Ask questions about the document you have open, or across the documents in your DocFila library. Answers are drawn from your own files and name the document they came from.
Open a contract, lease, manual or paper and ask about it. The answer is checked against numbered parts of the document and, when the extracted text has page breaks, names the page.
Ask across the documents in your DocFila library that have extracted text. Library-wide answers name the source document and show the passage. They do not give page numbers, and only the most relevant passages are used, so a question about “every” document can miss some.
The model is told to answer only from the passages it is given. Check important answers against the source document.
DocFila does not train, fine-tune or evaluate any model on your documents. While the AI setting is On, each saved document's text (up to roughly the first 9,000 characters) is sent to Google Vertex AI to build the search index, and to answer, the relevant passages go to Google's Gemini through DocFila's servers (see AI data handling). The stored text passages are encrypted with a key DocFila's servers can derive, so this is not end-to-end encryption, and the search vectors are stored unencrypted.
Vault items marked Encrypted (their key never leaves your device), files that only exist in a local folder on your computer, and scanned pages that have no extracted text.
DocFila indexes your library when you open Ask and again when it changes. For each document only the first 20 passages of about 500 characters are used, so the end of a long document is not searchable here.
Ask needs a signed-in account; it is not available in a guest session. See pricing for current plan and AI-credit limits.
Keep going with the document workflow that matches this page. Open the tool you need, then finish the file in DocFila.