pgEdge Now Supports pgvector, Introducing Similarity Search to PostgreSQL

pgEdge, the first company to offer a fully distributed database optimized for the network edge based on the standard and popular open source PostgreSQL database, is announcing newfound support for the pgvector extension, enabling PostgreSQL to leverage the power of open source vector similarity search.

The added support for pgvector opens a world of possibilities for AI and machine learning (ML) use cases for PostgreSQL.

“From our inception, we were inspired by AI and machine learning use cases for our distributed Postgres database but did not incorporate support for the pgvector extension until several customers began asking for it,” explained Phillip Merrick, co-founder and CEO of pgEdge.

The extension allows pgEdge to store vector embeddings, which is particularly useful for supporting natural language processing (NLP) applications that depend on large-scale, high dimensional data.

pgvector further offers support for three query operators designed specifically for similarity search, including Euclidean, negative inner product, and cosine distance. These are distance metrics that “provide distance calculations between two points, or between two vectors, and make it possible to know if two vectors are close to each other. If two vectors are found to be close, then we are able to infer that the objects they represent (e.g., pictures of a cat) are similar,” said Merrick.

Support for Euclidean, negative inner product, and cosine distance allows pgEdge users to represent and compare data with great efficiency, surfacing meaningful insights from diverse datasets and enabling users to make decisions based on similarity or dissimilarity, according to Merrick.

Additionally, pgvector introduces ivfflat, or inverted file with stored vectors, which helps to accelerate the search for similar vectors within a large dataset, ultimately increasing performance.

“The combination of pgvector and pgEdge makes for much more performant AI applications, particularly those that need to be deployed globally,” said Merrick. “For some applications, the latency of a centrally located database means they simply can't be deployed on a global basis —and for some that can be a literal game changer.”

To learn more about pgEdge and the pgvector extension, please visit