sapphire-retrieve

Full-text and semantic search library extracted from sapphire-journal.

What this crate provides

FTS5 — trigram full-text search over a SQLite database (RetrieveDb::search_fts)
Vector search — approximate nearest-neighbour search via sqlite-vec or LanceDB (RetrieveDb::search_similar)
Chunker — splits documents into overlapping text chunks for embedding (chunker::chunk_document)
Embedder trait — pluggable embedding backends (build_embedder)
- openai — OpenAI-compatible REST API
- ollama — local Ollama server
- fastembed (feature: fastembed-embed) — local ONNX inference, no server required
Config types — RetrieveConfig, VectorDb, EmbeddingConfig in sapphire_retrieve::config
LanceDB store (feature: lancedb-store) — high-performance columnar vector store

Feature	Default	Description
`sqlite-store`	no	SQLite FTS5 + sqlite-vec backend
`lancedb-store`	yes	LanceDB vector backend
`fastembed-embed`	yes	Local ONNX embedding via fastembed

MIT OR Apache-2.0