Viktor.comAn AI coworker that actually does the work
Promoted
Wolf PDF
Free PDF tools: merge, split, convert, edit, sign
14 followers
Free PDF tools: merge, split, convert, edit, sign
14 followers
Wolf PDF is a free, privacy-first toolkit for everyday PDF work. Merge, split, compress, convert (Word, Excel, PowerPoint, images), edit, redact, sign, add page numbers, and even ask questions about your PDF with cited answers. No watermarks, no forced sign-ups, and files are auto-deleted after 30 minutes
How does the AI question feature handle scanned PDFs or documents that are mostly images, does the OCR happen on your servers or locally?
@kbra84zm At the moment OCR for Ask PDF is not automatic. Ask PDF first tries to extract readable text from the PDF on our backend. If the document is mostly images or scanned pages, it asks the user to run OCR first.
The OCR itself is server-side in Wolf PDF’s conversion tools, not something happening in the browser. A smoother “OCR then Ask” flow is on my roadmap because it would make scanned documents much easier to query.
@kbra84zm Quick update on this: scanned PDFs now work in Ask PDF. When you upload a document with no readable text layer, it detects that automatically and offers one-click text recognition before you ask anything. The OCR runs server-side, same as our conversion tools, and files are still auto-deleted after your session. Would love for you to try it and tell me if the flow feels right.
Curious how the AI question-answering feature handles scanned or image-only PDFs - does it run OCR first or just fall back to "can't find that" if there's no text layer?
@sezerklaiptebx Right now Ask PDF works best with PDFs that already have a readable text layer. If the PDF is scanned or image-only and there is no extractable text, it will return a “run OCR first” style message rather than pretending it can read the image.
Wolf PDF does have OCR/conversion capability elsewhere, especially in the PDF-to-Word OCR flow, but automatic OCR-before-Ask is not yet wired into Ask PDF. That is a very sensible next improvement.
@sezerklaiptebx Following up on this one: it no longer dead-ends. Ask PDF now detects scanned or image-only PDFs on upload and offers one-click OCR before you ask a question, then answers with page citations as normal. Your comment was part of why this moved up the roadmap, so if you get a chance to test it I'd really value your take.
How does the AI question feature actually pull cited answers from a PDF, and does it work on scanned/image-only documents or just ones with selectable text?
@dnegayretztem It works best on PDFs that already have selectable text. We pull answers from the document text and include page references where we can. For scanned or image-only PDFs, Ask PDF doesn’t automatically run OCR yet, so the best flow today is to OCR the file first and then ask questions on the readable version. I do think automatically offering OCR before Ask PDF would be useful though, so that’s something I’ll look at.
@dnegayretztem Update: the "OCR first, then ask" flow I mentioned is now live. Ask PDF spots scanned or image-only documents automatically and offers one-click text recognition, then answers with page citations from the recognised text. If you try it on a scanned doc I'd love to hear how the citations hold up.
How does the "ask questions about your PDF" feature actually work behind the scenes, like is it running locally or sending text to an external model?
@yaz2013053 Great question. Ask PDF currently runs server-side rather than fully locally. We extract the readable text layer from the uploaded PDF, choose the most relevant page context, and send that text plus your question to our configured AI provider. In production that is currently.
We don’t expose API keys in the browser, and uploaded files are handled as temporary processing files rather than long-term training data. The feature is designed for document Q&A with citations, not as a local-only/offline AI system.
Curious how the AI question feature works under the hood — is it running locally on my machine or sending the document off to a server, and if it's the latter, what's actually happening during that 30 minute window before deletion?
@sefa3n8y Great question. It isn’t fully local at the moment. For Ask PDF, the file is uploaded temporarily so the server can extract the readable text and answer against the relevant pages. It stays available during that active session so you can ask follow-up questions, but it isn’t saved as a document library. When the session ends we try to clean it up, and the automatic cleanup catches abandoned sessions normally within 30 minutes.
Curious how the AI question-answering feature actually works in practice. Is it pulling answers only from the uploaded PDF, or does it pull from external sources too?
@zeyneplteklvlu Thanks dor the question. Right now it answers from the uploaded PDF rather than using external sources. We pull out the readable text, find the most relevant pages, and use that context to answer your question. I do think an optional external retrieval mode could be useful in future, but I’d want that to be something the user clearly switches on, so it’s obvious when the answer is based only on the PDF versus when extra context has been brought in.