Chat with your PDF
powered by open-source AI.
Upload a PDF, get an instant summary, then ask follow-up questions and get answers straight from the document. Runs on our own self-hosted open-source model — not ChatGPT, not a third-party API.
Drop a PDF here, or click to choose one
Needs selectable text — scanned PDFs should go through OCR first. Up to 15MB, ~60 pages read.
Summary
How to chat with a PDF online
- Drop in your PDF, or click to choose one. The text is read once — the file itself is deleted right after.
- Click Summarize this PDF for an instant overview, or skip straight to asking a question in the chat box below.
- Keep asking follow-up questions — each answer is based only on what's actually in the document, using the same extracted text, so you don't need to re-upload.
- Use the suggested questions for a quick start, or type your own — dates, figures, who's involved, what a clause means, anything the document covers.
Why VitePDF's AI is different
- Runs on our own self-hosted, open-source language model — your document is never sent to OpenAI, Google or any other AI company, and never used to train anything.
- No account, no API key of your own needed — it's built into the tool.
- The same document stays in memory for your whole conversation, so follow-up questions are instant to send — no re-uploading.
- Free — this trades a little speed (the model runs on our own hardware, not a paid third-party API) for not depending on any AI company's pricing or data policy.
Frequently asked questions
Is this powered by ChatGPT or another third-party AI?
No — VitePDF runs its own open-source language model (Qwen2.5), self-hosted on our own servers. Your document is never sent to OpenAI, Google, Anthropic or any other AI company, and it's never used to train any model.
Is my PDF stored or used for anything else?
No — the file is uploaded, its text is read, and the file itself is deleted right after. Only the extracted text stays in memory for the length of your session, to answer follow-up questions without re-uploading.
How accurate is the summary or the answers?
The model answers based only on the text it can read in your PDF, and is instructed to say so if something isn't in the document rather than guess. As with any AI tool, always double-check anything important — especially numbers, dates or legal wording — against the original document.
Does this work with scanned PDFs?
Only if the PDF already has selectable text. A scanned image with no text layer won't work — run it through our OCR PDF tool first, then upload the result here.
Is there a size or length limit?
Yes — up to 15MB per file and roughly the first 60 pages are read (longer documents are summarized/answered from that portion, and we'll tell you if a document was too long to read in full).
Why is it sometimes slow to respond?
The model runs on our own servers rather than a paid third-party API, so responses can take longer, especially for the first request after a quiet period. That trade-off is what keeps this tool free and independent of any AI company's pricing.