RPA (Automation)

Build a process once — a sequence of steps, visually, no code — then run it on any profile. It drives the exact same isolated, fingerprinted browser a normal launch would, so it's indistinguishable from you clicking it by hand.

Reference

The full reference is split by topic:

Command referenceEvery step type you can add to a process, with its parameters. Commands are grouped the way they appear in the builder. CSS & XPath selectorsEvery command that touches the page needs to find it first. This is how selectors work, and which ones keep working. Conditions, loops & dataWhat turns a straight list of clicks into something that reacts: branching on what the page shows, repeating over a list, and reshaping values along the way. Google SheetsUse a spreadsheet as the input list and the output log, so one process handles different rows on every run. Third-party tools & profile infoCommands that reach outside the browser — captcha services, OpenAI, email — plus reading the running profile's own details.

Building a process

Open Processes → New Process, name it, and add steps from the command palette one at a time. Steps run top to bottom; if and loop can nest other steps (including more if/loop) inside themselves.

Process settings

Setting	Options	What it does
Group	Any group, or none	Shares the same Groups list as profiles.
On error	Skip (default) / Stop	Skip: a failed step is recorded and the run continues past it — useful when an occasional optional element (a cookie banner) shouldn't abort everything after it. Stop: halts at the first failure, like an unhandled exception would stop a hand-written script.
After task completes	Clear tab / Save tab, and Quit Browser / Keep browser open	Two independent choices for what happens to the browser once the process finishes.

Running & debugging

Click Debug in the process editor's header — a popup lets you pick which profile to run it on, then launches (or reuses an already-running) profile and executes the process against it.

Every run is recorded in that process's Debug logs — the last 20 runs per process, showing the full step tree exactly as it executed, including which nested steps inside an if or loop ran and whether each one succeeded or failed. Screenshots taken mid-run appear inline.

Scheduled runs are headless by default — no window opens, so they stay out of your way. Tick Run visibly on the schedule to give its runs a real window instead. It is worth doing whenever the process is meant to resemble ordinary browsing, because a headless browser is served a different page and cannot scroll smoothly at all. Runs you start yourself from the editor are always visible.

Run logs are local-only. They stay on the machine that ran the process and are never synced to the cloud, even though the process definition itself is.

Import / export

Export a process to JSON to share it, back it up, or version it outside the app. Import accepts pasted JSON with a choice of Append (add to your existing processes) or Replace (overwrite). Since a hand-pasted or scripted import bypasses the editor's own UI limits, the same 10,000-loop cap is enforced when the process actually runs, not just when you build it in the editor.

Signed in, processes sync to your account the same way profiles do — build one on one machine, sign in elsewhere, and it's already there.