@voxire/glyph

Production-grade OCR and document intelligence platform.

Glyph is a fully composable, TypeScript-first SDK for document scanning, OCR, structured field extraction, and validation. It is designed for long-term extensibility with a Rust-powered preprocessing layer and ONNX runtime support.

---

Architecture

npm package (TypeScript API)
        ↓
Rust native preprocessing
        ↓
ONNX runtime
        ↓
OCR providers (RapidOCR, Tesseract, Textract, Cloud Vision)
        ↓
Document parsers (passport, invoice, receipt, custom)
        ↓
Validation layer (MRZ, IBAN, totals, custom)
        ↓
Normalized structured JSON

Monorepo structure

packages/
├── core        — Interfaces, types, pipeline builder
├── node        — @voxire/glyph main entry point
├── native      — Rust napi-rs bindings wrapper
├── providers   — OCR provider implementations
├── parsers     — Document parsers
├── validators  — Field + document validators
├── outputs     — Output formatters (JSON, CSV, text)
├── models      — ONNX model registry + downloader
├── examples    — Usage examples
└── playground  — Interactive playground

crates/
├── preprocess  — Image preprocessing pipeline
├── inference   — ONNX runtime abstraction
├── image       — Low-level image utilities
├── bindings    — napi-rs Node.js bindings
├── pdf         — PDF rasterization (Phase 3)
└── pipeline    — Rust-side orchestration

---

Quick start

pnpm add @voxire/glyph

Zero-config scan

import { scanDocument } from '@voxire/glyph';

const result = await scanDocument('/path/to/document.jpg');
console.log(result.documentType);   // 'generic'
console.log(result.confidence);     // 0.98
console.log(result.fields);         // extracted key-value fields

Full pipeline control

import {
  createPipeline,
  RapidOcrProvider,
  JsonFormatter,
} from '@voxire/glyph';
import { PassportParser } from '@voxire/parsers';
import { MrzValidator } from '@voxire/validators';

const pipeline = createPipeline()
  .useOCR(new RapidOcrProvider({ numThreads: 4 }))
  .useParser(new PassportParser())
  .useValidator(new MrzValidator())
  .useOutput(new JsonFormatter())
  .build();

const json = await pipeline.run('/path/to/passport.jpg');
console.log(json);
// {
//   "documentType": "passport",
//   "confidence": 0.98,
//   "fields": { "firstName": { "value": "Mohamad", "confidence": 0.98 }, ... },
//   "blocks": [...],
//   "warnings": [],
//   "metadata": { "provider": "rapidocr", "processingMs": 123 }
// }

await pipeline.dispose();

Hooks & plugins

const pipeline = createPipeline()
  .useOCR(provider)
  .useParser(parser)
  // Inline hook
  .on('afterOcr', (ctx) => {
    console.log(`OCR confidence: ${ctx.data.result.confidence}`);
  })
  // Plugin
  .usePlugin(new MyAuditPlugin())
  .build();

---

OCR Providers

Provider	Execution	Network	Languages
RapidOcrProvider	Local (ONNX)	No	en, zh, ja, ko, + more
TesseractProvider	Local (WASM)	No	100+
MockOcrProvider	In-process	No	— (testing)
TextractProvider	Cloud (AWS)	Yes	en, fr, de, es, + more
CloudVisionProvider	Cloud (GCP)	Yes	50+

---

Parsers

Parser	Document Type	Fields
PassportParser	passport	firstName, lastName, documentNumber, MRZ validation
InvoiceParser	invoice	invoiceNumber, total, IBAN, dates
ReceiptParser	receipt	merchantName, total, items
GenericParser	generic	Configurable regex patterns

---

Validators

Validator	Applies To
MrzValidator	passport
IbanValidator	invoice
InvoiceTotalValidator	invoice

---

Development

# Install dependencies
pnpm install

# Build TypeScript packages
pnpm build:ts

# Build Rust crates
pnpm rust:build

# Build native Node.js bindings
pnpm build:native

# Run tests
pnpm test

# Type check
pnpm typecheck

# Lint Rust
pnpm rust:clippy

---

Phases

Phase	Status
1 — Monorepo, TypeScript SDK, Rust bindings, preprocessing, OCR provider	✅
2 — Parser architecture, passport, invoice, validation layer	✅
3 — PDF support, layout analysis, table extraction	Planned
4 — Fraud detection, GPU, cloud/local hybrid, enterprise	Planned

---

License

MIT © Voxire