AI Search now has hybrid search and relevance boosting
Key Points
- Hybrid search combines vector and keyword search with configurable fusion
- Relevance boosting prioritizes results by document metadata fields
- Full control over tokenizers, match modes, and ranking strategies
Summary
Cloudflare AI Search now supports hybrid search combining vector and keyword search, plus relevance boosting to control result ranking based on document metadata.
Key Points
- Hybrid Search: Combines vector (semantic) search with BM25 keyword search in parallel queries, fusing results into a single ranked list
- Configurable Tokenizers: Choose between
porterfor natural language ortrigramfor code indexing - Match Modes: Configure keyword matching with
andfor precision ororfor recall - Fusion Methods: Select
rrf(reciprocal rank fusion) ormaxfusion strategy per instance - Relevance Boosting: Nudge rankings using up to 3 metadata fields (e.g.,
timestamp,priority) per instance or per request - Per-Request Overrides: Boost configuration can be customized at query time without changing instance settings
Configuration
Both features are configurable via TypeScript API with instance-level defaults and request-level overrides supported.