How the AI assistant works

The assistant sits in every project behind Search or ask AI… in the header. It reads the documents you are allowed to read, answers with citations you can open, builds tables and documents alongside the conversation, and asks before it changes anything in the project. Knowing those four habits is enough to use it well.

Essential concepts

• Scope. The assistant searches the project's processed documents, and only those you can open yourself, in both the Workspace and the Dataroom. It can therefore draw on internal working files that were never placed or shared with buyers, and for an organisation admin, who is Owner of every project, that means every document. To keep an answer to what buyers see, say so in the question, for example Answer only from the Dataroom or Use only folder 2 Financials. Redactions apply to it as they apply to you. A colleague with less access who reads the same conversation sees Restricted source where a citation points to a document they cannot open.
• Citations. Every fact drawn from a document carries a citation: a page of a PDF or Word file, a cell range of a spreadsheet, a web page if web research was used. Select it to open the source at that spot in the preview panel. Treat an answer without citations as the assistant's own reasoning, not as evidence.
• Conversations are private to you until you share one. Share gives named colleagues or teams read-only access; they can fork it to continue in their own copy. Sharing is refused with Can't share if a recipient lacks access to the project or to a document the conversation cites.
• Artifacts. Tables, charts and documents the assistant produces appear in the Artifacts rail next to the chat, with version history. They can be previewed, edited and exported as PDF or PowerPoint. See Export a questionnaire or an artifact.
• Confirmations. Actions that change something, such as renaming or moving files, creating folders, changing access or creating tasks, pause with AI wants to … and wait for Allow or Deny. Larger changes arrive as a Proposed changes review with the resulting structure, which you Approve or Reject, and a receipt with Revert afterwards. Nothing in the project changes without one of these.
• Agent Memory is a notes page the assistant reads at the start of every conversation: Global notes for your preferences and terminology, Project notes for this deal. It is yours alone. If you switch Use memories on for a questionnaire evaluation, the notes become part of that evaluation's output and visible to anyone who can open it.

Compact panel on the right; Expand to workspace gives it the whole window with a thread list.11.Composer22.Artifacts rail

What it costs

Each message spends credits from your organisation's balance, more for longer documents and deeper research. When the balance runs low a banner offers Top Up; when it is empty the assistant stops with You're out of credits until an admin adds credits. Guests cannot use the assistant at all. See Credits and billing.

Availability and limitations

• The header button appears on project pages only; ⌘K or Ctrl+K opens the same palette, and /chat opens conversations full-screen.
• The assistant sees processed documents. A file still on Processing or on Uploaded – can't be processed is invisible to it.
• Slides and documents are produced as editable HTML and exported to PDF or PowerPoint from the viewer, not generated as files directly.
• It can make mistakes. Check the cited page before an answer goes into a memo; the citation is there for exactly that.

Next actions

• Ask the AI about your documents walks through a first conversation.
• How questionnaires and templates work covers the structured alternative to chat.