ACTIONIST · LARGE ACTION MODEL
It doesn’t describe work.
It does it.
Actionist sees the screen, moves the cursor, and finishes the task — like a person. No API, no script, no integration.
One real task
▸Book the cheapest flight to Tokyo next Friday.
Read the screen
Opened the booking site
Searched Tokyo · next Fri
Compared 14 fares
Held the $612 fare
● Done · 38s
A real task, start to finish. It operates the interface like you do.
01
Sees
Takes a screenshot and reads the interface the way a person does. No DOM, no API, no brittle selectors.
02
Acts
Moves the cursor, types, clicks, navigates. It drives real software to finish real work.
03
Learns
Every completed task becomes a route on the Action Tree. The next agent reads it instead of relearning it.
Every task it completes makes the next one faster.
Give it a task. Watch it act.
Open Actionist, describe the work, and let the agent take the screen.