@czap/detect

Probes browser APIs — GPU, CPU cores, memory, motion and color preferences, viewport, network — and maps the results to the capability tiers the rest of LiteShip uses to gate output.

You usually don't install this directly — it arrives as a dependency of @czap/astro, which runs detection for you in the client runtime. Install @czap/astro instead unless you need the capability probes standalone.

Install

pnpm add @czap/astro # brings @czap/detect with it

If you do install it directly, effect must be the Effect 4 beta: pnpm add @czap/detect effect@beta.

30 seconds

import { Detect } from '@czap/detect';
import { Effect } from 'effect';

const result = Effect.runSync(Detect.detect());
console.log(result.tier);       // 'static' | 'styled' | 'reactive' | 'animated' | 'gpu'
console.log(result.motionTier); // 'none' | 'transitions' | 'animations' | 'physics' | 'compute'
console.log(result.confidence); // 0.5–1.0 — low means probes fell back to defaults

In a browser this logs the device's capability level, its motion tier (reduced-motion preference forces 'none'), and a confidence score. All probes are synchronous and never throw — Effect.runSync is safe here. Detect.watchCapabilities(onChange) re-detects on viewport and preference changes.

Where it sits

This package is host-adjacent — it touches browser APIs so nothing else has to. Its only @czap dependency is @czap/core, for the CapLevel and MotionTier types it maps detected hardware onto. What to do at each tier is decided elsewhere: @czap/quantizer gates outputs by motion tier, and @czap/astro applies detection during hydration. It also exports the capability-attribute vocabulary — CAP_AXES / capAxisAttr for the tier/motion/design axes — the single source @czap/edge and the client runtime project to data-czap-* attributes, so the emitted attribute name and the locals field name can't drift. See the package surfaces map for the full layout.

If it does nothing

Detection never throws; outside a browser (SSR, Node tests) every probe falls back, and you get mid-range defaults that look plausible. If results are identical across wildly different devices, check result.confidence — a low value means the probes fell back rather than measured. Run detection on the client.

Docs

• Getting started
• Authoring model — authoring for tiers, including reduced motion
• Glossary — the vocabulary used above
• API reference — generated from source

Part of LiteShip — powered by the CZAP engine (Content-Zoned Adaptive Projection), distributed as @czap/* packages.