RedisVentures / redis-vector-php

Redis Vector Library (RedisVL) enables Redis as a real-time database for LLM applications, based on Predis PHP client

Introduction

The Redis Vector Library (RedisVL) is a PHP client for AI applications leveraging Redis.

Designed for:

• Vector similarity search
• Recommendation engine

A perfect tool for Redis-based applications, incorporating capabilities like vector-based semantic search, full-text search, and geo-spatial search.

Getting started

Installation

composer install redis-ventures/redisvl

Setting up Redis

Choose from multiple Redis deployment options:

1. Redis Cloud: Managed cloud database (free tier available)
2. Redis Stack: Docker image for development

docker run -d --name redis-stack -p 6379:6379 -p 8001:8001 redis/redis-stack:latest

3. Redis Enterprise: Commercial, self-hosted database

What's included?

Redis index management

1. Design your schema that models your dataset with one of the available Redis data structures (HASH, JSON) and indexable fields (e.g. text, tags, numerics, geo, and vectors).

Load schema as a dictionary:

$schema = [
    'index' => [
        'name' => 'products',
        'prefix' => 'product:',
        'storage_type' => 'hash',
    ],
    'fields' => [
        'id' => [
            'type' => 'numeric',
        ],
        'categories' => [
            'type' => 'tag',
        ],
        'description' => [
            'type' => 'text',
        ],
        'description_embedding' => [
             'type' => 'vector',
             'dims' => 3,
             'datatype' => 'float32',
             'algorithm' => 'flat',
             'distance_metric' => 'cosine'
        ],
    ],
];

2. Create a SearchIndex object with an input schema and client connection to be able to interact with your Redis index

use Predis\Client;
use RedisVentures\RedisVl\Index\SearchIndex;

$client = new Client();
$index = new SearchIndex($client, $schema);

// Creates index in the Redis
$index->create();

3. Load/fetch your data from index. If you have a hash index data should be loaded as key-value pairs , for json type data loads as json string.

$data = ['id' => '1', 'count' => 10, 'id_embeddings' => VectorHelper::toBytes([0.000001, 0.000002, 0.000003])];

// Loads given dataset associated with given key.
$index->load('key', $data);

// Fetch dataset corresponding to given key
$index->fetch('key');

Realtime search

Define queries and perform advanced search over your indices, including combination of vectors and variety of filters.

VectorQuery - flexible vector-similarity semantic search with customizable filters

use RedisVentures\RedisVl\Query\VectorQuery;

$query = new VectorQuery(
    [0.001, 0.002, 0.03],
    'description_embedding',
    null,
    3
);

// Run vector search against vector field specified in schema.
$results = $index->query($query);

Incorporate complex metadata filters on your queries:

use RedisVentures\RedisVl\Query\Filter\TagFilter;
use RedisVentures\RedisVl\Enum\Condition;

$filter = new TagFilter(
    'categories',
    Condition::equal,
    'foo'
);

$query = new VectorQuery(
    [0.001, 0.002, 0.03],
    'description_embedding',
    null,
    10,
    true,
    2,
    $filter
);

// Results will be filtered by tag field values.
$results = $index->query($query);

Filter types

Numeric

Numeric filters could be applied to numeric fields. Supports variety of conditions applicable for scalar types (==, !=, <, >, <=, >=). More information here.

use RedisVentures\RedisVl\Query\Filter\NumericFilter;
use RedisVentures\RedisVl\Enum\Condition;

$equal = new NumericFilter('numeric', Condition::equal, 10);
$notEqual = new NumericFilter('numeric', Condition::notEqual, 10);
$greaterThan = new NumericFilter('numeric', Condition::greaterThan, 10);
$greaterThanOrEqual = new NumericFilter('numeric', Condition::greaterThanOrEqual, 10);
$lowerThan = new NumericFilter('numeric', Condition::lowerThan, 10);
$lowerThanOrEqual = new NumericFilter('numeric', Condition::lowerThanOrEqual, 10);

Tag

Tag filters could be applied to tag fields. Single or multiple values can be provided, single values supports only equality conditions (==, !==), for multiple tags additional conjunction (AND, OR) could be specified. More information here

use RedisVentures\RedisVl\Query\Filter\TagFilter;
use RedisVentures\RedisVl\Enum\Condition;
use RedisVentures\RedisVl\Enum\Logical;

$singleTag = new TagFilter('tag', Condition::equal, 'value')
$multipleTags = new TagFilter('tag', Condition::notEqual, [
    'conjunction' => Logical::or,
    'tags' => ['value1', 'value2']
])

Text

Text filters could be applied to text fields. Values can be provided as a single word or multiple words with specified condition. Empty value corresponds to all values (*). More information here

use RedisVentures\RedisVl\Query\Filter\TextFilter;
use RedisVentures\RedisVl\Enum\Condition;

$single = new TextFilter('text', Condition::equal, 'foo');

// Matching foo AND bar
$multipleAnd = new TextFilter('text', Condition::equal, 'foo bar');

// Matching foo OR bar
$multipleOr = new TextFilter('text', Condition::equal, 'foo|bar');

// Perform fuzzy search
$fuzzy = new TextFilter('text', Condition::equal, '%foobaz%');

Geo

Geo filters could be applied to geo fields. Supports only equality conditions, value should be specified as specific-shape array. More information here

use RedisVentures\RedisVl\Query\Filter\GeoFilter;
use RedisVentures\RedisVl\Enum\Condition;
use RedisVentures\RedisVl\Enum\Unit;

$geo = new GeoFilter('geo', Condition::equal, [
    'lon' => 10.111,
    'lat' => 11.111,
    'radius' => 100,
    'unit' => Unit::kilometers
]);

Aggregate

To apply multiple filters to a single query use AggregateFilter. If there's the same logical operator that should be applied for each filter you can pass values in constructor,
if you need a specific combination use and() and or() methods to create combined filter.

use RedisVentures\RedisVl\Query\Filter\AggregateFilter;
use RedisVentures\RedisVl\Query\Filter\TextFilter;
use RedisVentures\RedisVl\Query\Filter\NumericFilter;
use RedisVentures\RedisVl\Enum\Condition;
use RedisVentures\RedisVl\Enum\Logical;

$aggregate = new AggregateFilter([
    new TextFilter('text', Condition::equal, 'value'),
    new NumericFilter('numeric', Condition::greaterThan, 10)
], Logical::or);

$combinedAggregate = new AggregateFilter();
$combinedAggregate
    ->and(
        new TextFilter('text', Condition::equal, 'value'),
        new NumericFilter('numeric', Condition::greaterThan, 10)
    )->or(
        new NumericFilter('numeric', Condition::lowerThan, 100)
    );

Vectorizers

To be able to effectively create vector representations for your indexed data or queries, you have to use LLM's. There's a variety of vectorizers that provide integration with popular embedding models.

The only required option is your API key specified as environment variable or configuration option.

OpenAI

use RedisVentures\RedisVl\Vectorizer\Factory;

putenv('OPENAI_API_TOKEN=your_token');

$factory = new Factory();
$vectorizer = $factory->createVectorizer('openai');

// Creates vector representation of given text.
$embedding = $vectorizer->embed('your_text')

// Creates a single vector representation from multiple chunks.
$mergedEmbedding = $vectorizer->batchEmbed(['first_chunk', 'second_chunk']);

VectorHelper

When you perform vector queries against Redis or load hash data into index that contains vector field data, your vector should be represented as a blob string. VectorHelper allows you to create blob representation from your vector represented as array of floats.

use RedisVentures\RedisVl\VectorHelper;

$blobVector = VectorHelper::toBytes([0.001, 0.002, 0.003]);