Personal RAG - Private Semantic Search

Personal RAG is a private, offline-capable semantic search engine that runs entirely in your browser. It indexes your documents and lets you query them using natural language — finding meaning rather than just matching keywords. Under the hood, it chunks documents, generates embeddings using the all-MiniLM-L12-v2 model via ONNX Runtime Web, and stores the vector index in the browser's Origin Private File System (OPFS). Your documents never leave your device, making this ideal for private research, personal knowledge bases, or sensitive files you don't want in the cloud.

On upload, documents are parsed in a Web Worker (PDF via pdf.js, DOCX via mammoth, text files directly, images via Tesseract OCR). They are chunked using a recursive splitter and each chunk is embedded using all-MiniLM-L12-v2 running via ONNX Runtime Web. The vector index is stored in OPFS. On query, your search phrase is embedded and cosine similarity is computed against all chunks. Top-k results are returned with their relevance scores.