Skip to main content

Named Vectors

vectlite supports multiple vector spaces per record. For example, you can store separate embeddings for a document's title and body, then search across both.

Inserting Named Vectors​

Python​

db.upsert(
    "doc1",
    primary_embedding,   # default vector
    {"title": "Auth Setup", "text": "How to configure SSO..."},
    vectors={"title": title_embedding, "body": body_embedding},
)

Node.js​

db.upsert('doc1', primaryEmbedding,
  { title: 'Auth Setup', text: 'How to configure SSO...' },
  { vectors: { title: titleEmbedding, body: bodyEmbedding } }
)

Searching a Specific Vector Space​

Python​

results = db.search(title_query_embedding, k=10, vector_name="title")

Node.js​

const results = db.search(titleQueryEmbedding, { k: 10, vectorName: 'title' })

Search across multiple vector spaces in a single call with custom weights:

Python​

results = db.search(
    k=10,
    query_vectors={
        "title": title_query_embedding,
        "body": body_query_embedding,
    },
    vector_weights={"title": 1.5, "body": 1.0},
)

Node.js​

const results = db.search(null, {
  k: 10,
  queryVectors: {
    title: titleQueryEmbedding,
    body: bodyQueryEmbedding,
  },
  vectorWeights: { title: 1.5, body: 1.0 },
})