AI Search - AI Search now has hybrid search and relevance boosting
Key Points
- Hybrid combines semantic vectors + BM25
- Configure tokenizer, match mode, and fusion
- Boost ranking with up to 3 metadata fields
Summary
AI Search adds hybrid search (vector + BM25 keyword) and relevance boosting. Hybrid search runs semantic vector and BM25 keyword retrieval in parallel and fuses results; relevance boosting lets you nudge ranks using document metadata (e.g., timestamp, priority). Both are configurable per instance and can be overridden per request.
Key Points
- Hybrid search
- Runs vector and BM25 keyword retrieval in a single query and fuses results into one ranked list.
- Configure per instance with
index_method: { vector: true, keyword: true }and choosefusion_method(rrformax). - Tokenizer choices:
porterfor natural language,trigramfor code. Setkeyword_match_modetoand(precision) oror(recall).
- Relevance boosting
- Boost rankings using up to 3 metadata fields (set per instance or per request via
boost_by). - Typical uses: boost recent content (
timestamp) or high-priority content (priority); specifydirection(e.g.,desc).
- Boost rankings using up to 3 metadata fields (set per instance or per request via
- Practical notes
- Prefer
porterfor docs andtrigramfor code indexes; use per-request boost overrides for special queries. - See the Search modes and Relevance boosting docs for configuration details and examples.
- Prefer