Paste a contract, report, or research paper on the left and ask questions about it on the right. Answers come only from the text you provided, if the document doesn't say, it tells you that instead of guessing. The conversation remembers what you already asked, so you can say "what about the second one?" and it knows what you mean. Nothing is stored.
Your document is not stored. It stays in this browser tab for the session and is sent to the AI engine only when you ask a question. Close the tab and it is gone.
Yes. The last few question-and-answer turns are sent along with each new question, so follow-ups like "and the second clause?" resolve correctly. That memory lasts for the tab session and is never stored anywhere.
It is instructed not to. If the answer isn't in the text you pasted, it should tell you what's missing rather than fill the gap from general knowledge. Verify anything that matters against the document itself.
Not directly, the file loader takes plain text formats (.txt, .md, .csv, .json). For a PDF or Word document, open it, copy the text, and paste it in. Adding real PDF parsing would mean shipping a parser to your browser, which we would rather not do quietly.
Up to 20,000 characters, roughly 3,000 words or 8 pages. The counter under the box turns bold when you go over, and the Ask button disables until you trim it, so you never get an answer based on a silently truncated document.
No. It stays in your browser tab and is sent to the AI provider only at the moment you ask a question. We store nothing, close the tab and it is gone.