Menu-bar cockpit + chip-agnostic accuracy.
- Live 6-bar menu-bar glyph — CPU / GPU / ANE / Media / memory-usage / memory-bandwidth as colored mini bars with a stacked "SS" label, drawn as a bitmap for reliable rendering. The whole glyph blinks red on an alert (swap, memory-pressure critical, or GPU throttle).
- Revamped dropdown — six color-matched trend graphs mirroring the glyph, each on a fixed Y axis matched to its bar (no auto-scale exaggeration); memory usage is now a line graph; top processes; and an Open Dashboard button that brings the main window forward from the background. Tighter, denser layout.
- Honest AI attribution — distinguishes a loaded runtime from an idle daemon (
Ollama (idle)) and an unmanaged in-app / MLX-Swift workload (in-app / unmanaged); the dashboard no longer credits an idle daemon for GPU work done by another app. - Chip-agnostic bottleneck verdict — bandwidth-bound is judged against the machine's own observed achievable bandwidth peak (self-calibrating across M1…M5+), not a fixed fraction of the theoretical spec. Observed peaks decay slowly so a one-off spike no longer pins the scale.
- Compact dashboard — smaller card padding / heights / spacing and a narrower default window.
Notarized DMG below. Apple Silicon only, sudoless.