Documentation The current backend implements a broad search stack designed for commerce and structured catalog use cases.
Core search behavior
- hybrid lexical plus semantic blending
- FTS-only, vector-only, hybrid, and FTS-rerank modes
- prefix-aware autocomplete with query normalization
- spell correction before final retrieval
- synonym expansion based on schema configuration
- exact filters and numeric range filters
Result tuning
- per-index ranking controls
- lexical and semantic weight balancing
- boost rules and optional scoring webhooks
- pinned and curated merchandising results
- display field mapping for title, subtitle, image, and URL output
Typeahead path
/v1/typeahead uses a precomputed suggestions table instead of full FTS queries on every keystroke.
That makes it a strong fit for:
- nav search bars
- mobile query assist
- low-latency prefix suggestion UIs
Intelligence features
- query-time embeddings for semantic retrieval
- optional benchmark mode across vector backends
- knowledge graph scoring hooks
- personalization scores using stored user profiles
- query entity extraction for debug and graph expansion
When to use which endpoint
- Use
/v1/typeaheadfor fast label suggestions while the user is typing. - Use
/v1/autocompletewhen you need ranked product or content cards with richer payloads. - Use personalization and graph features when you have enough behavioral or catalog structure to benefit from them.