v0.6.0 - pgvector Support

What's New

pgvector Support (prax-pgvector) - New Crate

Full vector similarity search integration for AI/ML workloads via the PostgreSQL pgvector extension.

Vector Types

• Embedding - Dense float32 vectors wrapping pgvector::Vector
• SparseEmbedding - Sparse vectors wrapping pgvector::SparseVector
• BinaryVector - Binary bit vectors wrapping pgvector::Bit
• HalfEmbedding - Half-precision float16 vectors (feature-gated halfvec)

Distance Metrics

• L2 (Euclidean), Cosine, Inner Product, L1 (Manhattan)
• Hamming and Jaccard distances for binary vectors

Search APIs

• VectorSearchBuilder - Fluent API for nearest-neighbor similarity search with filtering, pagination, and metric selection
• HybridSearchBuilder - Combined vector + full-text search using Reciprocal Rank Fusion (RRF) scoring

Index Management

• HNSW and IVFFlat index creation with tuning parameters
• Concurrent index building support
• Pre-configured index profiles: high_recall(), balanced(), high_speed()

Utilities

• Client-side vector math: L2 norm, normalization, dot product, cosine similarity
• Extension management SQL helpers (CREATE/DROP/CHECK pgvector)
• VectorFilter and VectorOrderBy for prax-query WHERE/ORDER BY integration

Test Coverage

• 99 unit tests + 10 doc tests + 36 integration tests against a live PostgreSQL pgvector instance

Installation

[dependencies]
prax-pgvector = "0.6"

Full Changelog: v0.5.1...v0.6.0