The preview pane

Ask for the quarterly report and its first heading appears in a panel to the right of the conversation, then the first paragraph, then the rest, while you read.

That panel is the preview pane. It is a set of tabs on the right of the chat page, opened by your agent when it has something worth putting in front of you, and it refreshes on its own as the work changes. There is nothing to keep open and nothing to reload: the agent saves, the tab redraws. The tabs belong to the chat, so reload the page or open the same chat on your phone and the same tabs are there, in the same place.

What it can show

A file. Documents, spreadsheets, slide decks, PDFs, images, video, sound recordings, plain text and code all open in place, from whichever workspace they live in. Ask for a change in the chat and watch the same file rewrite itself beside you. Video and sound play in the tab. A web page saved as a file shows with its own styles, scripts and pictures, and its tab has a View and Source switch. A 3D model or CAD file (STL, OBJ, GLB, STEP and several more) opens in a viewer you can turn with the mouse or a finger, with lighting, view and export options on the tab's bar.

A web page. A site your agents are building is published from its workspace and shown running, so a change to a colour or a heading appears without you touching anything. Any other address they can reach opens the same way, and a YouTube or Vimeo address plays in its own player. This is what A website built beside you is made of.

A page you use rather than read. Where a list of options in chat would be clumsy, an agent can build a small page with real buttons, a form, a colour picker or a place to drop files. What you choose there arrives in the conversation as your next message, so the work carries straight on. For a plain choice or a short form it usually uses a card in the thread instead.

A card. An agent can keep a card in a tab rather than in the thread, as it does for a quiz, so the questions stay in view while you talk.

A flow, drawn. A flow opens as a live flowchart with the current step lit up, alongside panels for what triggers it and what it still needs. Edit the flow and the drawing redraws.

A secure browser window. When something needs a human sign-in, a real browser opens in a tab, already under your control: you type your password there, deal with any two-factor code, then press Hand back. If you only asked to watch, Take control hands you the keyboard. The password never passes through the agent. Log in for it once walks it through, and The browser covers the browser itself.

Tabs

Only one tab shows at a time. A strip of tab titles runs along the top, each with its own close button, and the agent gives every tab a short name for what it holds. The dashed + at the end of the strip opens another Files tab, so you can have several files open side by side and switch between them. Opening something that is already open takes you to its tab instead of making a second one.

While any tab is open the sidebar tucks away and the chat narrows beside the tabs. Rest the pointer on the sidebar icon at the top left and the sidebar floats back over the page. The round button at the top right, beside the bell, hides the tabs and brings them back; it reads Hide the tabs, Show the tabs with a count, or Open the files when none are open. Ctrl and \ (Cmd and \ on a Mac) does the same from the keyboard. Drag the seam between chat and tabs to any split you prefer; it is remembered in your browser. Full screen gives the tab the whole window.

Working with a file

You can edit a text file where it sits. Press Edit on the tab's bar and Markdown opens in a rich editor with a switch to its raw text, code opens in a code editor, and plain text in a plain box. Save and Cancel take the place of Edit, and Ctrl or Cmd and S saves. The page does not jump back to the top when you start or finish editing, and if the file changed on disk while you were typing, you are asked before saving over it. The agent is told when you save, so it reads the file again before it builds on it. You can also still ask for the change in chat.

The bar also holds Refresh and Download, and a menu for the rest. Print and Download as PDF are there for Markdown, web pages, pictures, text and code, and Download as PDF also covers Word, Excel and PowerPoint files and their OpenDocument equivalents; a PDF just downloads as it is. The printout and the PDF match. If your installation cannot do a conversion, the item is greyed out with a short reason. Rename, Cut, Copy and Delete are in the same menu. A file that is part of a share also has Sharing options there: see Sharing.

A PDF has Open, which shows it in your browser's own viewer in a new browser tab, and Download. A file with no viewer, such as an archive or an installer, shows a card with its name, kind and size and one Download button.

Closing and coming back

Close a tab whenever it is in the way. The agent's message keeps a link that opens the same thing again. A page published only for a while eventually expires, and its tab closes itself rather than leaving a dead frame. Some sites refuse to be shown inside another page; the tab says so and offers Open in a new tab.

Where it lives

On the chat page. Press Files (the folder icon) in the chat's controls to open your files in a tab, or ask your agent to show you something. On a phone the tabs sit behind a Preview tab beside Chat, appearing once something has opened.

The same tabs open on the Files page: click a file on Manage > Files (/workspaces), or on an agent's Workspace tab, a Memory Space's Files tab or a job's Files tab, and it opens in a tab beside that page. There, a Chat about this file button opens the Help panel on a conversation about it. Those tabs are kept in your browser, not with a chat.

Files shown here are ordinary workspace files, so the same ones are waiting under Manage > Files long after the conversation has moved on. Chat covers the conversation itself, and Make things shows what tends to end up in a tab.