GRAPH AND VECTOR

Graph and vector, one query

Model relationships as native graph edges and search meaning with HNSW vector indexes - in the same database, over the same records. Traverse connections and rank by similarity in a single SurrealQL statement, with structured filters narrowing the set first.

Read the docs Start free with SurrealDB

01 |CONNECTED DATA

Relationships and similarity, together

Edges are full documents you traverse with arrow syntax; vectors are an index on the same table. Because both live in one engine, one query can walk the graph and rank by meaning.

GRAPH RELATIONSHIPS

Edges are records you traverse

RELATE connects two records through an edge that carries its own fields. Arrow syntax walks those connections without a join.

RELATE customer:alice->purchased->product:widget_pro
    SET quantity = 2, date = time::now(), source = 'web';

SELECT ->purchased[WHERE date > time::now() - 30d]->product.name
    FROM customer:alice;

VECTOR SEARCH

Similarity on the same records

Define an HNSW index on a field of embeddings, then rank by distance - with a structured WHERE clause scoping the candidates before the vector search runs.

DEFINE INDEX product_embedding ON product
    FIELDS embedding HNSW DIMENSION 1536 DIST COSINE;

SELECT id, name, vector::distance::knn() AS distance
    FROM product
    WHERE category = 'tools'
        AND embedding <|10,40|> $query_embedding
    ORDER BY distance;

A graph database and a vector database in one

Native relationships, approximate-nearest-neighbour search, and full-text ranking compose in a single query language over a single copy of your data.

Native graph model

Nodes, edges, and properties as first-class records, traversed with arrow syntax to any depth.

Learn more

Recursive traversal

Walk chains and dependency trees with recursive graph queries, filtering edges as you go.

Learn more

HNSW vector index

Approximate-nearest-neighbour search with cosine, euclidean, and manhattan distance over your embeddings.

Learn more

Hybrid search

Combine full-text BM25 and vector similarity in one query with reciprocal rank fusion.

Learn more

Scope first, rank second

A structured filter narrows the candidate set before the vector search, so similarity runs on the rows that matter.

Learn more

One copy of the data

Graph edges, vectors, and documents share the same records - no syncing a vector store to a graph store.

Learn more

02 |ONE ENGINE

No vector store bolted to a graph store

The usual stack stitches a graph database, a vector database, and a document store together, then syncs them. SurrealDB makes all three the same records, queried the same way.