Search

Full-text, fuzzy, vector, and hybrid search are first-class query expressions in Omnigraph.

Omnigraph has five built-in search expressions plus Reciprocal Rank Fusion (rrf) for hybrid ranking. They are not separate services. They are expressions you use directly inside .gq queries, alongside traversal, filtering, and projection.

The search modes

Expression	Mode	Where it appears	What it does
`search(field, term)`	Full-text	`match`, `order`, `return`	Inverted-index keyword matching. In `match` it filters; in `order`/`return` it projects a score.
`fuzzy(field, term, distance)`	Fuzzy	`match`, `order`, `return`	Edit-distance-tolerant text matching for typos and near-matches.
`match_text(field, phrase)`	Text match	`match`, `order`, `return`	Same Lance full-text backend as `search`; tokenized matching, not strict phrase adjacency.
`bm25(field, term)`	BM25 ranking	`order`, `return`	BM25 relevance score over an `@index`-ed `String` field.
`nearest(field, query)`	Vector KNN	`order`, `return`	Nearest-neighbor distance over an `@index`-ed `Vector(N)` field. Accepts a `Vector(N)` parameter or a string that the engine auto-embeds. Requires `limit`.
`rrf(rank_a, rank_b, k?)`	Hybrid	`order`	Reciprocal Rank Fusion of two ranking signals. Optional third argument is the rank-fusion constant `k` (default `60`). Requires `limit`.

Where search expressions appear

Search expressions are valid in three places inside a query:

match block, search(), fuzzy(), and match_text() act as filters. They restrict which nodes are returned.
order block, bm25(), nearest(), and rrf() rank results by relevance. Any of the text expressions can also drive ordering.
return block. Any search expression can be projected as a per-row score column, so callers can see the underlying signal that produced the ranking.

A query can combine all three: filter in match, rank in order, expose the score in return.

query find_experts($term: String, $vec: Vector(1536)) {
    match {
        $p: Person
        search($p.bio, $term)
    }
    return { $p.name, $p.bio }
    order { rrf(nearest($p.embedding, $vec), bm25($p.bio, $term)) desc }
    limit 10
}

Index requirements

Search expressions require indexes on the fields they operate on:

Search type	Schema annotation	Index created
Full-text, fuzzy, text match, BM25	`@index` on a `String` field	Inverted index
Vector KNN	`@index` on a `Vector(N)` field	`ivf_flat` vector index with `L2` metric

Without the right index, the query compiler rejects the query at compile time.

node Person {
    name: String @key
    bio: String @index
    embedding: Vector(1536) @index
}

Combining search with traversal

Search expressions compose with the rest of the query language. You can filter by search, traverse edges, and project properties all in one query:

query experts_at_company($term: String) {
    match {
        $p: Person
        search($p.bio, $term)
        $p worksAt $c: Company { name: "Acme" }
    }
    return { $p.name, $c.name }
    order { bm25($p.bio, $term) desc }
}

Next steps

Full-text Search. Keyword search, fuzzy matching, text matching, and BM25 ranking.
Vector Search. Nearest-neighbor search over embeddings.
Hybrid Search. Combine BM25 and vector ranking with Reciprocal Rank Fusion.