What is the AI Document Q&A tool?

It is a privacy-first document question-answering tool that runs entirely in your browser. Upload PDF, Markdown, or TXT files and ask natural language questions. The AI finds relevant passages and generates accurate answers with citations - all without sending any data to external servers.

What file types are supported?

The tool supports PDF files (.pdf), Markdown files (.md, .markdown), and plain text files (.txt). PDF text is extracted page by page, Markdown preserves heading structure for citations, and text files track line numbers.

Is my data private and secure?

Yes, completely. All processing happens in your browser using Web Workers. Your documents, questions, and answers never leave your device. No data is sent to any server. The AI models run locally via WebGPU or WebAssembly, making this safe for confidential and proprietary documents.

How large can documents be?

There is no strict file size limit, but very large documents (over 100 pages) may take longer to process since every chunk needs to be embedded. For best performance, documents under 50 pages or 200KB of text work well. You can upload multiple smaller documents instead of one very large one.

Which AI models are used?

The tool uses three AI models: nomic-embed-text-v1.5 for semantic embedding (256 dimensions, q4f16 quantized), Xenova/bge-reranker-base as a cross-encoder for reranking search results, and Qwen2.5-0.5B-Instruct for generating natural language answers. All models use WebGPU acceleration with WASM fallback.

How accurate are the answers?

Answers are grounded in your documents through the RAG pipeline. The system retrieves the most relevant passages, reranks them with a cross-encoder for precision, then generates answers based only on those passages. However, small language models can sometimes produce imperfect responses. Always check the cited source passages shown below the answer.

Why does the first load take longer?

On first use, three AI models need to download (approximately 300-500MB total depending on the device). After the initial download, models are cached in your browser storage for instant loading on subsequent visits. Use WiFi for the first download to avoid mobile data charges.

AI Document Q&A

Ask questions about your PDF, Markdown & text documents using AI. Privacy-first RAG pipeline with vector search runs 100% in your browser

ai document qa

document question answering

pdf qa

Upload Documents

Drag and drop a document here

or click to select a file

Choose Document

Supports PDF, Markdown (.md), and Text (.txt) files

Ask a Question

Loading Document Q&A...

How to Use

Upload one or more documents (PDF, Markdown, or TXT files).
Wait for the document to be processed (text extraction, chunking, and embedding).
Type your question in the input field and press Enter or click Send.
View the AI-generated answer with cited source passages from your documents.

How it Works

This tool uses a full Retrieval-Augmented Generation (RAG) pipeline running entirely in your browser:

1. Extract Text

Text is extracted from uploaded PDFs, Markdown, or plain text files.

2. Chunk into Passages

Documents are split into overlapping passages using a sliding window approach.

3. Embed with AI

Each passage is embedded using nomic-embed-text-v1.5 into 256-dimensional vectors.

4. Vector Search

Your question is embedded and matched against passages using Orama vector search.

5. Cross-Encoder Reranking

Top candidates are reranked with bge-reranker-base for precision.