zvec-rust 0.1.0

Rust-native vector collection engine with a Python API (WAL durability, ANN, hybrid search).

zvec-rust

zvec-rust is a Rust-native vector collection engine with a compiled Python API. It provides durable storage (snapshot + write-ahead log), dense/sparse vector search, scalar filtering, full-text search, and hybrid (multi-route) retrieval with pluggable rerankers.

Install

pip install zvec-rust

Quick start

import zvec_rust as zvec

schema = zvec.CollectionSchema(
    name="docs",
    vectors=[zvec.VectorSchema("embedding", zvec.DataType.VECTOR_FP32, dimension=4)],
    fields=[zvec.FieldSchema("category", zvec.DataType.STRING)],
)

col = zvec.create_and_open("./data/docs", schema, durability_mode="interval")
col.insert([
    zvec.Doc("a", vectors={"embedding": [1.0, 0.0, 0.0, 0.0]}, fields={"category": "blog"}),
    zvec.Doc("b", vectors={"embedding": [0.0, 1.0, 0.0, 0.0]}, fields={"category": "news"}),
])

# Vector search with a scalar filter.
hits = col.query(
    zvec.Query(field_name="embedding", vector=[1.0, 0.0, 0.0, 0.0]),
    topk=5,
    filter="category == 'blog'",
)

# Hybrid (multi-route) search fused by a reranker.
hits = col.query(
    [
        zvec.Query(field_name="embedding", vector=[1.0, 0.0, 0.0, 0.0]),
        zvec.Query(field_name="embedding", vector=[0.0, 1.0, 0.0, 0.0]),
    ],
    topk=5,
    reranker=zvec.RrfReRanker(),
)

Highlights

• Durability modes: always, interval, manual (WAL + checkpoint + manifest).
• Crash-recovery: committed ops are replayed from the WAL on reopen.
• ANN index params: HNSW, IVF, Flat, DiskANN, Vamana, HNSW-RaBitQ.
• Hybrid retrieval: RrfReRanker, WeightedReRanker, CallbackReRanker.
• Non-throwing *_status API surface returning Status objects.

See the project repository for the full API reference, production guide, and parity audit.

Licensed under Apache-2.0.