New Desktop jobs now carry on outside the browser

Stop writing the guide.
Just do the job once.

Record one honest pass through a task in the apps you already have open. It writes down what you did, asks the few things watching could not tell it, then walks anyone else through the same job, in whatever order of tabs it takes.

Add to your browser See a run that already happened

Chrome and Brave today. Safari and Firefox build from the same source.

A finished run: nine steps, three times round the loop, and the screen the assistant was looking at when it took each one.

How it works

Three steps, and only one of them is yours.

Nobody writes a procedure document. Nobody records a video that goes stale in a month. You do the job the way you were going to do it anyway, and that pass becomes the thing your next hire runs.

1

Do it once, with record on

Open the extension and press record. Then work. Copy a number out of a call log, switch tabs, search for it, fill the form, save. It notes what you touched, what you took, and which row you took it from.

2

Answer a few questions

Watching cannot tell it everything. Which parts change every time? How far back should it look? When has a loop done enough? It asks in plain language, once, and writes your answers into the workflow.

3

Hand it to anyone

They open the same page and the assistant is there. It walks them through step by step, or does the whole thing while they watch. Either way the run is written down as it happens.

What it does

Most automation breaks on Tuesday. This one looks before it clicks.

It reads the screen, not a recording

A recorded selector is a bet that nothing will ever move. It loses the first time a vendor ships a redesign. This one looks at what is genuinely on screen at the moment it acts: the name on the button, what kind of control it is, the row it sits in, and whether it is even reachable.

  • Survives a layout change, because it never memorised the layout
  • Sees inside panels, drawers and embedded frames
  • Notices an overlay is in the way and clears it first
The assistant working inside a live app, with the exact field it found on the page ringed.

It follows the value, not just the clicks

The hard part of a real job is rarely the clicking. It is carrying something from one place to another and not losing it on the way. When you copy a phone number out of a call record and paste it into a search box two tabs over, it records that as one thought: take this from here, use it there. Tomorrow it does the same with tomorrow's number.

  • Knows the column and the row a value came from
  • Tells the difference between a value that changes and one that never does
  • Carries it across applications, not just across pages
A workflow that reads a phone number in one application and searches for it in another, step by step.

It knows when it is finished

Some jobs are a single pass. Others run down a list until the list is done, and the list is a different length every morning. You say so once, in a sentence, and it checks the live screen at the end of every lap rather than counting to a number somebody guessed six weeks ago.

  • Stopping conditions in plain language, checked against what is on screen
  • Every lap counted and kept separately in the record
  • It stops and asks rather than guessing its way through the rest
The second lap of a run, noting the next unhandled row and carrying that name into the sheet.

The record

When it goes wrong, you get to look instead of guess.

Every run keeps a page of its own. Each step, what the assistant could see, which control it chose and why, the screenshot taken at the moment it acted, how many laps a loop took, and every question it had to stop and ask. Nothing is summarised away.

 
 
 
 
 
 
 
 

Questions

The things people ask first.

Does it need our passwords or an API key for every tool?

No. It runs inside your browser, in the session you are already signed into, and acts as you. There is no credential to hand over and no integration to wait for a vendor to build. If a person can do it on the screen, it can do it on the screen.

What happens the day the app we use changes?

The step is written as an intention, not as a path. It says find the call record for this number and open it, so it goes looking for that each time. A moved button, a renamed tab or a new column does not break the job. A step that genuinely no longer exists does, and then it stops and says which one.

What does it do when it gets stuck?

It stops and asks. The question goes on the run page along with the screen it was looking at, so whoever owns the workflow can answer once and have that answer apply from then on. It does not quietly improvise the rest of the job.

Can someone watch it before trusting it?

That is the normal way to start. It walks a person through, one step at a time, and they press continue. When the same workflow has run clean a few times, let it drive. The record looks the same either way.

Does it work outside the browser?

Yes, on macOS. A small runner signs in from your Mac, connects out to the server and takes desktop work, so a job that ends in a native application keeps going instead of stopping at the last tab. Nothing has to be opened up on your network for it.

Which browsers can we install it in?

Chrome and Brave today. Safari is in testing and Firefox builds from the same source, with one codebase behind all of them. Installing is a download and a folder, with no store review in the way.

Onboarding is a recording nobody ever made.

Make it this afternoon, in the time it takes to do the job once.

Add to your browser Open the console