---
title: "Vector Search"
slug: "vector-search"
tags: ["5.4"]
updated: 2026-10-28T01:53:23Z
published: 2026-10-28T01:53:23Z
canonical: "docs.vastdata.com/vector-search"
---

> ## Documentation Index
> Fetch the complete documentation index at: https://docs.vastdata.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Vector Search

VAST Databases support vector operations. This includes a vector data type (an array of floats), and functions to calculate and compare the distance between vectors and return the nearest neighbors of a specific vector.

Vector operations are supported using the [​VAST Query Engine](/docs-5-4/docs/vast-query-engine)​​.

> [!NOTE]
> **Note**
> 
> This feature is introduced in version 5.4. For this version, vector searches are performed using brute force techniques, which are not yet optimized.

## Vector Functions

VAST Databases array (vector) distance functions ​`array_distance`​​, ​`array_cosine_distance`​​, and ​`array_negative_inner_product`​​.

## Filtering Vector Searches

You can filter vector searches using standard query filters (such as WHERE clauses). You can also filter rows based on VAST user permissions (see ​[VAST Row and Column Security](/docs-5-4/docs/vast-database-row-and-column-security) filters​​), so that rows (vectors) are visible in the search according to permissions for the user running the search.

## Example

This code snippet illustrates the use of distance functions.

```plaintext
import pyarrow as pa
import vastdb
import adbc_driver_manager
import datetime as dt

# Define parameters, for example:
# VASTDB_ENDPOINT = 'http://....'
# AWS_ACCESS_KEY_ID = 'AAAAAAAAAAAAAAAAAAAA'
# AWS_SECRET_ACCESS_KEY = 'BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB'
# BUCKET_NAME = 'my-bucket' # Should already exist.
# SCHEMA_NAME =  'my-schema'
# TABLE_NAME = 'my-table'
# VAST_ADBC_DRIVER_PATH = '/tmp/libadbc_driver_vastdb.so'

# Create the table and insert data using VastDB SDK, for more information https://github.com/vast-data/vastdb_sdk?tab=readme-ov-file
session = vastdb.connect(
    endpoint=VASTDB_ENDPOINT,
    access=AWS_ACCESS_KEY_ID,
    secret=AWS_SECRET_ACCESS_KEY)

with session.transaction() as tx:
    bucket = tx.bucket(BUCKET_NAME)

    # Create the schema in the bucket.
    schema = bucket.create_schema(SCHEMA_NAME)

    # Create the table.
    dimension = 5
    columns = pa.schema([("id", pa.int64()),
                         ("vec", pa.list_(pa.field(name="item", type=pa.float32(), nullable=False), dimension)),
                         ('vec_timestamp', pa.timestamp('us'))])
    table = schema.create_table(TABLE_NAME, columns)

    # Insert a few rows of data.
    arrow_table = pa.table(schema=columns, data=[
        [1, 2, 3],
        [[1,2,3,4,5], [6, 7, 8, 9, 10], [11, 12, 13, 14, 15]],
        [dt.datetime(2024, 4, 10, 12, 34),
         dt.datetime(2024, 4, 11, 12, 34),
         dt.datetime(2024, 4, 13, 12, 34)]
    ])
    table.insert(arrow_table)

# Query the table using the ADBC driver.
with adbc_driver_manager.dbapi.connect(driver=VAST_ADBC_DRIVER_PATH, db_kwargs= {
    "vast.db.endpoint": VASTDB_ENDPOINT,
    "vast.db.access_key": AWS_ACCESS_KEY_ID,
    "vast.db.secret_key": AWS_SECRET_ACCESS_KEY}
                                       ) as connection:
    with connection.cursor() as cursor:
        full_table_name = f'"{BUCKET_NAME}/{SCHEMA_NAME}"."{TABLE_NAME}"'
        queries = [
            # Select all the rows.
            f"SELECT * FROM {full_table_name};",
            # Euclidian distance.
            f"SELECT * FROM {full_table_name} "
            f"WHERE vec_timestamp < '2024-12-11 11:30:00' ORDER BY "
            f"array_distance(vec, [1.5, 2.5, 3.5, 4.5, 5.5]::FLOAT[5]) LIMIT 2;",
            # Cosine distance.
            f"SELECT * FROM {full_table_name} "
            f"WHERE vec_timestamp < '2024-12-11 11:30:00' ORDER BY "
            f"array_cosine_distance(vec, [1.5, 2.5, 3.5, 4.5, 5.5]::FLOAT[5]) LIMIT 2;",
        ]
        for query in queries:
            cursor.execute(query)
            output_as_pandas_dataframe = cursor.fetch_arrow_table().to_pandas()
            print(f'{query=} {output_as_pandas_dataframe=}')
```