lucisearch

The SQLite/DuckDB of search — an embeddable, in-process search engine written in Rust.

No cluster to manage. No HTTP layer. No JVM. Open a file and search.

Crate name vs. import name: this library publishes to crates.io as lucisearch (the bare luci name was already taken), but the crate itself is luci — add lucisearch and write use luci::…. Same story on PyPI: install lucisearch, import luci.

Features

• Full-text search with BM25 scoring, analyzers, phrase queries
• Vector search with HNSW, int8 quantization, pre-filtering
• Hybrid search with Reciprocal Rank Fusion
• Aggregations — terms, avg, sum, min, max, stats, range, histogram, cardinality, percentiles
• Geospatial — geo_distance, geo_bounding_box, geo_shape (intersects, within, disjoint, …)
• Nested documents with block-join queries and inner_hits
• Single-file storage — one .luci file, no directory sprawl
• Lazy result materialization — only compute source, highlight, explain on access

Quick Start

[dependencies]
lucisearch = "0.8"

use luci::index::Index;
use luci::search::expression::SearchExpression;
use serde_json::json;

let schema = json!({
    "properties": {
        "title": {"type": "text"},
        "tag":   {"type": "keyword"},
    }
});

let mut index = Index::create("my_index.luci", schema)?;

index.bulk(vec![
    json!({"title": "Hello world", "tag": "greeting"}),
    json!({"title": "Search engines", "tag": "tech"}),
])?;

// Queries are ES-compatible JSON.
let expr = SearchExpression::from_json(json!({"match": {"title": "hello"}}), 10)?;
let results = index.search(&expr)?;
for hit in results.iter() {
    println!("{:.2} {:?}", hit.score(), hit.source());
}

Field Types

Python

Also available as a Python package — pip install lucisearch (import luci) — with the same ES-compatible dict API plus a typed, Pydantic-validated query builder.

License

MIT