Vector Search

Configure dense vector fields for semantic and similarity search using SearchProjectionConfig, with support for cosine, dot product, and L2 norm similarity metrics.

February 15, 2026

Vector search enables semantic and similarity search by storing embedding vectors alongside traditional text fields. NeoArc Studio configures vector search at the property level via SearchProjectionConfig, allowing each field to define its dimensionality, algorithm profile, and similarity metric.

Configuration Fields

To enable vector search on a property, set searchFieldType to dense_vector and provide the required vector parameters.

Field	Type	Required	Description
searchFieldType	'dense_vector'	Yes	Must be set to dense_vector for vector fields
dimensions	number	Yes	Number of dimensions in the embedding vector (e.g. 768, 1536)
vectorSearchProfile	string	Recommended	Engine-specific vector search algorithm profile name (free-text input)
similarityMetric	'cosine' \| 'dot_product' \| 'l2_norm'	No	Distance or similarity function used for nearest-neighbour queries

Similarity Metrics

The similarityMetric field determines how the search engine calculates distance between vectors.

Cosine

Measures the angle between two vectors. Suitable for normalised embeddings where magnitude is irrelevant.

Dot Product

Computes the sum of element-wise products. Fastest metric, but requires normalised vectors for meaningful results.

L2 Norm (Euclidean)

Measures the straight-line distance between two points in vector space. Sensitive to vector magnitude.

Engine Support

Engine	Vector Search Supported
Elasticsearch	Yes
Azure Cognitive Search	Yes
OpenSearch	Yes
Typesense	Yes
Algolia	No
Custom	Yes

Configuration Steps

Example Configuration

The following JSON shows a vector search configuration on a property's search projection.

{
  "included": true,
  "searchFieldType": "dense_vector",
  "dimensions": 1536,
  "vectorSearchProfile": "my-hnsw-profile",
  "similarityMetric": "cosine",
  "stored": true,
  "indexEnabled": true
}

Validation Rules

NeoArc Studio enforces two validation rules specific to vector search fields.

Rule	Severity	Condition
search-vector-missing-dimensions	Error	searchFieldType is dense_vector but dimensions is not set
search-vector-missing-profile	Warning	Vector field does not have a vectorSearchProfile specified

Common Embedding Dimensions

Model	Dimensions
all-MiniLM-L6-v2	384
BERT base	768
BERT large	1024
OpenAI text-embedding-ada-002	1536
OpenAI text-embedding-3-small	1536
OpenAI text-embedding-3-large	3072
Cohere embed-english-v3	1024