VECTOR SEARCH
Find similar products, images and documents.Find similar products, images and documents.
Create embeddings in your application. Keldra stores and queries those vectors alongside your objects, so your application can find similar items by meaning, not just exact words or filenames.
Similarity search over the embeddings you supply.
HOW IT WORKS
Embeddings live alongside your data in Keldra.
Your application converts text, images or other content into embeddings. A vector index searches those numerical representations while the original products, documents and files stay in object storage.
1
Create embeddings in your application using the model that suits your content.
2
Store the vectors in indexed fields linked to your original objects. Declare their dimensions and supported distance measure.
3
Submit a query vector to find nearby items. Fetch the original objects your application needs.
PRODUCT DISCOVERY
Find visually or semantically similar products.
Input
Blue running shoe
Your application creates an embedding from the product image or description and sends it to Keldra.
Similar content
A vector query retrieves nearby products from your indexed catalogue, subject to the caller's authorisation.
Input
Blue running shoe
Your application creates an embedding from the product image or description and sends it to Keldra.
Similar content
A vector query retrieves nearby products from your indexed catalogue, subject to the caller's authorisation.
REAL BUSINESS SCENARIOS
Recommend alternatives and find related content.
Catalogue alternatives
Help customers find related products when an item is out of stock, or discover alternatives without knowing the exact product name.
Related knowledge
Retrieve relevant internal documents, policies and support articles even when their wording differs from the question.
Content similarity
Search image, video or document libraries using your chosen model's embeddings. Keep exact field filters and full-text search available for the questions similarity alone cannot answer.
Catalogue alternatives
Help customers find related products when an item is out of stock, or discover alternatives without knowing the exact product name.
Related knowledge
Retrieve relevant internal documents, policies and support articles even when their wording differs from the question.
Content similarity
Search image, video or document libraries using your chosen model's embeddings. Keep exact field filters and full-text search available for the questions similarity alone cannot answer.
Add catalogue and document similarity search.
Run it on your own servers.
Your model supplies the embeddings; Keldra does not generate them or recognise images automatically. Approximate nearest-neighbour search trades search effort against recall. Measure results with your own data; a similarity score does not establish identity or permission.