Range Search

Overview

Range Search retrieves all vectors within a specified distance/similarity threshold from a query, instead of returning a fixed k-nearest neighbors. Returns variable number of results based on similarity criteria.

Comparison

k-NN Search

"Find 10 nearest neighbors" → Always returns 10 results

Range Search

"Find all vectors with similarity > 0.8" → Returns 0 to N results

Use Cases

• Quality thresholding ("only show high-confidence matches")
• Duplicate detection ("find all near-duplicates")
• Anomaly detection ("nothing similar exists")
• Variable-size result sets
• Similarity-based filtering

Example

# Qdrant
results = client.search(
    collection_name="my_collection",
    query_vector=query_embedding,
    score_threshold=0.8,  # Only results with score > 0.8
)

# May return 0, 5, 100, or any number of results

Benefits

• Quality control (no poor matches)
• Flexible result sizes
• Natural similarity boundary
• Better for strict requirements

Challenges

• Unpredictable result count
• Need to choose appropriate threshold
• May return zero results
• Less efficient than k-NN for some indexes

Pricing

Not applicable (query type supported by most vector databases).