Skip to content

Welcome to vectorstores

vectorstores is a TypeScript framework for building AI applications on top of your data. It’s an abstract layer on top of vector stores, providing a unified interface for working with them. It gives you composable building blocks for ingestion and retrieval, so you can ship production-grade AI applications getting their context from your data.

Built for modern JavaScript runtimes like Node.js, Deno, Bun, Cloudflare Workers, and more.

LLMs are powerful, but they don’t know your private or domain data by default. Context engineering is the practice of selecting the right slices of your data and injecting them into the prompt at the right time.

The most common pattern is Retrieval-Augmented Generation (RAG):

  • Indexing stage: ingest raw inputs, chunk them into Nodes, and compute embeddings.
  • Querying stage: retrieve the most relevant Nodes for a query, optionally rerank them, then generate an answer grounded in that context.

Learn the mental model in the RAG concepts guide.

  • Documents & Nodes: your data as Documents, split into retrievable Nodes. See Documents and Nodes.
  • Ingestion pipelines: configurable transformations (splitters, embeddings). See Ingestion Pipeline.
  • Retrievers: fetch relevant Nodes for a query. See Retriever.
  • Postprocessors: rerank, filter, or otherwise transform retrieved nodes. Example: Cohere Reranker.

The fastest way to get started is in StackBlitz—no local setup required:

StackBlitz IDE showing vectorstores examples

We 💜 contributors! View our contributing guide to get started.