# `AshNeo4j.Functions.VectorSimilarity`
[🔗](https://github.com/diffo-dev/ash_neo4j/blob/v0.10.1/lib/functions/vector_similarity.ex#L5)

Normalised cosine similarity between a stored vector attribute and a query
embedding, matching Neo4j's `vector.similarity.cosine/2`.

Returns a float in `[0.0, 1.0]` — `1.0` identical direction, `0.5` orthogonal,
`0.0` opposite — i.e. `(1 + raw_cosine) / 2`. Higher is closer. Typically used
in `sort` to rank results by relevance:

    Item
    |> Ash.Query.sort({calc(vector_similarity(embedding, ^q), type: :float), :desc})
    |> Ash.read!()

Requires Cypher 25 (Neo4j ≥ 2025.06). The data layer pushes this down to
`vector.similarity.cosine(s.embedding, $q)`. `evaluate/1` mirrors that exact
normalisation in Elixir so the data layer's in-memory correctness re-filter
agrees with the pushdown (and so filters/sorts also work without pushdown).

# `args`

# `has_partial_evaluate?`

---

*Consult [api-reference.md](api-reference.md) for complete listing*