No documents yet. Add a PDF, a URL or some text to start.
Chat with PDF answers questions about your documents using retrieval-augmented generation (RAG) that runs entirely in your browser. There is no upload, no account and no API key: your documents are indexed locally with SQLite, and answers come from Gemini Nano (built into Chrome) or an open Qwen3 model running on your GPU through WebGPU.
Extraction, search and the AI model all run in your browser. Your files, questions and answers never reach a server.
Every answer points to the passages it used. Click a citation to read the source text.
If your documents don't contain the answer, it tells you instead of making one up.
Desktop Chrome uses Gemini Nano, its built-in AI. Other browsers can download Qwen3 1.7B once.
Drop PDF, DOCX, TXT, Markdown or HTML files, paste text, or add a URL.
Documents stay in your browser between visits, so you don't re-add them. Clear everything with one click.
Text is extracted from each document in your browser (pdf.js for PDFs, mammoth for Word files).
The text is split into overlapping passages, each turned into a vector by a small embedding model and stored with a keyword index in SQLite, all on your device.
Your question is matched against the passages by meaning and by keywords; the best matches from both are combined.
The top passages go to the AI model with instructions to answer only from them and cite each one.
No. Text extraction, search and the AI model all run inside your browser. Your files, the questions you ask and the answers stay on your device. The only downloads are the tool's code and, if you choose one, an AI model's files.
In desktop Chrome it uses Gemini Nano, the AI model built into Chrome, so there is nothing to download. In other browsers with WebGPU you can download Qwen3 1.7B (~0.9 GB) once; it is cached for later visits. No account or API key is needed.
Your documents are split into passages and indexed two ways: by meaning (a small embedding model) and by keywords (SQLite full-text search). For each question the best-matching passages are combined and given to the AI with instructions to answer only from them and to cite each one. If the passages don't contain the answer, it says so instead of guessing.
PDF, Word (DOCX), plain text, Markdown and HTML files, pasted text, and web pages by URL. Scanned PDFs without a text layer are not supported yet. The free tool holds up to 5 documents and about 300 pages in total.
Most websites tell browsers not to let other sites read their pages (a security rule called CORS). This tool fetches URLs directly from your browser and never through a server, so those sites can't be read. Copy the text and use Add text, or save the page as a PDF. Wikipedia, GitHub and arXiv links are handled automatically.
Yes, in your browser's private storage for this site, so you don't need to re-add them. Chat history is not saved. Use "Clear all my data" to delete everything.
We use cookies for analytics and personalized ads to help keep these tools free. Until you accept, ads stay non-personalized and analytics cookies are off. See our Privacy Policy.