AI Chat

Press "AI Chat" in the vertical bar at the right edge of the screen to open the chat on the right; press it again to close it. While the AI is working, the icon shows it even when the chat is closed.

The header reads "AI CEO". The dot next to it shows the connection: green is connected, amber means the AI's authentication has expired, red means disconnected. If authentication has expired, ask an administrator to re-authenticate.

Use "Width" to switch between "Side" and "Wide".

The chat has three views.

• Chat — the normal view. Type in the box at the bottom and press "Send". To cut an answer short, press "Stop".
• Menu — open it with the button left of the header. It lists recent conversations ("History"), the full list of "Smart Actions", and "Clear".
• History — open it with "Show all" in the menu. It lists "Conversations in this context"; pick one to go back to it.

"New conversation" above the input box starts a fresh conversation. "To Knowledge" harvests knowledge candidates from the conversation. They land in the "Knowledge Inbox", where a person decides whether to adopt them.

The short buttons marked "↳" just above the input box are Smart Actions. They change with the app and screen you are on and with the object you are talking about. Up to three show at first; "Show all" lists the rest.

Examples:

• In Knowledge: "Knowledge health", "Check gaps"
• In Marketing: "Marketing overview", "Brief status", "Compose a send", "Messages this month"
• In Sales: "Opportunities", "Next action"; in Support: "Check cases", "Check SLA"
• With a page open in the Page Designer: "Rewrite a section", "Tidy chapter titles", "Draft the page", "Review the copy" and more

When an object is open on screen, the empty chat invites you to "Ask anything about …" that object, and the Smart Actions are about it.

Pressing a Smart Action does not run it right away. It goes into the input box, so you can add your own instruction before you press "Send".

When the AI wants to change something, an action card appears below its reply. The AI's message above the card explains what will change.

• Approve — run it as proposed.
• Reject — do not run it.
• Skip — move on without deciding now.

When you asked through a Smart Action, the card asks "Run “…”?" and the button reads "Run". When a workflow is about to start, it asks "Run the “…” workflow?".

Changes that are hard to undo are confirmed in two steps. A card that sends a message outside shows the number of recipients, the message and this month's message count, and you confirm the count ("Yes, send to N") before it goes out. A card you have decided on shows "Decided".

Typing "yes" in the chat does not run anything by itself. If something you wrote sounds like an approval or a rejection, the chat says "Nothing has been executed." and asks you to decide with the buttons.

Tick "Save this conversation to Knowledge" on a card before you approve or reject, and the conversation also becomes a knowledge candidate.

The AI can only use the same registered operations people use, and it works inside the permissions of the person who asked.

• Reading and internal writes happen under the permissions of the person who asked.
• Sending outward, publishing, moving money and deleting need a person's approval or a mandate a person granted. If the AI tries one of these without a mandate, the work stops, is recorded, and a card asks "The AI asked to run … — approve?". Approving issues a one-time ticket valid for 15 minutes and runs exactly the recorded input under your own permissions. Rejecting records the refusal.
• The AI cannot issue credentials or permissions, and cannot make decisions such as adopt or reject on its own.

If approving the same kind of action every time gets tedious, "Widen the mandate" on the card opens the "AI delegation" screen, where you can hand it over with a scope and a budget.

Below the chat, your remaining AI credits are shown as "N / M credits left" with a ring gauge. "Manage AI providers" next to it takes you to the settings.

When the chat runs on your own AI credential, no credits are used, and the chat says "Running on your own AI credential — credits are not being consumed." If the balance cannot be read, no number is shown.

Conversations are kept as sessions. You can return to them from "History" in the menu, and they are also listed on the "Sessions" tab of the panel at the bottom of the screen. There, each conversation shows whether it is "Active", "Dormant" or "Archived", with a "Back to where it started" button. Conversations of other members are marked "Another member".

When you open a different object on screen, the chat switches to the conversation about that object.

Besides "AI Chat", the bar at the right edge has two more buttons. Only one of them is open on the right at a time.

• Inspector — shows the attributes of the object you have open. It is read-only; edit the object on its own app screen. With nothing selected, it says "Select an object to see its attributes here."
• Editor — appears only where a screen has an editing panel. In the Page Designer it opens as the "Page editor", where you change the look and the text of the part you selected.