Chat with documents and see every step behind the answer
ChatDocument does one thing strictly: it answers from the documents you give it and from nothing else. Here is what happens between the moment a file arrives and the moment an answer with page citations appears on your screen.
Add documents by upload or by connecting a drive
Drop files into a space, add a web page by its URL, or connect a folder in Google Drive, SharePoint and OneDrive, or Dropbox. Connected folders sync on a schedule, so new and changed files reach the space without anyone uploading them again.
- PDF, Word (DOCX), Excel (XLSX, CSV), PowerPoint (PPTX), plain text and Markdown
- Each space keeps its own documents and its own members
- Only the folders you select are synced
Reading every page and keeping its number
Text is read page by page, and each passage is stored with the document name and the page it sits on. Slides keep their slide number and spreadsheets keep the sheet and cell range. Scanned PDFs without a text layer go through OCR first, and the page numbers of the original file stay intact. A new page counts once per billing period, and a scanned page counts as 3.
Finding the passages that matter for your question
When you ask, ChatDocument searches only the documents in the space you selected. It ranks passages by how closely their wording and meaning match the question and passes the strongest ones on. Documents in other spaces are never searched, even when you have access to them.
Writing the answer only from those passages
The AI model receives your question and the retrieved passages, with the instruction to answer from them alone and to tag every statement with its source. Nothing from general knowledge is allowed to fill a gap. You choose the style: a short answer, an answer with quotes, or a bullet summary.
Checking every citation before you see it
Before the answer appears, code compares each citation against the indexed document. The cited page must exist and must contain the quoted text. A sentence whose citation fails is removed rather than shown. Then each remaining statement is displayed with a chip such as p. 4, and clicking it opens the document at the highlighted passage. The cited answers page shows this view in detail.
When the passages do not contain the answer
If nothing in the retrieved passages answers the question, or no sentence survives the check, ChatDocument tells you plainly that the answer is not in your documents. Below that message it lists the closest passages it read, each with its page, so you can see where it looked and decide whether a document is missing from the space.
Asking follow up questions in the same thread
A thread keeps the context of earlier questions, so you can ask "and what about the renewal term" without repeating which contract you mean. Each follow up goes through the same retrieval and citation check, and each one counts as a question.
What ChatDocument does not do
Knowing the limits is part of trusting the answers.
- It does not answer from the internet or from general knowledge.
- It does not read documents outside the selected space.
- It does not show a citation that failed the check.
- It does not give legal, financial or medical advice; it tells you what your documents say and where.
- It does not train AI models on your files.
FAQ
How long does indexing take?
Most documents are ready within moments of upload. Long files and scans that need OCR take longer, and you can ask as soon as a document shows as ready.
Can I ask across many documents at once?
Yes. A question in a space searches every document in it. To work across documents side by side, use compare documents for two to five files or extract to table for many.
Does a follow up question cost the same as the first?
Yes. Every cited answer is one question, whether it starts a thread or continues one.
See the whole process on one PDF
Drop a file, ask a question and click the page chip to check the source yourself.