Device Capability

The checkCapability() function probes the device’s hardware in milliseconds, letting you make UI decisions before loading any models.

Basic Usage

import { checkCapability } from '@webllm-io/sdk';

const report = await checkCapability();

if (report.webgpu) {
  showLocalAIToggle();  // All grades (S/A/B/C) support local inference
} else {
  showCloudOnlyBadge();
}

CapabilityReport

checkCapability() returns a CapabilityReport with these fields:

interface CapabilityReport {
  webgpu: boolean;          // WebGPU API available
  gpu: GpuInfo | null;      // GPU vendor, name, estimated VRAM
  grade: DeviceGrade;       // 'S' | 'A' | 'B' | 'C'
  connection: ConnectionInfo;
  battery: BatteryInfo | null;
  memory: number;           // navigator.deviceMemory (GB), 0 if unavailable
}

GpuInfo

interface GpuInfo {
  vendor: string;  // e.g. 'nvidia', 'apple', 'intel'
  name: string;    // GPU adapter name
  vram: number;    // Estimated VRAM in MB
}

ConnectionInfo

interface ConnectionInfo {
  type: string;     // 'wifi', '4g', 'unknown', etc.
  downlink: number; // Estimated bandwidth in Mbps
  saveData: boolean; // Data saver mode
}

BatteryInfo

interface BatteryInfo {
  level: number;    // 0 to 1
  charging: boolean;
}

Device Grades

The grade is based on estimated VRAM:

Grade	VRAM Threshold	Typical Device
S	≥8192 MB	Desktop with dedicated GPU (RTX 3060+, M1 Pro+)
A	≥4096 MB	Gaming laptop, M1 MacBook
B	≥2048 MB	Integrated GPU, mid-range mobile
C	<2048 MB	Low-end mobile, older hardware

All grades support local inference — grade C uses a lightweight model (Qwen2.5-1.5B).

Conditional Features

Use the report to progressively enable features:

const report = await checkCapability();

// Only offer local AI on capable devices
if (!report.webgpu) {
  config.local = false;
}

// Warn on metered connections
if (report.connection.saveData || report.connection.downlink < 2) {
  showDataWarning();
}

// Skip local on low battery
if (report.battery && report.battery.level < 0.2 && !report.battery.charging) {
  config.local = false;
}

Pre-flight Pattern

Check capabilities before creating the client:

import { checkCapability, createClient } from '@webllm-io/sdk';

const cap = await checkCapability();

const client = createClient({
  local: cap.webgpu ? 'auto' : false,
  cloud: { baseURL: 'https://api.example.com/v1' },
});

This avoids unnecessary WebGPU initialization on unsupported devices.