Browser-based graphical shell and control plane for a Linux server you own — real terminal, file manager, live metrics and BYOK AI in one mobile-first workspace. Next.js 16 · React 19 · node-pty · Tailwind v4 · shadcn/ui. Public alpha, self-hosted.
TypeScript browser-shellcontrol-planefile-managerhomelabmobile-firstnextjsptyreactself-hostedserver-managementshadcn-uisysadmintailwindcssterminaltypescriptvpsweb-desktopweb-terminal
/* Material 3 Expressive motion — the Android shell's spring tokens, in one place.
WHAT CHANGED AND WHY. This shell animated with `cubic-bezier(0.2,0.9,0.3,1.06)`
at a flat 300ms, which is the pre-2025 Material model (duration + easing).
M3 Expressive (m3.material.io, 2026) replaced that: motion is now described by
a spring — a stiffness/damping pair — in two families. "Spatial" springs drive
anything that moves or resizes and are allowed to overshoot; "effects" springs
drive colour and opacity and must NOT, because an alpha that overshoots past 1
clips flat and reads as a flicker rather than as momentum.
CSS has no spring primitive. Each easing below is a `linear()` SAMPLED off the
real spring's unit-step response (mass 1, as M3 specifies), truncated where the
curve settles inside ±0.5% of rest.
THE CEILING OF THIS APPROXIMATION, stated plainly: a sampled linear() is still
a fixed-duration easing. A real spring can be interrupted mid-flight and
re-targeted, carrying its current velocity into the new animation. A CSS easing
always restarts from zero velocity and always runs its full duration, so
flicking the shade open-closed-open looks like three separate animations rather
than one continuous mass. Anything that genuinely needs interruptible physics
has to leave CSS for WAAPI or a JS spring; do not try to fix it by adding more
keyframes here.
ONLY THREE CURVES, not six. Normalised to 0..1 a spring's SHAPE depends only on
the damping ratio; stiffness merely scales time. So M3's six spring tokens
collapse to three easings plus six durations:
ζ=0.6 spatial fast → 9.3% overshoot
ζ=0.8 spatial default/slow → 1.5% overshoot
ζ=1.0 every effects token → critically damped, no overshoot
DURATIONS ARE calc(var(--shell-dur) * K), NOT RAW MS, ON PURPOSE. globals.css
collapses --shell-dur to 1ms under prefers-reduced-motion. Before this change
the only Android motion that honoured that was the one animation already
written in terms of --shell-dur; every press/enter animation added here would
have silently opted OUT of the reduce-motion contract if the ms were hard-coded.
K = settle time ÷ the 300ms Android --shell-dur:
spatial fast 322ms (k=800, ζ=0.6) ×1.073
spatial default 371ms (k=380, ζ=0.8) ×1.237
spatial slow 511ms (k=200, ζ=0.8) ×1.703
effects fast 121ms (k=3800, ζ=1.0) ×0.403
effects default 186ms (k=1600, ζ=1.0) ×0.620
effects slow 263ms (k=800, ζ=1.0) ×0.877
The tokens themselves now live under [data-shell="android"] in app/globals.css,
so every Android surface inherits them from the shell root with no provider and
no per-render style object. This module keeps only the class that consumes them.
BROWSER FLOOR: linear() is Chrome 113 / Safari 17.2 / Firefox 112. Older engines
drop the whole timing-function declaration and fall back to `ease`, which is a
correct-looking degrade, not a break — so no @supports guard. */
/* Touch press-down. Real Android answers a tap with a ripple, which CSS cannot
draw without a pseudo-element sized from the touch point; a spring scale-down
is the honest substitute and carries the same "the surface is physical" signal.
`transition-transform` also covers `scale` (Tailwind 4 puts scale in the
transform group), and it deliberately REPLACES the shadcn Button base's
`transition-colors` via tailwind-merge — every call site here is
`hover:bg-transparent`, so there is no colour transition to lose, and keeping
`transition` (all properties) would have put a 9.3%-overshoot SPATIAL curve on
colour, which is exactly the family mix M3 forbids. */
export const M3_PRESS =
"transition-transform duration-[var(--m3-dur-spatial-fast)] ease-[var(--m3-spatial-fast)]";