Skip to content

Setting up

Requirements

  • Drupal 10.3 or 11.
  • The AI module 1.2 or later, with AI Search, and Key.
  • Solr 9.6 or later. Solr 10 adds quantized vectors (scalar and binary), which hold several times as many vectors in the same memory.

  • A core for the vectors, made from Solr's _default configset. For example, in Docker:

    docker run -d --name solr -p 8983:8983 solr:10.0.0 solr-precreate vectors

    The provider adds its vector fields and their types through the Schema API, and turns off the configset's guessing of new fields.

To run Solr on a server of its own, see Solr on DigitalOcean.

Installing

composer require 'drupal/ai_vdb_provider_solr:^1.0@alpha'
drush pm:install ai_vdb_provider_solr

Connecting to Solr

At Configuration » AI » Vector Database Providers » Solr Configuration (/admin/config/ai/vdb_providers/solr):

Setting What to give
Scheme http, or https behind TLS.
Host The Solr server's host name, without the scheme.
Port 8983 for Solr itself, 443 behind an HTTPS proxy.
Path to Solr The path before /solr/, usually /.
Core or collection The core made for the vectors, such as vectors.
User name For basic authentication; leave empty without it.
Password A Key that holds the basic authentication password.
Vector quantization For new collections: none, scalar or binary. Scalar and binary need Solr 10.
Commit within (ms) How soon indexed chunks become searchable.
Timeout (s) How long each request to Solr may take.

The form pings the core before it saves. The core's address is the scheme, host and port, then the path, /solr/ and the core: with a path of /, https://solr.example.com:443/solr/vectors.

The same settings can be set with Drush:

drush config:set ai_vdb_provider_solr.settings host solr.example.com -y

Creating a collection

  1. At Configuration » Search and metadata » Search API, add a server.
  2. Choose the AI Search backend and an Embeddings Engine.
  3. Choose Apache Solr as the Vector Database.
  4. Under Vector Database Configuration, name a Collection and choose the Similarity Metric.
  5. Leave Include raw embedding vector in results off: the provider indexes vectors without storing them, so it cannot return them.
  6. Save the server.

The collection is created in the core when the server is saved, or, for a server imported from configuration, when it first indexes. A collection's dimension, similarity and quantization are fixed when it is created.