README

Open-source semantic (vector) search for Bagisto — 100% self-hosted, no paid APIs.

Stack: Qdrant (vector DB, Apache-2.0) + fastembed running BAAI/bge-small-en-v1.5 (ONNX, CPU) as a small local embedding service.

Features

• Semantic product search — overrides ProductRepository::getAll() for the storefront search box (query is embedded → nearest vectors → results in relevance order, with category/price filters preserved). Falls back to Bagisto's DB search if the vector service is down.
• Category-grouped autosuggest — GET /vector-suggest?query= returns products grouped by category (name, price, image, attributes) served entirely from the Qdrant payload.
• Vector-powered related & up-sell — overrides /api/products/{id}/related and /up-sell: manual admin relations first, gap filled with vector nearest-neighbours (visibility-filtered).
• Auto re-index — listens to catalog.product.create/update/delete.after; keeps the index in sync on every product save.

Requirements

• Qdrant reachable at VECTOR_QDRANT_URL (default http://127.0.0.1:6333).
• Embedding service (fastembed FastAPI) at VECTOR_EMBED_URL (default http://127.0.0.1:8500) exposing POST /embed {texts, is_query}.

Install (local package)

1. Add to root composer.json autoload: "Rbk\\VectorSearch\\": "packages/Rbk/VectorSearch/src" then composer dump-autoload.
2. Register Rbk\VectorSearch\Providers\VectorSearchServiceProvider::class in bootstrap/providers.php.
3. Add env:

   VECTOR_SEARCH_ENABLED=true
   VECTOR_QDRANT_URL=http://127.0.0.1:6333
   VECTOR_EMBED_URL=http://127.0.0.1:8500
   VECTOR_COLLECTION=products
   VECTOR_DIM=384
   VECTOR_LIMIT=80
   

4. Create the Qdrant collection (size 384, Cosine), then build the index:

   php artisan vector-search:index
   

Set VECTOR_SEARCH_ENABLED=false to cleanly revert to Bagisto's native search/related.

License

MIT