index("embedding", IndexType::Vector(VectorIndexCfg {
            algorithm: VectorAlgo::Hnsw { m: 16, ef_construction: 200 },
            metric: VectorMetric::Cosine,
        }))
    .apply()
    .await?;

let rows = db.query_with(
    "CALL uni.vector.query('Document', 'embedding', $q, 10) YIELD node, score RETURN node, score"
)
.param("q", vec![0.1_f32, 0.2, 0.3])
.fetch_all()
.await?;

println!("{:?}", rows);
# Ok(())
# }

import uni_db

db = uni_db.Database("./my_db")

db.schema() \
    .label("Document") \
        .property("title", "string") \
        .vector("embedding", 384) \
        .index("embedding", "vector") \
        .done() \
    .apply()

rows = db.query(
    "CALL uni.vector.query('Document', 'embedding', $q, 10) YIELD node, score RETURN node, score",
    {"q": [0.1, 0.2, 0.3]}
)
print(rows)

Use Cases¶

Semantic search for documents or products.
RAG retrieval over knowledge graphs.
Similarity search over embeddings generated in-app.

When To Use¶

Choose vector search when you need semantic similarity rather than exact matching. Pair it with graph traversal for contextual results.