Skip to main content

Hybrid Search

vectlite supports dense vector search, sparse keyword search (BM25), and hybrid fusion of both.

Cosine similarity with automatic HNSW indexing. Small collections (<128 records) use exact search.

Python​

results = db.search(query_embedding, k=10)

Node.js​

const results = db.search(queryEmbedding, { k: 10 })

BM25-scored inverted index for keyword retrieval. Use sparse_terms() to generate term vectors from text.

Python​

terms = vectlite.sparse_terms("How to configure SSO authentication")
results = db.search(sparse=terms, k=10)

Node.js​

const terms = vectlite.sparseTerms('How to configure SSO authentication')
const results = db.search(null, { sparse: terms, k: 10 })

Hybrid Fusion​

Combine dense and sparse search with linear combination or reciprocal rank fusion (RRF).

Linear Combination (default)​

results = db.search(
    query_embedding,
    k=10,
    sparse=vectlite.sparse_terms("SSO authentication"),
    dense_weight=1.0,
    sparse_weight=0.5,
    fusion="linear",
)

Reciprocal Rank Fusion (RRF)​

results = db.search(
    query_embedding,
    k=10,
    sparse=vectlite.sparse_terms("SSO authentication"),
    fusion="rrf",
    rrf_k=60,
)
const results = db.search(queryEmbedding, {
  k: 10,
  sparse: vectlite.sparseTerms('SSO authentication'),
  fusion: 'rrf',
  rrfK: 60,
})

Inserting Sparse Vectors​

Provide sparse terms when inserting to enable keyword search:

Python​

db.upsert(
    "doc1",
    dense_embedding,
    {"title": "Auth Setup", "text": "How to configure SSO..."},
    sparse=vectlite.sparse_terms("How to configure SSO authentication"),
)

Node.js​

db.upsert('doc1', denseEmbedding,
  { title: 'Auth Setup', text: 'How to configure SSO...' },
  { sparse: vectlite.sparseTerms('How to configure SSO authentication') }
)

Text Helpers​

For convenience, upsert_text() and search_text() handle embedding generation and sparse term extraction in one call:

Python​

def embed(text):
    # Your embedding function (e.g., OpenAI, sentence-transformers)
    return model.encode(text)

vectlite.upsert_text(db, "doc1", "Auth setup guide for SSO", embed, {"source": "docs"})
results = vectlite.search_text(db, "how to authenticate", embed, k=5)

Node.js​

function embed(text) {
  // Your embedding function
  return model.encode(text)
}

vectlite.upsertText(db, 'doc1', 'Auth setup guide for SSO', embed, { source: 'docs' })
const results = vectlite.searchText(db, 'how to authenticate', embed, { k: 5 })