# Vectra

A local, file-backed vector database for Node.js with an optional gRPC server for cross-language access, aimed at applications that want a Pinecone-like index stored on disk.

Vectra is an in-memory vector database backed by files on disk, where each index is a folder, queried with MongoDB-style metadata filtering and cosine-similarity ranking at sub-millisecond latency for small indexes. A browser and Electron build adds IndexedDB storage and local embeddings through Transformers.js, so a project can generate embeddings with Hugging Face models and no API key.

An opt-in gRPC server exposes its functionality to other languages, and a code generator can scaffold client bindings for several of them. It also includes a FolderWatcher that keeps a document index synced with a watched directory, and ships an llms.txt file summarizing its API for coding agents.

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

Website: https://github.com/Stevenic/vectra