Configure HNSW parameters

Configure HNSW parameters when creating a collection to balance search speed, accuracy, and memory usage. Custom parameters are useful when default settings do not meet your recall or latency requirements.

Create a collection with custom parameters

The following example creates a collection with custom m and ef_construct values inside HnswConfigDiff.

from actian_vectorai import VectorAIClient, VectorParams, Distance, HnswConfigDiff

# Connect to VectorAI DB server (use context manager to ensure proper connection)
with VectorAIClient("localhost:6574") as client:
    # Create collection with custom HNSW parameters
    client.collections.create(
        "my_collection",  # Collection name
        vectors_config=VectorParams(size=384, distance=Distance.Cosine),  # Vector configuration
        hnsw_config=HnswConfigDiff(
            m=32,              # More connections for better recall
            ef_construct=200   # Higher quality index
        )
    )
    print("✓ Collection 'my_collection' created successfully")

import { VectorAIClient } from '@actian/vectorai-client';

async function main() {
    // Connect to VectorAI DB server
    const client = new VectorAIClient('localhost:6574');

    // Create collection with custom HNSW parameters
    await client.collections.create('my_collection', {
        dimension: 384,
        distanceMetric: 'COSINE',
        hnswConfig: {
            m: 32,              // More connections for better recall
            efConstruct: 200    // Higher quality index
        }
    });

    console.log("Collection 'my_collection' created successfully");
}

main().catch(console.error);

Higher parameter values improve accuracy at the cost of increased memory usage, slower index construction, and longer search times. Lower values prioritize speed and memory efficiency but can reduce recall.

Create a collection with default settings

When you omit hnsw_config, VectorAI DB applies balanced default settings that work well for general-purpose vector search. Test with your data to determine if custom parameters provide improvements.

from actian_vectorai import VectorAIClient, VectorParams, Distance

# Connect to VectorAI DB server
with VectorAIClient("localhost:6574") as client:
    # Collection with default index settings
    client.collections.create(
        "my_collection",  # Collection name
        vectors_config=VectorParams(size=384, distance=Distance.Cosine)  # Vector configuration
        # hnsw_m=16 (default)
        # hnsw_ef_construct=200 (default)
        # hnsw_ef_search=50 (default)
    )
    print("✓ Collection created successfully")

import { VectorAIClient } from '@actian/vectorai-client';

async function main() {
    // Connect to VectorAI DB server
    const client = new VectorAIClient('localhost:6574');

    // Collection with default index settings
    await client.collections.create('my_collection', {
        dimension: 384,
        distanceMetric: 'COSINE'
        // hnswConfig uses defaults:
        // m: 16
        // efConstruct: 200
    });

    console.log("Collection created successfully");
}

main().catch(console.error);

Collections

Points

Vectors

Payload

Search

Filtering

Semantic search

Hybrid search

Distance metrics

Indexing

Configure HNSW parameters

Create a collection with custom parameters

Create a collection with default settings

​Create a collection with custom parameters

​Create a collection with default settings

Create a collection with custom parameters

Create a collection with default settings