TurboVec VDB Provider Adds Self-Hosted Vector Search for Drupal AI

Developers working with the Drupal AI module can test a self-hosted vector database option through TurboVec VDB Provider, a contributed module created by Nick Opris, senior Drupal developer. The module connects Drupal AI’s vector database layer with turbovec-docker, a Dockerised HTTP API for storing, searching, and deleting vectors without relying on an external cloud service.

The project matters for Drupal AI adoption because semantic search often requires a vector store, and many site teams need local or self-hosted testing before committing to a managed service. The Drupal.org project page positions TurboVec VDB Provider for development, staging, and small-to-medium production deployments. It also states that the project is not covered by Drupal’s security advisory policy.

The provider supports semantic vector search through the AI Search backend, collection creation and deletion, entity insertion, filtered search, metadata-only queries, entity deletion, offset-based pagination, optional bearer-token authentication, and entity ID filtering. The entity ID allowlist is intended to keep search results scoped to Drupal content the current user is permitted to see after access-check retries. The project also includes safeguards for corrupted index states where vectors exist without matching metadata.

Requirements include the Drupal AI module, including the ai_search submodule, the Key module, a running turbovec-docker instance, and Drupal 10.2 or Drupal 11. The project page instructs site builders to start the Docker service withdocker compose up --build, enable the module with drush en ai_vdb_provider_turbovec and drush cr, then configure /admin/config/ai/vdb_providers/turbovec. DDEV users can point the provider to http://host.docker.internal:8000.

The GitHub repository describes turbovec-docker as a Dockerised HTTP API built on Ryan Codrai’s TurboVec project using the Python IdMapIndex binding. The API listens at http://localhost:8000, stores index files under ./data as .tvim files, and exposes routes for health checks, collection operations, entity insertions, vector searches, metadata queries, and deletions. Authentication is optional through API_BEARER_TOKEN.

The repository also includes Drupal-style node test data using stable uint64 identifiers and metadata fields such as title, body, type, and URL. Its README says 59 tests cover collections, insert, search, query, delete, bearer authentication, and full Drupal end-to-end flows. The test suite runs inside the Docker container rather than requiring a local Python installation.

In a LinkedIn post, Nick presented the provider as a Drupal AI vector provider for TurboVec and linked both the Drupal.org project and GitHub repository. The post includes a performance comparison with FAISS, but no benchmark methodology is supplied in the source material. That comparison should be treated as a project claim unless independent test details are published.

For Drupal teams evaluating retrieval-augmented generation and semantic search, TurboVec VDB Provider creates a local testing path around Drupal AI’s vector database provider layer. Its practical value will depend on production hardening, security review, and the project's handling of large datasets, access-aware search, and long-term maintenance.