Some of your work only exists in a browser.
The vendor portal with no API. The council site that will never have one. Record the task once and Claritty does it there, the way you did.
The browser extension is coming soon. Everything on this page is what it does; until it is out, describe the task and Claritty builds the same automation.
One recording, as stored
A recording keeps the name of the control you touched, never the value you put into it. The table’s column names are kept, because the automation has to know which column it is reading. No cell of it is.
You do the task. It writes down the steps.
A pill sits in the corner of the page. Start it, do the job the way you always do, and stop. It follows you across tabs and hosts, and the feed shows you what it is keeping as you go.
It asks at most one question, at the moment a step is ambiguous, while you still remember why you did it. Your answer is attached to that step.
vendorportal.com
Invoices
Blurred here for the same reason they are not in the recording.
Read the overdue invoices
vendorportal.com
Escalate it when it is more than 30 days late
the finance channel
Send the reminder when it is 30 days or less
the vendor on that row
Ready to send 2 of 3 invoices.
One escalation, one reminder. The paid one is left alone.
That recording is now an automation.
The steps you took, the rule you said out loud, and the cadence you described. You read it back before it exists, and you can change any part of it in plain words.
Then it runs on its own. Each run reads first and shows you what it is about to send, and the paid invoice it decided to leave alone is as much a part of the job as the two it acted on.
You record it once
A browser extension follows the task across tabs and hosts.
You check the steps
Every step, the sites it touches, the accounts it needs. Nothing is created until you say so.
It runs somewhere else
On a schedule, in a Claritty-controlled browser. Not the one on your desk.
It does not run on your machine.
You connect the site once. That session is encrypted, and it is loaded into a Claritty-controlled browser at the moment of the run. Never your own Chrome, and the runner keeps nothing when it is done.
It refuses to open an address on a private network, so a task pointed at your own infrastructure stops rather than proceeds.
A run that has not been approved sends nothing.
Every run happens twice. The first time it reads, works out exactly what it would send, and sends none of it. You see the messages before they exist. The second time it does what you approved and nothing else.
It reads, and stages every message it would send.
You approve, in the app or from Slack.
It performs exactly what was staged, once.
If it could only do part of the job, it says so rather than reporting success.
How we know any of this is true.
An automation that does nothing and reports success is the failure that matters here, because it looks exactly like one that worked. So the recorder, the model, the compiler and the runtime are run together against recorded tasks, and the result is compared against a written statement of what that automation has to do.
Real parts, not stand-ins
The browser extension that ships, the model that ships, the compiler that ships, and the engine a deployed automation really runs on.
Judged on what it did, not what it looks like
Over a known set of invoices, the test asserts which vendors were emailed and at which address. An automation that emails nobody cannot pass by having the right shape.
The data-blind rule is re-checked every run
The password, the address typed into the form, and every cell of the table are searched for in the recording and in the finished automation.
When it is wrong, it says which part was wrong
The recorder, the prompt, the model, the repair passes, the compiler and the runtime are separable, and a failure is attributed to one of them rather than reported as a whole.
This is also why there is no accuracy figure on this page. The suite covers a handful of recorded tasks rather than a representative sample, and a number from it would say more than it knows.
Show it the thing you keep doing.
The extension is coming soon. Describe the task today and Claritty builds the same automation.
Describe it insteadFree to start, no credit card. Questions? Talk to us