NextJS-RAG – opinionated SQLite RAG (github.com)

NextJS-RAG is a lightweight, opinionated RAG toolkit for Next.js that uses SQLite (via sqlite-vec) as an embedded vector store so you don’t need an external DB. It provides a CLI to index text files (npx nextjs-rag init ./docs), incremental re-indexing for changed files, and a tiny runtime API (queryRag) that returns combined context and source citations. It’s built for serverless deployments—on Vercel/Netlify it auto-detects serverless environments, copies a bundled .rag/sqlite.db to /tmp on cold starts, and can either commit the DB into your repo for faster starts or regenerate at build time. It supports OpenAI embedding models (text-embedding-3-small/large, ada variants), configurable chunk size/overlap, topK, and simple CLI querying. For developers this matters because it removes vector DB infrastructure friction: fast vector search inside SQLite lowers cost and complexity for building knowledge-backed apps in Next.js. Key tech notes: uses sqlite-vec for vector ops, supports smart chunking (recommend 500–800 chars for precision or 1500+ for context), incremental reindexing only updates changed files, and the package exposes configuration for embedding model, chunking, and DB path. Limitations: no native image/PDF/video support without preprocessing. Licensed MIT and designed to be drop-in for small-to-medium scoped RAG use cases on Node/serverless hosts.