Use case · For everyone

Show your Agent a chore once

Some jobs are easier to show than to explain. Record a demonstration by doing the task once on your own computer, and your Agent learns the steps. It then walks you through what it understood, you correct anything it got wrong, and the result becomes a skill your Agent keeps using.

Start with one Agent More use cases

“Do it the way you normally would, and I’ll ask about any part I didn’t follow.”

Wren

You filed one expense report the long way. Since then, each report has been filled in for you and held for your OK.

Teaching Wren the monthly expense report

Tue 4:00 pm
YouYour MacBook
Tells Wren “Watch me do this,” then files March’s expenses the usual way: the receipts folder, the expense site, three categories, submit.
Tue 4:12 pm
WrenYour MacBook
Plays back the steps in plain words and asks one question: “Why did the taxi go under travel instead of client costs?”
Tue 4:14 pm
YouChat with Wren
Explains the rule, and Wren saves the whole job as a skill.
Apr 30, 9:00 am
WrenYour MacBook
Gathers April’s receipts, fills in the report, and stops before submitting.
Apr 30, 9:03 am
WrenNeeds you
“April’s report is ready: $612.40 across 14 receipts. Submit it?”

How it works

The pieces behind the story

Where it runs

Your computer itself becomes Wren’s home, so Wren learns in the same apps you use.

Who does it

Wren turns what they saw into a written skill, and asks before using it the first time.

What you control

Demonstrations record only when you tell them to. Wren sees the steps you take but not what you type, and you review whatever Wren keeps.

Demonstration recordings stay on your computer, and the skill you approve is the only thing kept.

Set it up

Getting set up

1
Turn on “Use this computer as an Agent” in Geyser Host.
2
In Wren’s settings, set “When to watch” to “When I demonstrate a task.”
3
Tell Wren “Watch me do this,” and do the job once.

Keep exploring

More use cases

Browse all use cases

Make a start

Start with one Agent and one real job

Set up a workspace, give an Agent something useful to do, and see how far it gets.

Start with one Agent See pricing