ORCA guide · about 15 minutes, one page you already use
The first workflow teaches you more than the manual does. Do it on a task you would otherwise do by hand today, follow the five steps, and read the log at the end — that log is what makes the second workflow fast.
Open the page you would work on by hand and start the visual element picker. Click one control — a button, an input, a link. ORCA stores that element as a target with a label, not as a coordinate on screen.
Type a value, choose a dropdown option, extract a field, submit. Then make it robust: a retry when the click is refused, a wait for an element to appear, a condition on something being enabled. Two rules here usually matter more than ten actions.
Run it once. Read the log line by line: what was found, what was typed, what had to be retried. A run you have read is a run you can trust.
A target that moved: re-pick it and give the step a wait. A value that came back empty: pick the container instead of the row. Most first workflows need exactly these two repairs.
Save it as a workflow profile. From then on it is one click in the panel, or a recurring run on the built-in scheduler if the work arrives on a timetable.
Five things account for most of them. None of them mean you did it wrong.
“Element not found”The page changed shape, or the element loads late. Re-pick it with the picker and add a wait for that container before the click.
A half-typed valueSome frameworks reject instant input. Add a small dynamic delay before the field and let ORCA retry the step.
A run that stops at a verification gateThat is the guard rail working. ORCA pauses instead of solving a CAPTCHA or a 2FA prompt: sign in yourself, then start the run in that tab. Never script a login — credentials belong in your password manager.
A scheduled run that did not fireThe scheduler fires inside your browser profile, so the browser has to be running and the plan active on that machine. If a run matters, make it idempotent so an extra firing is harmless.
Empty rows from an extractionYou picked the leaf cell rather than the row. Pick the row container, then the fields inside it, and extract per row.
Saved profiles can run on the built-in scheduler, and rule-based runs do not spend your AI balance. Check the permissions page once, then forget about us.
