Asking the AI about a document
Last updated:
Translated from the Italian guide. Narciso itself is in Italian, so the names of buttons, menus and fields are shown exactly as they appear on screen, with a translation in brackets where it helps.
When a client or a supplier sends a long document, a quote, a price list, a contract, you can ask Narciso's AI assistant to answer questions about its content without reading all of it: «Qual è il totale?» (What is the total?), «Entro quando va pagato?» (By when must it be paid?), «Che condizioni di consegna indica?» (What delivery terms does it state?).
Contents
Opening the assistant#
- Open the attachment with a click (Attachment viewer).
- At the top left, among the buttons, press Chiedi all'AI (Ask the AI) (the button only appears for PDFs, Word documents, presentations and spreadsheets; not for photos, videos and audio).
- On the right, the MEDIA EXPLORER grid gives way to the chat with the assistant. Pressing the button again takes you back to the grid.
If you change attachment with the arrows or with the thumbnails, the chat closes and, if there were already questions for the new document, they reappear.
The document must be ready#
The assistant only answers when the document has been read by the program. You can tell from the first message:
| Assistant message | Meaning |
|---|---|
| «Sto leggendo il tuo documento...» (I am reading your document...) | The document is still being prepared: the writing field is locked. The program checks again every 3 seconds for about ten minutes and unlocks by itself |
| «Ciao! Sono l'assistente AI di Narciso. Cosa vuoi sapere su questo documento?» (Hello! I am Narciso's AI assistant. What do you want to know about this document?) | Ready: you can write |
| «Non sono riuscito a leggere questo documento.» (I could not read this document.) | The preparation failed: it cannot be queried. The original file can still be downloaded (Attachment viewer) |
Asking a question#
- Write in the «Chiedi qualcosa sul documento...» (Ask something about the document...) field.
- Press Invio (Enter) or the button with the paper plane. The question appears on the right, on an orange background.
- While the assistant is working you read «Analisi del documento in corso...» (Analysing the document...) for the first question and «Sto formulando la risposta...» (I am putting together the answer...) for the following ones. In the meantime you cannot write.
- The answer appears on the left, in a grey bubble, in plain text (no bold or formatted lists).
Later questions take the earlier ones into account: you can say «e la seconda?» (and the second one?) or «spiegamelo meglio» (explain it better) as in a conversation.
A question can be at most 2000 characters long. The program uses the last 20 questions and answers as the memory of the conversation.
How it answers#
- It relies only on the content of the document: if the information is not there, it says so instead of making it up.
- For documents of normal size (up to about 400 thousand characters of text) it reads the whole document. For longer documents it reads the extracts closest to the question: the answer may therefore say that a piece of data is not there even though it is in another part. Try again by rewording the question with words that appear in the place you are looking for.
- For very long documents, if the search index is not ready yet (usually right after the document arrives), «Il documento è molto grande: lo sto preparando per le domande. Riprova tra qualche minuto.» (The document is very large: I am preparing it for questions. Try again in a few minutes.) may appear.
The sources#
Under the answer the line Fonti: (Sources:) may appear with small buttons Fonte 1, Fonte 2... (Source 1, Source 2...) (at most 6). Each one corresponds to a passage of the document the answer is based on; hold the mouse over it to read the passage. The program checks that each passage is really copied from the document and discards those that are not.
When you click a source, the viewer looks for the passage in the PDF, scrolls to it and highlights it. It works in documents shown in pages (PDFs, and converted Word and PowerPoint files), not in spreadsheets. Possible warnings:
| Warning | What it means |
|---|---|
| «La fonte si può evidenziare solo nei documenti PDF già aperti.» (The source can only be highlighted in PDF documents that are already open.) | The document is not shown in pages (for example it is a spreadsheet) or it has not loaded yet |
| «Non ho trovato questa frase nel testo del PDF: potrebbe essere divisa tra due pagine o scritta in modo diverso nel file.» (I did not find this sentence in the text of the PDF: it may be split across two pages or written differently in the file.) | The passage is not found word for word: read it by hand |
| «Non sono riuscito a cercare la fonte nel documento.» (I could not search for the source in the document.) | The search in the PDF did not work; try again |
The conversation stays saved#
Questions and answers are saved: if you close the window, change page or reload, when you reopen the same attachment you find the conversation again. You see at most the last 200 entries.
The conversation belongs to that document, on that channel, in that location: colleagues of the same location who open the same attachment see it too, and what you ask stays there for everyone. There is no button to delete it.
Limits and privacy#
- 30 questions every 10 minutes for the whole business: beyond that, the assistant answers «Hai superato il limite di richieste. Attendi un momento.» (You have exceeded the request limit. Wait a moment.) and you try again after a few minutes.
- To answer, the text of the document is sent to an external artificial intelligence service (External services). If the document contains personal or confidential data, consider first whether it is appropriate to use it.
- The assistant can make mistakes: for important figures, dates and clauses, check the source in the document.
- There is no restriction by role: Titolare, Responsabile and Operatore can use it.
What to do if...#
| Situation | What to do |
|---|---|
| The field is grey and says «Sto leggendo il tuo documento...» | Wait: it unlocks by itself as soon as the document is ready |
| «Assistente AI temporaneamente non disponibile» (AI assistant temporarily unavailable) appears | The artificial intelligence service is not responding: try again shortly |
| «Sto ancora leggendo il documento: riprova tra poco.» (I am still reading the document: try again shortly.) | The preparation is not finished; try again after a few moments |
| «Non sono riuscito a leggere questo documento: l'elaborazione è fallita.» (I could not read this document: processing failed.) | The document could not be converted: read the original by downloading it |
| It answers that a piece of information is not there, but it is | The document is long and it only read some extracts: reword the question with the words of the place you are looking for |
| The Chiedi all'AI button is not there | The attachment is a photo, a video or an audio file |