Getting started

pip install recern-vector           # Python 3.11+
cargo add recern-vector             # Rust library
cargo install recern-vector-cli     # the recern-vector command

import numpy as np
import recern_vector as rv

with rv.Database.open_or_create("docs.rvec") as db:      # saved on a clean exit
    docs = db.create_collection("docs", dim=384, metric="cosine")

    # ids, a float32 array of shape (n, 384), and one metadata dict per record
    docs.upsert_many(ids, embeddings, metadatas)

    hits = docs.search(query, k=5, filter={"lang": "en"})
    for hit in hits:
        print(hit.id, hit.distance, hit.metadata)

    print(docs.explain(query, k=5))      # strategy, nodes visited, time

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
embeddings = model.encode(texts, normalize_embeddings=True)   # float32, (n, 384)
query = model.encode("how do I make my queries faster?", normalize_embeddings=True)

use recern_vector::{CollectionConfig, Database, Filter, Metric, SearchOptions};
use serde_json::json;

let mut db = Database::open_or_create("docs.rvec")?;
let docs = db.create_collection("docs", CollectionConfig::new(384, Metric::Cosine))?;
docs.upsert("doc-1", &embedding, Some(json!({"lang": "en"})))?;
docs.upsert_many(records)?;   // iterator of (id, vector, Option<metadata>)

let options = SearchOptions::default().filter(Filter::eq("lang", "en"));
let hits = docs.search(&query, 5, &options)?;
db.save()?;

recern-vector init docs.rvec
recern-vector create-collection docs.rvec docs --dim 384 --metric cosine
recern-vector insert docs.rvec docs records.jsonl
recern-vector query docs.rvec docs --like doc-42 -k 5 --explain
recern-vector inspect docs.rvec