verifyfetch

Download large files. Verify them. Resume when it fails.

Install

npm install verifyfetch

Quick Start

import { verifyFetch } from 'verifyfetch';

const response = await verifyFetch('/model.bin', {
  sri: 'sha256-uU0nuZNNPgilLlLX2n2r+sSE7+N6U4DukIj3rOLvzek='
});
// Throws if hash doesn't match

The Problem

Download a 4GB AI model. Network drops at 3.8GB. Start over.

Also:

• Native crypto.subtle.digest() buffers entire file = 4GB model needs 4GB RAM
• No way to detect corruption until after downloading everything

The Solution

Use Cases

• AI models - WebLLM, Transformers.js, ONNX (multi-GB files)
• WASM modules - Game engines, video codecs
• Large data - Fonts, images, datasets

API

// Basic verification
verifyFetch(url, { sri: 'sha256-...' })

// Streaming (constant memory)
verifyFetchStream(url, { sri: 'sha256-...' })

// Resumable (survives page reload)
verifyFetchResumable(url, { chunked: {...}, persist: true })

// Multi-CDN failover
verifyFetchFromSources(sri, path, { sources: [...] })

// Service Worker (automatic verification)
createVerifyWorker({ manifestUrl: '/vf.manifest.json' })

Important Notes

• WASM for streaming: True constant-memory streaming requires WASM. Without it, files >50MB are buffered (warning shown in console).
• SRI format: Hashes use Subresource Integrity format (sha256-BASE64...)
• Resumable downloads: Requires server support for HTTP Range requests.

Full Documentation

See GitHub README for:

• Complete API reference
• Manifest format
• CLI commands
• Examples

Generate Hashes

npx verifyfetch sign ./public/*.wasm
npx verifyfetch sign --chunked ./large-model.bin  # For resumable downloads

License

Apache-2.0