pgvector-knowledge-patch

What problem does it solve?

This Skill patches gaps in model knowledge about pgvector by providing concise, verified updates to APIs, vector types, indexing options, and query patterns introduced after the model training cutoff so agents do not generate outdated or broken SQL and index designs.

Core Features & Use Cases

• Updated Vector Types: Describes halfvec (half-precision), sparsevec (sparse storage), and bit indexing including dimensional limits and operator classes.
• Indexing & Query Patterns: Explains expression indexes, binary quantization and re-ranking strategies, subvector extraction, L1 support for HNSW, and operator classes for distance metrics.
• Iterative Scan Guidance: Documents iterative index scan modes, scan limits, and CTE patterns to avoid overfiltering when combining approximate search with WHERE clauses.
• Use Case: Migrate an embedding-based search schema to pgvector 0.7+ by adding appropriate expression indexes for quantized or half-precision storage and applying iterative scan settings to maintain recall with selective filters.

Quick Start

Load the pgvector-knowledge-patch and summarize the API, new vector types, operator classes, iterative scan settings, and recommended SQL migration patterns since 0.6.x.