Introducing Superbase: Scaling AI Applications with PG Vector

As the CEO and co-founder of Superbase, I'm excited to share how our open-source company is empowering builders to scale their AI applications with PG Vector, a powerful extension that's revolutionizing vector storage and indexing.

  • * Copple is the CEO and co-founder of Superbase, a back-end-as-a-service platform.
  • * Superbase provides a full Postgres database, authentication, edge functions, large file storage, realtime system, and vector offering for every project launched on the platform.
  • * Superbase has been operating since 2020 and is open source, with everything MIT licensed Apache 2 or Postgres.
  • * The company supports existing communities and contributes to existing tools like PG Vector.
  • * Superbase's growth was accelerated by the addition of PG Vector, with thousands of AI applications being launched every week using it in some way.
  • * PG Vector is an open source tool for storing embeddings, powered by PG Vector and contributed to by Superbase.
  • * The sausage-making process in an open source company involves responding to emails from developers like Greg, who created the PG Vector extension, and collaborating with them to merge the extension.
  • * After merging the PG Vector extension, Superbase released Clippy as a doc search interface, which was followed by others in the industry.
  • * Superbase and PG Vector have become part of the AI stack for many builders, working well with platforms like Vercel and Netlify.
  • * Superbase is launching around 12,000 databases a week, with thousands of them using PG Vector in some way.
  • * Some apps launched using Superbase and PG Vector have scaled to millions of users in a short amount of time.
  • * The app being used for the talk was also powered by Superbase.
  • * A tweet criticizing PG Vector's speed and accuracy compared to other vector databases prompted Superbase to work with the ORAL and AWS teams to build in hnsw, resulting in increased queries per seco
  • * Superbase is bullish about Postgres and PG Vector for this particular use case because of the power and extensibility of Postgres.
  • * Joseph, CEO of Roofflow, highlighted the potential of Postgres in a related but slightly different application involving an embedding store for images.
  • * Using Postgres, Copple created a solution involving partitions to segment good and bad cats based on similarity to a canonical cat.
  • * Triggers can be attached to tables in Postgres to run functions for every upload, such as comparing the distance to the canonical cat.
  • * Superbase is extensible and has 30 years of engineering, making it ideal for building AI applications.
  • * PG Vector itself is not built into Postgres but is just an extension that was merged into Superbase as a community contribution.
  • * Postgres offers row-level security, which allows declarative rules to be written on tables for user data and defense at depth.
  • * Storing embeddings next to operational data in Postgres can result in faster fetches with a single round trip.
  • * PG Vector is currently an extension but may be merged into PG core eventually.
  • * Superbase is focused on more enterprise use cases, such as storing billions of vectors, and is working on sharding with the cus extension.
  • * The company is also offering free credits and swag to those who sign up for their platform.

Source: AI Engineer via YouTube

❓ What do you think? What role do you think AI-powered databases, like superbase's PG Vector, will play in shaping the future of data storage and processing? Feel free to share your thoughts in the comments!