Guided actions
When you ask the assistant to do something on a page ("create a new folder", "fill in this form", "open Online Services"), it can answer with a guided action plan: a short, ordered list of steps it walks you through right on the page. Each step highlights one element and tells you exactly what to do, in order.
What a plan looks like
A plan is a list of steps. Every step is a single action on a single thing:
- Click a button, link, or menu entry.
- Type a value into a field.
- Choose an option from a dropdown.
- Hover to reveal a hidden menu (the assistant asks you to hover, because browsers only show hover menus for a real pointer).
The steps run in order. The assistant shows a moving highlight (a "comet") that lands on the element for the current step, so you always know where to look next.
You stay in control
- Nothing happens until you approve. The plan is shown first; you choose how the assistant runs it: Do it for me (quick, minimal delay), Walk me through (it acts while narrating briefly), or Teach me (it explains each step before doing it). In every mode the assistant performs the steps for you; the modes differ only in pace and how much it explains, not in who acts. The one thing it cannot do for you is hover to reveal a hidden menu (browsers only show those for a real pointer): there it points at the element and waits for you to hover.
- The plan never starts over. If you switch tabs, scroll, or click ahead, the assistant catches up to where you actually are and continues from there. It will not redo a step you already completed, and it never silently undoes a navigation or a save.
- The plan follows you backwards too. If you wander off mid-plan (say you close the menu the tour just opened by clicking somewhere else), the assistant notices within a few seconds and walks back with you: the highlight returns to the step that reopens what was closed (the "+ New" button, not the menu entry that is no longer on screen). It only ever steps back through things that are safe to redo, like opening a menu; it never re-runs a save or a navigation.
- Steps that reveal other steps are handled for you. Some buttons only appear after you open a menu or hover over something first. The assistant includes that opening step as an ordinary earlier step in the plan, so the sequence always flows from what is visible right now to what you are trying to reach.
- You can act ahead at any time. The assistant performs the steps for you, but if you click the highlighted element yourself before it does, that counts and the plan simply moves on. You never have to dismiss an explanation bubble first.
You always see the next step coming
The assistant never fires an action out of nowhere. Before each step it shows a short, visible countdown of a few seconds, for example "Next step in 6s", together with the name of the step about to run. The countdown means the same thing in every mode: the assistant will do this step in N seconds (then it clicks, types, or selects for you). The countdown lives on the assistant itself, the small progress pill when it is collapsed, or the bottom of the panel when it is open, never buried in the chat. You can:
- Go now to run the step immediately, if you do not want to wait.
- Hover the countdown to freeze it while you catch up; it resumes when you move away.
- Pause to stop the whole tour where it is; resuming continues from the same point.
- Cancel to end the plan.
The same surface works on every site, whether you use the browser extension or the assistant embedded in the page.
When you pick the Teach me pace, the explanation for each step appears right next to the element it is about to use, with the countdown beside it, so you can read while it counts down. That explanation is the description the app's team wrote for that element, shown in your language; a step whose element has no written description simply runs with its highlight and countdown, without an explanatory note. When the countdown ends the assistant performs the step for you, exactly as in the other modes; Teach me simply gives you a longer countdown and an explanation each time. Go now runs it immediately, Need more time (or hovering the explanation) pauses the countdown (the button then makes way for a quiet "Auto-advance paused" note), and Got it runs the step whenever you are ready. (If the element is tucked inside a menu that is still closed, the explanation shows at the bottom of the panel instead, so you always see it.)
Why a step sometimes waits
A step can pause and wait when the thing it needs is not on screen yet, for example a menu that has not opened or a tab you have not switched to. The assistant waits for that element to appear (or asks you to perform the gesture that reveals it) and then continues. A pause is not an error; it is the plan keeping in step with the page.
If what the step needs cannot appear anymore (the menu it lived in was closed), the assistant does not keep you waiting: it steps back to the action that brings it back (see "The plan follows you backwards too" above). The tour only pauses and asks you to press Retry when it genuinely cannot find a way forward or back, and the pause message always names the exact element involved so you know what to look for.
A whole plan can also wait when it belongs to a different page than the one you are looking at. In that case it does not start at all: the plan card tells you where to continue ("This plan is waiting for you on Files"), includes the assistant's directions for getting there when it knows the way, and wakes up on its own as soon as you arrive.
Live page mode (sites without curated content)
Most of the web has no setup for Appilot yet. On those sites you are in Live page mode: the assistant reads the page in real time to answer your questions and guide you, instead of relying on content someone prepared in advance. You will see a short note the first time you open the assistant on a new site, and the page chip reads "Live page mode" rather than naming a known area.
Guided steps still work here. When you ask the assistant to walk you through something, it looks at the page, finds the buttons and fields involved, and builds the steps from what it sees. Those steps are marked "From this page" so you can tell them apart from steps an app's team has verified in advance. Because the assistant inferred these steps from the live page rather than from prepared instructions, they always run one at a time, with your confirmation: the assistant points at the element and you perform the action. There is no "run everything automatically" option for inferred steps, which keeps you in control on a page nobody configured.
If you would like a site you use to get first-class support, open the assistant on it and choose Request coverage for this site. We tally the requests, so the more people ask, the sooner a site rises up the list.
You can also pause Appilot on a specific site from the assistant's settings if you would rather it not appear there. To bring it back, right-click the page and choose Resume Appilot on this site. What the assistant reads from any page is privacy-masked before it reaches the AI provider; see Privacy & Security for how that works.
For configurators
If you administer an app, you can make a control easier for the assistant to reach by giving it prerequisites in the Backoffice control editor: the ordered prior steps (a hover that opens a drawer, a click that opens a menu) that reveal it. Most controls need none. Add prerequisites only when a control is hidden until another action is taken and the assistant could not find it on the live page on its own. See Managing controls for the authoring guide.
You also control what the Teach me pace says about each step: it reads the control's description. Write that description for the person clicking through (plain language, no internal names), translate it for the languages you serve, and the on-page explanation appears in the user's language automatically. Leave a control's description empty and its Teach step runs with no explanation, so fill in the descriptions for the controls your plans walk people through.