# vlite

A simple vector database implemented in plain numpy, with no server or configuration to set up, that stores embeddings in a single portable context file.

vlite is a vector database that needs no separate server or setup: installing the package and importing it is enough to start indexing and searching text. It bakes in a small embedding model, embed-xsmall, and runs entirely in numpy, its only dependency, chunking long text into the model's token window automatically and supporting metadata filtering on stored entries.

Data is stored in a CTX file, a portable format meant to be carried between machines similarly to a browser cookie; saving writes the whole index to one file, and reopening the same name loads it back. It was built to support embeddings used by ChatGPT plugins running for a large number of users.

It is written in Python and released under the MIT license.

Website: http://embeddingsapi.com