Crate prax_pgvector 

prax-pgvector

pgvector integration for the Prax ORM — vector similarity search, embeddings, and index management for PostgreSQL.

This crate provides type-safe wrappers around pgvector functionality, integrating with the prax-postgres driver for seamless vector operations.

Features

• Vector types: Embedding, SparseEmbedding, BinaryVector (and [HalfEmbedding] with the halfvec feature)
• Distance metrics: L2, inner product, cosine, L1, Hamming, Jaccard
• Index management: IVFFlat and HNSW index creation with tuning parameters
• Query builder: Fluent API for vector similarity search
• Hybrid search: Combined vector + full-text search using Reciprocal Rank Fusion
• Filter integration: Vector filters for prax-query WHERE/ORDER BY clauses

Quick Start

use prax_pgvector::prelude::*;

// Create an embedding
let query = Embedding::new(vec![0.1, 0.2, 0.3, /* ... */]);

// Build a similarity search
let search = VectorSearchBuilder::new("documents", "embedding")
    .query(query)
    .metric(DistanceMetric::Cosine)
    .limit(10)
    .ef_search(200) // Tune HNSW recall
    .build();

let sql = search.to_sql();
// SELECT *, embedding <=> $1 AS distance FROM documents ORDER BY distance LIMIT 10

Index Management

use prax_pgvector::index::{VectorIndex, HnswConfig};
use prax_pgvector::DistanceMetric;

// Create an HNSW index
let index = VectorIndex::hnsw("idx_doc_embedding", "documents", "embedding")
    .metric(DistanceMetric::Cosine)
    .config(HnswConfig::high_recall())
    .concurrent()
    .build()
    .unwrap();

println!("{}", index.to_create_sql());
// CREATE INDEX CONCURRENTLY idx_doc_embedding ON documents
//   USING hnsw (embedding vector_cosine_ops)
//   WITH (m = 32, ef_construction = 128)

Feature Flags

Feature	Description
halfvec	Enable half-precision (float16) vector support

Re-exports

pub use error::VectorError;

pub use error::VectorResult;

pub use ops::BinaryDistanceMetric;

pub use ops::DistanceMetric;

pub use types::BinaryVector;

pub use types::Embedding;

pub use types::SparseEmbedding;

Modules

error

Error types for pgvector operations.

filter

Vector filter operations for integration with the prax query builder.

index

Vector index management for pgvector.

ops

Distance operators and similarity metrics for pgvector.

prelude

Prelude for convenient imports.

query

High-level query builder for vector similarity search.

types

Core vector types wrapping pgvector with Prax ORM integration.