pgvector : Open-source vector similarity extension for PostgreSQL

pgvector is an open-source PostgreSQL extension that adds support for vector similarity search within the database. It allows developers to store and query high-dimensional vector embeddings directly in PostgreSQL, enabling semantic search, recommendation systems, and AI-powered retrieval tasks without needing external vector databases.

By bringing vector operations natively into a relational database, pgvector helps simplify application architecture, maintain consistency across data systems, and use existing PostgreSQL features such as indexing, transactions, and security. It’s a strong option for teams already using PostgreSQL who want to add AI capabilities with minimal overhead.

Key benefits include:

• Native vector storage and search in PostgreSQL
• Simple integration with existing PostgreSQL apps and tools
• Flexible indexing for accurate and efficient nearest neighbor queries

What are the main features of pgvector?

Native vector type in PostgreSQL

pgvector introduces a new vector column type, allowing direct storage of fixed-length float vectors.

• Store vectors like embeddings (e.g., from OpenAI, Hugging Face, etc.)
• Supports common operations such as dot product, cosine similarity, and Euclidean distance
• Fully integrated with SQL syntax and PostgreSQL tooling

Similarity search within SQL

pgvector enables k-nearest neighbor (k-NN) queries directly in SQL using familiar operators.

• Use <-> for Euclidean distance, <#> for cosine distance, and <=> for inner product
• Perform filtering and ordering in combination with vector similarity
• Seamlessly combine structured and unstructured data queries

Indexing for efficient search

To accelerate similarity queries, pgvector supports indexing strategies optimized for performance.

• ivfflat index for approximate nearest neighbor (ANN) search
• Supports filtering on other columns (e.g., metadata) while using the index
• Index build requires offline training (clustering centroids)

Works with PostgreSQL extensions and tools

pgvector is fully compatible with the broader PostgreSQL ecosystem.

• Use alongside extensions like PostGIS or full-text search
• Supported by ORMs such as Django, SQLAlchemy, Prisma, and more
• Deployable on major platforms including AWS RDS, Azure Database for PostgreSQL, and Supabase

Lightweight and easy to deploy

pgvector adds minimal complexity to your PostgreSQL setup.

• Just install the extension and create vector columns
• No external services, APIs, or separate databases required
• Ideal for full-stack apps, SaaS platforms, and internal tools

Why choose pgvector?

• Integrated with PostgreSQL: Leverages the reliability, tooling, and familiarity of a mature relational database
• All-in-one storage: Store vectors and structured data together in a single system
• Efficient similarity search: Supports both exact and approximate nearest neighbor queries
• Flexible and developer-friendly: Easy to query, index, and combine with other SQL features
• Open source and production-ready: Actively maintained and used in real-world AI applications