Chat with a PDF extracts the text of each page into memory and answers from those words. If a question is not in the document, the reply is that it could not be found. There is no general-knowledge fallback. Page buttons jump to the extracted passage. A browser on-device prompt is used only when it exists and only with the passages already retrieved. A downloadable private model is not published with this version, so the button is not offered. The index is cleared when you close the document. Summaries, dates, and clause hints are searches over the same text, still cited. Scans without a text layer produce little to search, and this version does not OCR them in secret.

Chat with a PDF stays on this device. The download is the copy you keep, and Weesize does not store a second one. Close the tab when you are finished and the bytes are gone.

• No upload: the answer is limited to text read from the file in this tab.
• Speed: the file does not travel to a server and back, so the wait is only the work itself.
• Quality: the original is kept when the new file would be worse or is not actually smaller.