Guide via Screenshots
A last-resort way for a habit to finish a step by looking at a picture of the app, when Plainwork's precise on-device methods can't.
last updated
Plainwork normally drives apps precisely and entirely on your device. Some apps don't cooperate — a list that only exists while you scroll it, a window drawn as one big picture, a remote desktop. Guide via Screenshots is what a habit falls back to there: Plainwork takes a picture of that one app's window, sends it to its cloud AI model, and lets the model click and type the way you would by hand.
It is off by default, and it is always the last thing tried — never the first.
Turn it on
- Open Settings → Advanced.
- Switch on Guide via screenshots.
- Grant Screen Recording if you're asked. Nothing can be photographed without it.
What it does during a run
A blue ring follows the point Plainwork is about to click, so you can see where its attention is. It works one small action at a time, taking a fresh picture after each one, until the step is done or it gives up.
If the rest of a habit has stopped matching what's on screen, Plainwork can hand the whole remainder over at once instead of one step at a time.
Anything it can't undo stops and asks
Before sending, posting, submitting, deleting, or paying, the run pauses and shows you the actual picture it is about to act on, plus a plain description of what happens next. Nothing fires until you approve it. Decline and the run stops there — every later action is refused, not just that one.
The exception is a habit you've set to run without approval, which is the point of that setting.
What gets sent
Only the window of the app being automated: never your whole screen, never a second monitor, never a blocked app. Pictures are used in the moment and never stored. This is the one part of Plainwork that lets the model act by sight, which is why it is an explicit opt-in.