Initializing secure environment…
Initializing secure environment…
Ask questions about a document and get answers with page references, so you can check each one. You supply your own free Gemini API key, and the raw PDF is never uploaded: the text layer is extracted locally and only the text of the pages you are asking about is sent, in a form you can see before it goes. Answers cite page numbers. No sign-up with us, no subscription, and nothing about your document is retained after you close the tab.
To chat with a PDF using AI free, add the file, paste your own Gemini API key, and ask questions. Answers come back with page references you can verify. The raw PDF file itself is processed locally — only text content is transmitted, and only the pages you asked about, in a form you can see before it goes.
Ask questions that have a locatable answer — 'what is the termination clause', 'what does the appendix say about fees' — and check the page reference. Vague questions produce vague summaries of the whole document, which is less useful than looking at the document. The tool is a search interface with a fluent interface, not an analyst.
Anything requiring arithmetic across pages, anything counting occurrences, and anything where the answer depends on a table's structure rather than its text. A model reading extracted text sees '1,250.00' without seeing that it is a subtotal. For numbers, read the table with the table tools.
Check what the document is before you send chunks of it to a third-party model. The raw file stays here, but the extracted text does leave your machine. For a contract under an NDA, that is a disclosure to a model provider and belongs in your own assessment, not in this tool's marketing.
If the document cannot leave your device in any form, there is no AI chat tool that is right for it, including this one. Extract the text and read it, or use a local model on a machine you control. Being clear about that boundary is more useful than a tool that quietly sends the text anyway.
No. The file is parsed in your browser and the raw PDF is never transmitted. What reaches the model is text extracted from the pages relevant to your question, and the extraction is shown before the request is made. That is a narrower disclosure than uploading the document, and it is the only way this claim can be true.
Because the alternative is a server that holds your document and someone else's key, which is exactly what this site is built not to be. A free Gemini API key costs nothing and takes a minute to create. It stays in your browser's local storage, is sent only to Google, and can be removed with one click.
Not until it has a text layer, because the model reads text and not pixels. Run OCR first, then chat with the result. OCR quality caps the answer quality — a badly recognised page produces confident nonsense, which is the failure mode to watch for.
Good on extraction and summary questions, unreliable on anything requiring arithmetic across the document, and confident when it should be uncertain. That is why every answer carries page references. Treat it as a fast way to locate information, not as an authority on it.
The text sent is handled under your API key and your provider's terms, which is the part you should check for your own compliance requirements. What this site can state precisely is that it has no copy: the request is made from your browser, and this site neither stores nor logs it.
Yes — set a page range, which also keeps the request small and fast. For a 400-page manual, asking about chapters 12 to 14 is more accurate and cheaper than sending the whole thing.
It is deleted along with everything else in local storage. It was never on a server here, so there is nothing to revoke beyond Google's own key management if you want to rotate it.
The tool is free with no account here. The model call is free within Google's own free-tier limits, which are generous for a document of this size. If you exceed them, the limit is Google's and is visible on your key's dashboard.
More ai & smart tools — all free, no upload.