Toast notifications were designed for one event at a time. AI agents generate dozens. When notifications pile up, you stop reading them. When audio plays, you can't not hear it.
Desktop notifications work well for occasional, important events. A calendar reminder. A message from a colleague. A deploy finishing. One toast appears, you read it, you act or dismiss. The system was designed for this cadence: a few events per hour, each worth your attention.
AI agents break this cadence. Three agents running in parallel generate fifteen to thirty events per hour. Each event fires a toast. The toasts stack in your notification center. You glance at the pile, see a wall of similar-looking messages, and dismiss them all with one swipe. Somewhere in that pile was a build failure. You didn't see it.
This is the pile-up problem: when the volume of notifications exceeds your willingness to read them, you switch from reading to dismissing. The notification system is technically working. It delivered every event. But you stopped processing the deliveries. The system failed at the thing it was supposed to do: make you aware.
Audio is temporal. Events arrive one at a time, play for a few seconds, and disappear. There is no unread count. There is no pile to dismiss. You can't "swipe away" a voice that already spoke.
Radio Agent plays events sequentially on an audio stream. When three events arrive close together, they play one after another, each getting its own moment. A completion plays, then a tone for a start, then a voice for a failure. Each is distinct in time. None gets buried under the others.
The critical difference: notifications occupy space (they stack on your screen), while audio occupies time (events play in sequence). Space piles up. Time doesn't. You can ignore a stack of toasts. You can't ignore a voice saying "build failed" while you're standing in the kitchen.
The most important event in a multi-agent session is a failure. Everything else is nice to know. A failure is need to know. The question is: how quickly do you learn about it?
With desktop notifications, failure detection depends on you reading the toast before it disappears (typically 5-10 seconds), or scrolling through the notification center to find it among the completions and status updates. If you were focused on another window when the toast appeared, you might not see it for minutes or hours.
With Radio Agent, failures are impossible to miss. They use a different voice (male instead of female) and a dissonant tone. Your ear recognizes "something went wrong" before your brain processes the words. This works whether you're at your desk, in the kitchen, or in the next room. No reading required. No scanning required.
| Desktop notifications | Audio (Radio Agent) | |
|---|---|---|
| Delivery | Visual toast, stacks in notification center | Spoken voice over music stream |
| At 30 events/hr | Pile up. You dismiss in bulk. | Play sequentially. Each gets its moment. |
| Failure visibility | Same toast style as completions. Easy to miss. | Different voice + dissonant tone. Unmissable. |
| Away from desk | Silent. Toasts stack unseen. | Audible from any room. |
| Attention model | Demands reading. Interrupts focus. | Peripheral. No context switch. |
| After dismissal | Gone, unless you open notification center. | Dashboard wire feed shows recent history. |
| Sound | Same ding for everything, or silent. | Distinct tones per event type. |
Desktop notifications put events in a stack and hope you read them. Radio Agent puts events in a stream and speaks them aloud. One is optimized for low-volume, high-importance alerts. The other is optimized for continuous, ambient monitoring of busy agent sessions.
If you've ever swiped away a pile of agent notifications without reading them, you've already outgrown what toasts can do.
curl -sSL https://radioagent.live/install.sh | bash
Runs on your machine. No cloud service, no notification permissions to configure. Plays music, speaks announcements, and streams to any device on your LAN.