Kevin Quillen Releases Alpha7 of Drupal Solr Dense Vector Module

Designed to support vector-based search with semantic ranking, this module integrates AI-generated embeddings into Apache Solr and leverages the Drupal AI module for provider and model selection. Its core aim is to help users get meaningful results even when queries use natural, conversational language.

The new alpha7 release introduces a flexible plugin-based ranking system. Previously, vector ranking logic was hardcoded. Now, developers can define their own algorithms via plugins and apply them per Solr index. The release ships with a default plugin called PureVector and includes an example BlendedRerank plugin to show how Solr’s scoring parameters can be extended.

This update also brings per-index configuration, allowing developers to select embedding providers, models, vector dimensions, and similarity functions. It was tested with OpenAI and Ollama, but supports any provider compatible with the Drupal AI module. If a provider or embedding model changes, Solr must be updated and reindexed.

Solr’s current limitation allows only one dense vector field per index, so the module is best used for specific search cases where semantic ranking is crucial. Full plugin error handling and logging have also been improved.

The module requires Solr 9.6+, Search API Solr 4.x, and the Drupal AI module. This alpha series may continue to evolve, but it marks a major shift toward enabling intelligent, AI-assisted search capabilities in Drupal.