adrianmtanase/laravel-vector-store

Vector store abstraction layer for the Laravel framework.

0.0.25 2024-01-30 15:26 UTC

Requires

• php: ^8.1
• probots-io/pinecone-php: ^0.0.3
• spatie/laravel-package-tools: ^1.14.0
• timkley/weaviate-php: 0.6.2

Requires (Dev)

• laravel/pint: dev-main
• orchestra/testbench: ^8.18
• phpstan/phpstan: ^1.4.7
• phpunit/phpunit: ^10.5.3
• vlucas/phpdotenv: ^5.6@dev

MIT

• Adrian Tanase

• php
• store
• laravel
• vector
• ai
• weaviate
• adrian
• adrian tanase
• adrianmtanase

README

This package provides an implementation of multiple vector databases (e.g. Pinecone.io).

Support us

If this helped you, consider supporting my development over on Patreon or on Github.

Installation

Requires PHP ^8.1

composer require adrianmtanase/laravel-vector-store

Currently supports

• Pinecone.io
• Weaviate

Plans to implement

• MySql - once it's ready

Usage

Using the VectorStore facade, you can easily access any provider and execute operations.

VectorStore::dataset('vector-store')
           ->namespace('general')
           ->upsert(
               PineconeUpsertRequest::build()
                   ->id('1')
                   ->values([
                       -0.002739503,
                       -0.01970483,
                       -0.011307885,
                       -0.011125952,
                       -0.023119587,
                       0.0016207852,
                       -0.003981551,
                       -0.029249357,
                       0.00983842,
                       -0.023721369
                   ])
                   ->metadata([
                       'text' => 'Vector store is lit!'
                   ])
           );

The default provider is Pinecone.io, this can be easily switched using the facade VectorStore::provider(VectorStoreProviderType::PINECONE), or directly in the vector-store config.

Weaviate

As Weaviate runs through GraphQL, the query language is complex. There are several useful methods in WeaviateQueryRequest that will help you query data more efficiently. For example:

VectorStore::provider(VectorStoreProviderType::WEAVIATE)
           ->dataset()
           ->namespace('general')
           ->query(
               WeaviateQueryRequest::build()
                   ->vector([
                       -0.002739503,
                       -0.01970483,
                       -0.011307885,
                       -0.011125952,
                       -0.023119587,
                       0.0016207852,
                       -0.003981551,
                       -0.029249357,
                       0.00983842,
                       -0.023721369
                   ])
                   ->properties(['text'])
                   ->withId()
                   ->withParameters(WeaviateQueryParameters::build()->group('type: closest, force: 1'))
           );

As the system is complex, the package also supports a rawQuery form, or even getting a direct instance of the undelying client.

Weaviate raw query

VectorStore::provider(VectorStoreProviderType::WEAVIATE)
           ->dataset()
           ->namespace('general')
           ->rawQuery('
               {
                Get {
                  General(nearVector: {
                    vector: [
                        -0.002739503,
                       -0.01970483,
                       -0.011307885,
                       -0.011125952,
                       -0.023119587,
                       0.0016207852,
                       -0.003981551,
                       -0.029249357,
                       0.00983842,
                       -0.023721369
                    ]  
                  }) {
                    text
                  }
                }
            }
           ');

Underlying Weaviate client

VectorStore::provider(VectorStoreProviderType::WEAVIATE)
           ->dataset()
           ->client()
           ->batchDelete('general')