HYBRID TEXT AND VECTOR SEARCH
Combine keyword and vector search.Combine keyword and vector search.
Hybrid search combines full-text relevance with vector similarity. Match useful terminology while finding related products, documents and answers that use different words, within your index scope and authorisation rules.
Combine words with meaning
Tune the ranking balance
Return authorised results
How hybrid search works
Text search
Match analysed words in the indexed text fields.
Vector search
Compare embeddings supplied by your application.
Hybrid search
Combine the two relevance signals using the index's configured text and vector weights.
Scope and authorisation
Search the selected bucket and index. Custom-realm queries apply the supplied subject and authorisation relation.
Relevant results linked to the original objects
TURN SEARCH INTO PROGRESS
Less searching. More solving.
People do not always use the language in your catalogue or knowledge base. Exact words and semantic similarity each solve part of that problem. Hybrid ranking lets your application use both without running two unrelated search services.
Better discovery
Help customers and support teams find useful content when their wording differs.
Keep access separate from relevance
A high similarity score does not grant permission to read an object.
One approach, several uses
Apply text and vector ranking to commerce, support content and internal knowledge.
REAL SITUATIONS. PRACTICAL SEARCH.
Where hybrid search makes a difference.
Commerce
“headphones for the commute”
Match product descriptions while using embeddings to find related catalogue items whose names do not repeat the customer's phrase.
Rank useful alternatives with both text and semantic signals.
Support
“I cannot get into my account”
Find relevant help articles even when the customer describes a problem differently from the article title.
Keep terminology useful without depending on identical wording.
Multi-tenant applications
“policy for refunds”
Apply the query's custom-realm subject and relation so related content remains subject to the application's access rules.
Return relevant content that the selected subject may see.
Combine text and an application-supplied vector in a hybrid query. Use typed JSON queries for declared exact-field filters such as category or availability. Your application supplies embeddings and any text extracted from images or documents.
SEARCH THAT WORKS FOR YOUR APPLICATION
Rank results with both text and vector similarity.
Keep text, embeddings and original objects in one platform. Choose a ranking balance using representative queries from your own application.
