Provides a PostgreSQL extension that enables scalable vector search, supporting vector lengths up to 65,535 and binary vector retrieval with 30x reduced memory usage. This solution allows data scientists to build similarity-based applications with high recall, low latency, and full SQL support for filtering and joining data seamlessly.
Funding
Funding not disclosed
Founders
Product
Problem
Existing vector database solutions often lack seamless integration with traditional SQL databases, requiring complex data pipelines and limiting the ability to perform filtered vector searches with high recall. Furthermore, some vector search implementations in PostgreSQL have limitations in filtering capabilities, leading to suboptimal search results.
Solution
PGVecto.rs is a PostgreSQL extension that enables scalable vector search directly within existing Postgres databases, eliminating the need for separate vector stores. It supports vector lengths up to 65,535 dimensions and offers binary vector retrieval, reducing memory usage by up to 30x. The extension allows developers to build similarity-based applications with high recall and low latency, leveraging full SQL support for filtering and joining vector data with other relational data. This approach allows for more powerful and seamless search capabilities, including infinite scroll functionality, by providing as many results as requested for a set of filters.
Target Audience
The primary users are data scientists and software engineers building AI-powered applications who require scalable vector search capabilities within a PostgreSQL environment.
Features
- Supports vector lengths up to 65,535 dimensions, accommodating the latest AI models.
- Binary vector search utilizes adaptive retrieval techniques to maintain accuracy while reducing memory footprint.
- Full SQL support enables filtering and joining vector data with other tables without limitations.
- Hybrid search capabilities combine full-text search with vector search within a single query.
- Offers 4x faster performance compared to other PostgreSQL-based vector search solutions.
- Compatible with existing PostgreSQL infrastructure, minimizing integration efforts.