ai

The control in an assistant interface that picks which model answers, usually a menu carrying a short description of each option.

The composer at the bottom of an assistant interface: a growing text area with attachment, model and send controls attached to it.

The three bouncing dots shown while someone is composing a reply, or while an assistant is preparing one.

An interface operated in turns of language rather than by controls, where the transcript is the state and the only visible affordance is a box to type in.

An interface whose parts are chosen or composed by a model at runtime, so a request is answered with a built form, chart, or panel instead of a paragraph of text.

Completing the reader's text inside the field itself, with the guessed remainder selected so the next keystroke replaces it.

An interface that accepts and answers in more than one channel at once, letting a person speak, type, touch, or point and pick whichever suits the moment.

An interface whose input is ordinary language, typed or spoken: the user states an intent in their own words and the system parses it, instead of picking from controls.

Typing a forward slash in a composer to open a menu of actions, where choosing one runs it rather than inserting the words into the message.

A paradigm where the screen stops being the interface and the interaction happens through voice, gesture, presence, or sensors, so the system responds without anything to look at.

Revealing text one character at a time, as if it were being typed, usually with a blinking caret trailing the last letter.

How a screen reader is told about text that arrives token by token, holding the region quiet while it streams and speaking once when the answer is complete.