filtered-search

Filtered search in Milvus applies scalar field constraints before or alongside vector similarity matching. Configure attribute filters on indexed fields like category, price, or date, then run your query to retrieve semantically relevant results that also satisfy your metadata conditions.

Pre-filter scalar fields before executing vector search to enforce constraints while maintaining relevance ranking. This approach improves latency and result quality by narrowing the search space to records matching your metadata criteria first.

Yes. Filtered search supports multi-field constraints and category hierarchies, enabling faceted workflows where you combine vector similarity with filters across multiple metadata dimensions simultaneously.

Pre-filtering is more efficient in large datasets because it reduces the number of vectors to compare before ranking. Filter first on scalar fields to constrain results, then execute vector search on the filtered subset for faster latency and better relevance.

Filtered search enforces common comparison operators and supports array checks on scalar fields, allowing you to match tags, lists, and range-based constraints like price ranges or date windows.

Filtering by attributes like product category, price tier, or document type ensures results are both semantically similar and contextually appropriate. This combination eliminates irrelevant matches and narrows results to your exact domain or business constraints.