Technology · Widgets

The canvas your agent builds.

A widget is a self-contained folder — its own UI, its own data, its own logic. Your agent creates them, operates them, and some keep working quietly even when you're not looking.

Zaelar's own code Background execution
Fullscreen

Three separate things happen to a widget, and Zaelar never confuses them. Operating its data — adding a task, marking something done — is instant, handled by the FlashBrain calling a declared action. Creating or modifying its code — a brand new widget, or changing how one works — is real engineering work, done by a Brain Worker that writes and validates code. Opening, closing or deleting a widget on the canvas is a lightweight, deterministic action — deleting always asks for confirmation first.

Refreshed by events, never polling

A widget's card re-renders exactly once when its data actually changes — pushed over a live event stream, not a timer. Nothing on the canvas polls.

Some widgets keep working while closed

A widget can declare a background cycle — a periodic tick that runs whether or not its card is open. That's how messaging stays triaged and memory stays current even if you never opened the widget: it keeps reading its connectors, deciding what matters, and writing it to memory, isolated from everything else so a stuck widget never affects voice or another widget.

Code and data live in separate folders on purpose — regenerating or modifying a widget's code never touches what it has stored.