On the fifth attempt to build a dashboard for a How-To Geek writer’s home, Codex changed what it meant to select a control. Instead of looking for a button that performed a particular task, the writer could trace a path across the screen, choosing where to act, which device to operate and what it should do in one gesture.

How a gesture becomes a command

The writer used the study light to illustrate how the interface works. Starting a line at the study brings up a selection of that room’s devices in the next column. Extending the line through the light reveals a set of actions in the last column. The choices depend on the light’s current state: they might include switching it off or setting its brightness to 20%. The full lighting controls are also accessible from there.

On a computer, the writer traced the path with a mouse or trackpad; on a phone, a finger did the same job. The writer found the gesture intuitive but left open whether the dashboard would see long-term use. The claim to novelty was similarly narrow: the writer hadn’t encountered the idea in Home Assistant before, not that nobody else had built it.

Why the earlier designs missed the brief

The original request was selective about function but open about form. Codex was to inspect the home’s devices, entities and automations, then include only devices the writer regularly operated by hand. The writer granted fairly broad access through Home Assistant’s Model Context Protocol tool, which supports both reading and writing. The writer noted a privacy trade-off: that access can expose considerable information about a home to a tool that processes data in the cloud.

What stood out to Codex wasn’t necessarily what needed a manual control. The writer had built elaborate automations to monitor and reduce mold risk in an old, damp house. Its first design gave that concern much of the dashboard. The second shifted attention to what various automations were about to do. A third kept two frequently used evening controls visible — the bedside lamp and electric blanket — but paired them with a list of unresolved issues. To the writer, it still felt more like a status display.

A request to move beyond button cards produced a camera motif. A shutter framed the video doorbell feed, while different lenses brought up controls for lighting, heating or audio. The writer’s objection was that switching lenses changed the theme without fundamentally changing how the controls worked. With little usage left on the Codex plan, the writer asked for one more design. The resulting route interface addressed that specific objection: selecting a room and passing through a device determined which actions the writer could reach next.