Artificial intelligence computer use connects a model to screenshots, accessibility information, or interface state and lets it issue mouse, keyboard, and navigation actions. The model can operate applications that lack a purpose-built application programming interface, including websites and desktop tools.
Graphical interfaces are ambiguous and can change between observation and action. Safe computer use checks the current state before each mutation, constrains accessible applications and data, requests confirmation for consequential actions, and stops when prerequisites or expected evidence are missing.
Acronyms and aliases
AI computer use variantcomputer-use agent variant
Specialised terms
Related terms
Frequently asked questions
What actions can an artificial intelligence computer-use system take?
It can inspect interface state and perform bounded actions such as clicking, typing, scrolling, selecting controls, and navigating between views.
Why is computer use harder than calling an application programming interface?
Interfaces can be visually ambiguous, change unexpectedly, and provide weaker structured guarantees about what an action will affect.
Videos explaining artificial intelligence computer use
