All tests run on an 8-year-old MacBook Air.

The first version of HiyokoLogcat's AI diagnosis UI was a fixed panel at the bottom of the screen.

It took up 30% of the vertical space — permanently. Even when you weren't using AI. Users hated it.

The second version: a full-screen overlay that appears on demand and disappears with ESC. Here's why it works better and how I built it.

The problem with fixed panels

A fixed AI panel makes a promise: "AI is always here." But most of the time, you're reading logs, not diagnosing them.

30% of screen space dedicated to a feature you use 10% of the time is a bad trade.

The overlay model makes a different promise: "AI appears when you need it and gets out of the way when you don't."

The React overlay

interface DiagnosisOverlayProps {
  isVisible: boolean;
  result: string;
  isLoading: boolean;
  onClose: () => void;
}

export function DiagnosisOverlay({
  isVisible, result, isLoading, onClose
}: DiagnosisOverlayProps) {
  // Close on ESC key
  useEffect(() => {
    const handleKey = (e: KeyboardEvent) => {
      if (e.key === 'Escape') onClose();
    };
    if (isVisible) {
      window.addEventListener('keydown', handleKey);
    }
    return () => window.removeEventListener('keydown', handleKey);
  }, [isVisible, onClose]);

  if (!isVisible) return null;

  return (

 {
        if (e.target === e.currentTarget) onClose();
      }}
    >


        ✕
        {isLoading ? (

🐥 診断中...

        ) : (

{result}

        )}




  );
}

Click the backdrop → close. Press ESC → close. Click X → close. Three ways out, none of them surprising.

The loading state

Gemini takes 2–6 seconds. The overlay needs to communicate that something is happening:

.loading {
  display: flex;
  align-items: center;
  gap: 8px;
  animation: pulse 1.5s ease-in-out infinite;
}

@keyframes pulse {
  0%, 100% { opacity: 1; }
  50% { opacity: 0.4; }
}

A pulsing "diagnosing..." indicator. Simple. Users know the app is working.

Preserving the last result

Don't clear the diagnosis when the overlay closes. If the user closes it and re-opens it, they should see the same result without waiting for another API call:

const [lastResult, setLastResult] = useState('');
const [isOverlayOpen, setIsOverlayOpen] = useState(false);

const handleDiagnose = async (logLine: LogLine) => {
  setIsOverlayOpen(true);
  setIsLoading(true);
  const result = await diagnose(logLine);
  setLastResult(result);
  setIsLoading(false);
};

The result persists until a new diagnosis is requested. Closes and reopens instantly.

The lesson

UI for AI features should follow the same principles as any other UI: take up space proportional to how often the feature is used, and get out of the way when it's not.

An overlay that appears on demand beats a permanent panel every time.

HiyokoLogcat is free and open source → github.com/hiyoyok/HiyokoLogcat
X → @hiyoyok