Store vectors

Store dense vectors with metadata in a collection.

This page shows how to store dense vectors with metadata in a collection and configure collections with different distance metrics.

Install the Python client library: pip install actian-vectorai-client. The examples on this page use random vectors for demonstration. In production, generate vectors from an embedding model.

Store a dense vector

The following example creates a collection and stores a vector with metadata. In production, you generate vectors using pre-trained embedding models like Sentence Transformers, OpenAI embeddings, or custom neural networks. The embedding model determines the vector dimension and semantic properties.

Python

import random
from actian_vectorai import VectorAIClient, VectorParams, Distance, PointStruct

DIMENSION = 384
COLLECTION = "documents"

# Connect to VectorAI DB server
with VectorAIClient("localhost:6574") as client:
    # Create a collection for dense vectors
    client.collections.create(
        COLLECTION,  # Collection name
        vectors_config=VectorParams(size=DIMENSION, distance=Distance.Cosine)  # Vector configuration
    )
    
    # Generate a dense vector (in practice, from an embedding model)
    vector = [random.gauss(0, 1) for _ in range(DIMENSION)]
    
    # Store the vector with metadata
    point = PointStruct(
        id=1,  # Point ID
        vector=vector,  # Vector embedding
        payload={  # Metadata (optional)
            "text": "VectorAI DB enables semantic search",
            "category": "documentation"
        }
    )
    
    # Upsert point to collection
    client.points.upsert(COLLECTION, [point])
    print(f"Stored {DIMENSION}-dimensional dense vector")

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

const DIMENSION = 384;
const COLLECTION = 'documents';

async function main() {
  const client = new VectorAIClient('localhost:6574');

  // Create a collection for dense vectors
  await client.collections.create(COLLECTION, {
    dimension: DIMENSION,  // Vector dimension
    distanceMetric: 'COSINE',  // Distance metric
  });

  // Generate a dense vector (in practice, from an embedding model)
  const vector = Array.from({ length: DIMENSION }, () => Math.random() * 2 - 1);

  // Store the vector with metadata
  await client.points.upsert(COLLECTION, [
    {
      id: 1,  // Point ID
      vector: vector,  // Vector embedding
      payload: {  // Metadata (optional)
        text: 'VectorAI DB enables semantic search',
        category: 'documentation',
      },
    },
  ], { wait: true });

  console.log(`Stored ${DIMENSION}-dimensional dense vector`);
}

main().catch(console.error);

Store vectors with different distance metrics

The distance metric you choose determines how VectorAI DB measures similarity between vectors. The following example creates two collections using different metrics and inserts test vectors into each.

Python

import random
from actian_vectorai import VectorAIClient, VectorParams, Distance, PointStruct

DIMENSION = 128

# Connect to VectorAI DB server
with VectorAIClient("localhost:6574") as client:
    # Create collections with different distance metrics
    for metric in [Distance.Cosine, Distance.Euclid]:
        collection_name = f"vectors_{metric.name.lower()}"
        
        # Create collection with specific metric
        client.collections.create(
            collection_name,  # Collection name
            vectors_config=VectorParams(size=DIMENSION, distance=metric)  # Vector configuration
        )
        
        # Insert test vectors
        points = [
            PointStruct(
                id=1,  # Point ID
                vector=[random.gauss(0, 1) for _ in range(DIMENSION)]  # Generate vector
            ),
            PointStruct(
                id=2,  # Point ID
                vector=[random.gauss(0, 1) for _ in range(DIMENSION)]  # Generate vector
            )
        ]
        
        # Upsert points to collection
        client.points.upsert(collection_name, points)
        
        print(f"Collection '{collection_name}' uses {metric.name} distance")

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

const DIMENSION = 128;

async function main() {
  const client = new VectorAIClient('localhost:6574');

  // Create collections with different distance metrics
  const metrics = ['COSINE', 'EUCLIDEAN'];

  for (const metric of metrics) {
    const collectionName = `vectors_${metric.toLowerCase()}`;

    // Create collection with specific metric
    await client.collections.create(collectionName, {
      dimension: DIMENSION,  // Vector dimension
      distanceMetric: metric,  // Distance metric
    });

    // Insert test vectors
    const points = [
      {
        id: 1,  // Point ID
        vector: Array.from({ length: DIMENSION }, () => Math.random() * 2 - 1),  // Generate vector
      },
      {
        id: 2,  // Point ID
        vector: Array.from({ length: DIMENSION }, () => Math.random() * 2 - 1),  // Generate vector
      },
    ];

    // Upsert points to collection
    await client.points.upsert(collectionName, points, { wait: true });

    console.log(`Collection '${collectionName}' uses ${metric} distance`);
  }
}

main().catch(console.error);

The distance metric cannot be changed after collection creation. For details, see Vectors overview.