Expand description
Retrieval operations: local embedding generation and hybrid search with RRF fusion.
Structs§
- Document
Embedding Outcome - What the embedding service actually received for one document.
- Embedding
Truncation Report - Aggregate truncation accounting for one logical write or reindex batch.
- Hybrid
Search Outcome - Result of
KhiveRuntime::hybrid_search_outcome: the fused hits — text hits alone when the vector arm failed — plus the vector arm’s error, if any. - Search
Hit - A unified search result combining vector and text signals.
- Search
Signals - Retained component scores before fusion and strategy-local modifiers. An absent retrieval leg has no score, which is distinct from a measured zero. Scores belong to the backend and model that produced the retained hit; vector similarities from different embedding models are not comparable.
Enums§
- Rank
Score Kind - The strategy that produced a hit’s ordering score, including local modifiers.
- Search
Source - Which retrieval path(s) contributed to a hit.
Constants§
- EMBEDDING_
INPUT_ TRUNCATED_ WARNING - Advisory emitted by write verbs when only the embedding input was bounded.
Functions§
- bounded_
embedding_ input - Return the longest UTF-8-safe prefix of
textwithinmax_bytesand whether it was shortened. - document_
embedding_ budget - Maximum document bytes accepted before the embedding service adds its model prefix.