Your assistant's own place

Your assistant has a screen of its own rather than a chat, keeps every conversation you have had with it, lets you archive the ones you are done with, and answers a shortcut from anywhere.

Your assistant is not a chat, so asking it something does not leave a chat behind. It has a screen of its own instead, at /omni. The chevron left of the bolt at the top takes you back to Home, or leave it the way you leave any other screen, by going somewhere else in the sidebar.

That screen opens by itself when you send from the question box, so you can watch the answer arrive. To go back to an older one, open it from the list on Home or the assistant's sidebar panel. That panel opens off the assistant's own row in the left sidebar, above Chats. See Finding your way around for the rest of that sidebar.

That row carries a count when your assistant has something for you: a reply that landed after you moved on, or the welcome it sends a new account. The count goes away once you have read what is waiting. Often it is the only thing on screen telling you your assistant has spoken, since its conversations are not chats and so raise nothing in your chat list.

What makes it different from an agent

The question mark at the top of the assistant's sidebar panel, between "Create new agent" and the close button, explains this without leaving the panel. Hover it and a card appears beside the panel; click it to keep the card open, and press Escape or click away to put it back. It says the four things that are easy to miss: it is yours alone and acts as you, completely private and invisible to everyone else unless you approve an individual message or invitation it drafts; it can see across every chat, agent, workspace and file you have access to, and nothing you could not open yourself; anything it would send or change comes back to you as a card to approve first; and it is on every screen, a keyboard shortcut away, whether you want data pulled out, something set in motion, or an explanation of the screen in front of you. That last one carries the shortcut itself, in whichever form your Enter setting has left it.

Your assistant is deliberately not presented like one of your agents. It has no picture and no badge, only the same green bolt that marks it in the sidebar, its name, and a line saying what it is, because it is not a teammate you set up. It is your own reach into everything you can already get to, and it can see exactly what you can see and nothing else.

Conversations with your assistant have no threads. Its answers always arrive in the conversation itself, and if you ask it to reply in a thread it will say threads are not available there and answer in the conversation instead. Threads belong to chats, where the thread panel is; see Finding your way around for how those work.

Finding a conversation you have already had

Every conversation you have had is kept, and there are two ways in: clicking the assistant's sidebar row so its Omni requests panel opens beside it, or the clock icon in the question box on Home, next to the paperclip and the microphone. Once you are on the assistant's own screen there is a third: the conversation's title at the top is itself the switcher, and the chevron beside it drops the same list open. The list drops open under whichever of these you used rather than taking over the screen, so what you were reading stays put and anything you had started typing is still there when you close it. Conversations are listed newest first and titled by what they were about. Hovering one shows the end of it, the last few things said, so you can find the conversation you meant without opening several to check. Opening one picks it straight back up: you can carry on from where it stopped. Leaving the assistant's screen and coming back, or reloading the page, returns you to the conversation you were last reading rather than to the newest one. Opening the list to look through it does not count as reading one, so a glance at your history leaves that unchanged. "New request", at the top of the Omni requests panel and again on the assistant's own screen, takes you to the question box on Home with the cursor already in it. Nothing is created until you actually ask something, so pressing it and changing your mind leaves no empty conversation behind. Asking a new question from the question box always starts a fresh conversation, which is what keeps the history worth reading.

Archiving a conversation

Conversations you are done with can be put away. Right click one in the Omni requests panel and choose Archive: it leaves the list, stops counting towards the unread badge on the assistant's row, and is no longer what that row opens. It is not deleted. An archive button appears at the top of the panel, beside "Create new agent", carrying the number you have put away; it is there only when something is archived. Press it and the panel lists those instead of your live conversations. Right click one there and choose Unarchive to bring it back, or press the button again to go back to the live list. That button is the only way back to an archived conversation, since the Archived Chats view in your account menu holds chats and your assistant's conversations are not chats.

Asking your assistant from anywhere

You do not have to be on Home to ask. Press Cmd+Enter (Ctrl+Enter on Windows) from anywhere and a question box opens out of your assistant's row in the left sidebar, in the same panel that its conversations appear in. It behaves like the Home box: each question starts its own new chat with your assistant, so an ask typed from a dashboard or a settings screen never gets appended to an unrelated conversation.

The shortcut does the same thing everywhere now, including inside a chat. If that panel was already showing something, the question box takes its place while you type, and the panel goes back to what it was showing as soon as you close the box. On a phone, and on any screen with no left sidebar, the box opens over the middle of the screen instead.

If you have set Enter to start a new line rather than send, plain Cmd+Enter is your send key, so the assistant shortcut moves to Cmd+Shift+Enter (Ctrl+Shift+Enter on Windows). That version works whichever way the setting is set, so it is the safe one to learn.

Pressing it while you are reading a chat brings that conversation with it: the box shows the chat's name, and your assistant can read it while it answers. Take the name off if the question is about something else. Inside your assistant's own screen there is nothing for it to open, so it takes you to Home with the caret in the "What's next?" box, ready for the next request. Shift+Enter is still the way to start a new line in that message box.

The box stays open when you send, and a small panel appears under it saying what is happening to the question you just asked: your assistant working, and then its reply once it lands. Click that panel to open the conversation and read the whole answer. Close the box with the x in its top right corner, with Escape, or by clicking away from it. The conversation is waiting in your assistant's history either way, and your assistant's row in the left sidebar shows a count once the reply lands, until you have read it.

The Ask Omni action on a single message opens that same box, with the message quoted inside it. Hover a message and press the bolt, or pick Ask Omni from the menu you get by right clicking it. The question lands the same way the shortcut does: it starts its own new chat with your assistant and carries the chat the message came from with it, so asking about the call you have just finished does not turn up underneath last week's question. Selecting part of a message before you press it quotes only that part.

So on a computer every question to your assistant is in the one place, out of its row in the left sidebar, whatever you were doing when you thought of it. The message box of a chat is for that chat. On a phone there is no left sidebar, so asking about a message switches the chat's own message box into asking your assistant instead, and swiping a message to the left does the same.

When you ask about a quoted message, your assistant can attach a private insight to that message for you to revisit. Only you can see that card. A reply draft is separate and waits for your approval before it is sent to the chat.

You can ask your assistant to remember a preference for future requests, such as how you like meeting summaries structured. Ask what it remembers when you want to check or correct that context. See Memory and the Wiki for how saved memory differs from your conversation history.