Works inside anything with a screen
NetSuite, GoHighLevel, Zoho, CallTools, ReadyMode, or a system with no name left on the login page — none of them need to expose an API, because Operator never asks for one.
Lia Operator
Does the work. Then stops and asks.
The moment
The same twelve clicks, for the ninth time today, in a system that was never going to change for you.
Operator works at the screen level — it clicks, types and navigates the way a person does. That means it works inside software that has no API, no integration and no intention of ever offering one. If a person can do it on screen, it can be handed over.
One thing worth being precise about, because it trips people up: Operator itself runs on an AI model, and that needs its own key and its own budget — that is the engine. What it does NOT need is an API from the software you are pointing it at. NetSuite, GoHighLevel, Zoho, CallTools, ReadyMode, that twenty-year-old system nobody supports: none of them need to expose anything.
It runs as a queue you control. Add a task, run one or run the lot, pause mid-run, resume, or veto a single action a moment before it happens. There is a stop that works from anywhere.
And it does not finish the job. Deliberately. It will never click Send, Submit, Post, Publish, Pay or Confirm Order — it prepares everything and hands it back for a human to approve. That is enforced in the code, not promised in a brochure.
What it actually does
Runs the same clicks so you don’t have to
Point it at the screen you use every day and it repeats the exact steps, every time, without getting bored on click eleven.
Pauses before every action
A real veto window before each step, with a kill switch that works from anywhere in the app.
Never sends anything
Send, Submit, Post, Publish, Pay, Confirm Order — all refused, by design, every time. You finish it.
Stays out of the screens that can hurt you
Banking, payment, and password-manager screens are recognized and avoided through a blocklist that’s actively maintained — not a promise, a running list we add to the moment a new one is flagged.
Logs every step
Every run and every action is recorded and reviewable afterwards. Nothing happens that you can’t go back and read.
What it doesn’t do
- It is not autonomous and is not trying to be. Every outbound action ends with a human.
- It is still weaker in desktop applications than in a browser. Said plainly because you’d find out anyway.
- The dangerous-screen blocklist is maintained, not exhaustive. If it ever misses one, tell us and it gets added — that is a real gap, not a hidden one.
Every product page has this section. If a company won’t tell you where its software stops, that’s the tell.