rishabhguptajs/pgvec-studio

pgvec-studio is a tool for looking at the vector embeddings stored in a PostgreSQL database that uses the pgvector extension. Embeddings are long lists of numbers that represent things like text or images in a way a computer can compare, and normally there is no easy way to see what they actually look like. This project gives you a visual way to explore them. You paste in a database connection string, choose a table that has a vector column, and click a button to visualize it. The tool takes the high dimensional numbers and squishes them down into a 2D scatter plot using a technique called UMAP, which tries to keep similar items close together on the chart. That means embeddings that represent similar content will cluster near each other, so you can actually see patterns in your data instead of guessing. Beyond the chart, you can click on any point to see its metadata, search for the nearest neighbors of a specific vector, filter rows before visualizing so you are not overwhelmed by millions of points, and compare two rows directly to get a cosine similarity score. There is also an export option so you can save the filtered data as CSV or JSON for use elsewhere. Everything runs on your own computer through a local Next.js server. Nothing is uploaded anywhere, there is no account to create, and your connection string only lives for the length of your browser session. The app only reads from your database, so a read-only database user works fine. To use it, you need PostgreSQL with the pgvector extension already set up and at least one table with a vector column, plus Node.js version 18 or newer installed on your machine. Setup is a standard clone, install, and run sequence. The project is built with Next.js, TypeScript, and Tailwind CSS, with node-postgres handling the database connection and umap-js doing the dimensionality reduction in JavaScript. It is released under the MIT license.