WeftKitVec
Vector Similarity Search
Disk-resident approximate nearest neighbor search for RAG pipelines, semantic search, and recommendation systems. Handles billion-point datasets on a single machine.
flat search (100 vectors, 128-dim)
Key Features
What makes WeftKitVec stand out.
HNSW Graph Index
Hierarchical Navigable Small World graph with delta updates for real-time insertions. SIMD-accelerated distance kernels (AVX2/NEON).
DiskANN / Vamana
SSD-resident graph index for billion-scale datasets. In-memory navigation graph with on-disk full-precision vectors for re-ranking.
Product Quantization (IVF-PQ)
k-means clustering with PQ compression. 8–32x memory reduction with configurable recall vs. memory tradeoff.
Metadata Filtering
Pre-filtering and post-filtering with adaptive strategy based on filter selectivity. Combine vectors with structured attributes.
Hybrid Search
Combine vector similarity with keyword BM25 scores via reciprocal rank fusion for enhanced retrieval quality.
Multiple Distance Metrics
Cosine similarity, Euclidean (L2), inner product (dot), Hamming (binary), and custom metrics via function pointer.
Performance Targets
Real numbers from Criterion benchmarks on reference hardware.
Wire Protocol
WeftKitVec supports the gRPC protocol. Your existing tools just work.
Get Started in Seconds
Add WeftKitVec to your Rust project and start querying.
use weftkitvec::prelude::*;
// Open or create a database
let db = Database::open("my_app.db")?;
// Use the module API
let result = db.query("SELECT * FROM users WHERE active = true")?;
for row in result.rows() {
println!("{:?}", row);
}Where to Use It
RAG pipeline semantic retrieval
Recommendation systems
Image and audio similarity search