Guide
Chat with a PDF
Reading a long document to find one clause is a poor use of an afternoon. Open the PDF, ask what you actually want to know, and keep asking — each answer comes from the document itself, not from what a model happens to remember.
Open a PDF and start asking PDF, Word, PowerPoint, PNG, JPG, WebP, Text · XLSX, CSV, Parquet, JSONOpen the document
Drop the PDF onto the page. The first page is drawn in your own browser, so you can confirm it is the right file before anything is sent. Scans and photographed pages work too — the text in the image is read directly.
Ask in your own words
“What is the notice period?”, “what does it say about late payment?”, “who signed this and when?”. The answer is built only from the document, and where the document is silent it says so instead of filling the gap with something plausible.
Keep going
One question rarely settles anything. Ask the next one, and the one after — a summary first, then the detail, then the clause you were actually worried about.
What a chat with your document costs you
With most tools, the answer is a copy of your document on someone's server. Here the PDF passes through to the model and the answer comes straight back: no document store, no document database, no content log, and nothing used to train a model. The page tells you the file is being sent before you press the button, because that is the honest thing to say about a PDF — Word and text files never leave the device at all.
Does it remember the document between visits?
No, and that is the point. Nothing is stored, so a document opened today is not sitting anywhere tomorrow. Open it again to ask again.
How long can the document be?
Up to about 3.5 MB in one request. Longer reports are best split, or asked about a section at a time.
Do I need an account?
Not for the first few files each week. An account carries a plan when you want more, and signing in is a link in an email — there is no password to lose.
Can it tell me where in the document the answer came from?
Ask it to cite the section or the page and it will. Answers are drawn from the document rather than from what the model already knew, which is what makes that possible.
Can I ask in a different language than the document?
Yes. Ask in English about a German contract and the answer comes back in English, with no translation step in between.
What about a 300-page document?
It reads the whole thing rather than the first few pages. Narrow questions work better than a summary at that length, simply because a useful summary of 300 pages is several pages long.