This is a language model answering from your document — it can misread or misphrase things.It only reads the text you uploaded, is told to say when the document doesn't contain an answer instead of guessing, and remembers earlier questions in this conversation so a follow-up doesn't need to repeat context. It is not a substitute for reading the document yourself when something matters. The AI runs on our own hardware; your file is never sent to a third-party AI API. Nothing about this conversation is saved on our side — refresh the page and it's gone.
By ShinobiTools Team · Last updated: August 2026
A contract with a clause you need explained, a report you'd rather ask about than read cover to cover, a manual where you just need one answer — upload the PDF and ask it directly. Ask a follow-up and it remembers what you asked before.
Every question you ask is sent to the AI together with the document's text and the earlier questions and answers from this same conversation — that's how a follow-up like "and the second one?" gets answered correctly. None of that is saved on our side: your browser tab holds the conversation, and the file is sent along with each question rather than stored on a server. Close the tab or refresh the page and the conversation is gone, by design, not by accident.
A scanned page is a picture, not text — there is nothing for the AI to read until it has a text layer. This tool detects that case and points you to Image to Text instead of guessing at an answer it can't actually see — that tool reads a scanned PDF with OCR and gives you the words back directly, since we can't yet add a text layer to a PDF you already have. If your document already has a text layer (a report, an exported document, a contract), this works directly.
Your PDF is processed entirely in memory on our own machine and handed back in the response — it is never written to our disk for this tool, and nothing is kept afterwards. The extracted text is sent to a language model running on our own hardware, not to an outside AI provider.
Want one written summary instead of a back-and-forth? Summarize PDF reads the same document and writes a short, medium or long summary. Need the structure — headings, lists — kept intact instead of turned into an answer? PDF to Markdown does that. Handed a scan with no text layer at all? Image to Text pulls the words out with OCR — we can't yet add a text layer to a PDF you already have.
Upload the PDF, then type a question about it and press Ask. Ask a follow-up question and it remembers what you asked before, so you can say things like "and what about the second one" without repeating context.
Yes to both. A language model reads your document and answers in its own words. It is told to say so plainly when the document doesn't contain an answer rather than guess, but it can still misread something or phrase an answer imprecisely. For anything you'll rely on, check the actual document.
Only if it already has a text layer. A scanned page is just a picture, and there is no text for the AI to read — and we can't yet add a text layer to a PDF you already have. Image to Text reads a scanned PDF with OCR and gives you the words back as copyable text instead.
No. The AI model runs on our own hardware. Your file is not sent to OpenAI, Google or any other outside AI provider.
Nowhere on our side. The conversation only exists in this browser tab; refreshing the page clears it. Nothing is saved to a database, and your file is only ever handled in memory here, never written to our disk.
25 pages per file. There is no account, no watermark and no daily cap.