Skip to content

GlyphPDF v1.0.0 — First Public Release

Choose a tag to compare

View all tags
@eliets2 eliets2 released this 11 Jun 12:29
· 5 commits to main since this release
v1.0.0
346a554

GlyphPDF v1.0.0 — First Public Release

GlyphPDF is an open-source, privacy-first PDF workstation for Windows. No telemetry. No cloud
upload. No subscription. Your documents stay on your machine.

Licensed under Apache-2.0.

What's in v1.0.0

Core PDF Engine

  • Three-backend architecture: PoDoFo for editing/annotations/forms/encryption,
    PDFium for high-quality rendering (3-tier LRU tile cache, 256 MB), qpdf for
    linearization and repair-on-open.
  • Incremental save preserves digital signatures.
  • Corruption recovery with graceful repair-on-open warnings.

Security & Redaction

  • Edact-Ray defense: glyph-advance normalization closes the side-channel attack
    (Bland et al., PETS 2023 / arXiv 2206.02285) — redacted text cannot be reconstructed
    from cursor-position gaps.
  • Invisible OCR text scrub: Tr==3 invisible text within redaction rectangles is
    excised from the content stream, not merely painted over.
  • PAdES B-LT/B-LTA digital signatures with real DSS/VRI embedding, trust-chain
    verification against the Windows system root store, OCSP response verification, and
    ByteRange overlap detection (PDF shadow attack prevention).
  • 15+ sanitization vectors: metadata, JavaScript, embedded files, and more.
  • Content-literal injection fixed: pdfEscapeLiteralString hardens watermark,
    annotation, header/footer, and Bates numbering against injection.

OCR — Workstream 1

  • PP-DocLayoutV2 layout detector (ONNX, Apache-2.0): 11-region-type detection
    (Title/Paragraph/Table/Figure/List/Header/Footer/Equation/Reference/Caption/Other).
  • Tesseract 5.5.2 + RapidOCR PP-OCRv5 dual-engine fanout via LaneScheduler.
  • ROVER word-level fusion for multi-engine consensus — the default OCR engine
    is the ROVER ensemble (Tesseract primary + RapidOCR secondary, IoU-matched,
    confidence-weighted), with automatic fallback to Tesseract-only when the
    bundled PP-OCRv5 models are unavailable.
  • OCR result overlay in the viewer, with re-OCR on demand from the Edit ribbon.
  • All models bundled — the installer ships the PP-OCRv5 and layout ONNX
    weights plus Tesseract language data, so OCR works offline with no downloads.
  • Cross-page pipeline: layout ‖ OCR ‖ fusion overlap via CrossPagePipeline.

Djot Document Interchange — Workstream 2

  • Dual-model architecture: Structural PDF object graph ↔ Semantic SemanticDocument.
  • OCR→Djot mapping (OcrDjotMapper): reading-order layout regions → block structure,
    per-word provenance encoding for MRC sandwich-text alignment.
  • Annotation rich text in Djot with /RC XHTML + /Contents plain text + /PieceInfo
    Djot sidecar — perfect GlyphPDF roundtrip plus Acrobat/Foxit interoperability.
  • Comment threading: /IRT in-reply-to linking, review-state roundtrip, depth-indent
    view with date/status/author filters.
  • ProvenanceGuard refuses Djot save-back to signed born-PDF documents.
  • Vendored Lua 5.4 (MIT) + Djot reference parser for spec-conformant parsing.

MRC Layered Compression — Workstream 3

  • 30× compression ratio measured on synthetic A4 scanned page.
  • JBIG2 lossless foreground (agl/jbig2enc, Apache-2.0) — pattern-matching mode
    disabled (Xerox 2013 / German BSI ban).
  • JPEG2000 background at configurable ratio (Lossless 10:1 / Balanced 30:1 /
    Aggressive 50:1) via OpenJPEG 2.5.4 (BSD-2-Clause).
  • Invisible sandwich text from WS1 OCR word boxes — output is searchable PDF/A-2b.
  • veraPDF validation gate before finalization.

AI Assistant

  • Local AI via Ollama (llama3 default): real async HTTPS calls to a locally running Ollama
    instance — no data leaves your machine.
  • API key storage via Windows Credential Manager; "Test connection" verifies reachability before saving.
  • Cloud providers (Anthropic, OpenAI, Google Gemini) are planned for a future release.

Office & Image Import

  • Office→PDF via LibreOffice subprocess (docx/xlsx/pptx/odt/ods/odp/rtf/csv/txt).
  • Images→PDF via PoDoFo PdfImage XObject embedding (PNG/JPEG/TIFF).

Modes & Tools

  • 7 mode tabs: Home, View, Edit, Pages, Convert, Forms, Security
  • Pages mode: split (by-page / every-N / range expression), reorder with drag-drop.
  • Batch mode: 7 operation types, per-file progress, continue-on-failure, CSV/JSON
    error log export.
  • Form Builder: 7 field types (Text/Checkbox/Radio/Dropdown/ListBox/Date/Numeric)
    with tab-order editor and undo/redo. Signature widget and pushbutton fields are planned
    for a future release.
  • Compare mode: Myers LCS diff with move detection (green/red/orange), NEXT/PREV
    navigation.
  • Signature mode: PAdES B-LT/B-LTA signing + validation with trust indicators.
  • PDF/A mode: veraPDF subprocess validation, conformance levels, export.
  • Compress mode: MRC Off/Lossless/Balanced/Aggressive selector + live size estimate.

Branding & UI

  • App icon, splash screen, and system-tray icon (Show/Quit).
  • Dark, Light, and High Contrast themes.
  • WCAG-aligned accessibility: screen reader names, F6 region cycling, F1 help dialog.
  • Autosave every 5 minutes (configurable 1-30 min); crash recovery on next launch.

Localization

GlyphPDF v1.0 ships in English. Arabic, French, and German translations are planned for a future release.

Translation framework wired for Arabic (RTL), French, German (1394 source strings each; 0 translated in v1.0). Human-translated packages are pending commissioning for a future release.

Installation

System requirements: Windows 10/11 x64, ~500 MB disk space.

Option A — Installer (recommended)

  1. Download GlyphPDF-1.0.0-x64.msi below.
  2. Verify the SHA-256 checksum (see SHA-256 Verification below).
  3. Run the MSI installer.
  4. Launch GlyphPDF from the Start Menu or by double-clicking any .pdf file.

Option B — Portable (no installation)

  1. Download GlyphPDF-1.0.0-x64-portable.zip below.
  2. Verify the SHA-256 checksum.
  3. Unzip to any folder (e.g. C:\GlyphPDF or a USB drive).
  4. Run GlyphPDF.exe directly from that folder.

The portable edition is fully self-contained — all runtimes, models, and language data are included. Nothing is written outside the unzipped folder.

Optional dependencies (not bundled):

  • LibreOffice (for Office→PDF import)
  • veraPDF CLI (for PDF/A live validation in the PDF/A panel)
  • Ollama (for local AI assistant without an API key)

SHA-256 Verification

SHA-256: 79286BA3F8756B166C7ED4F8D1F2342CE0AE41EA9018C73A26E16BE010AF827B
File:    GlyphPDF-1.0.0-x64.msi

SHA-256: 530D24541824362CCD2EAFD55F1BB15ABABE461EB1662E32D80B1CDEB6E17BB7
File:    GlyphPDF-1.0.0-x64-portable.zip

PowerShell verification:

(Get-FileHash "GlyphPDF-1.0.0-x64.msi" -Algorithm SHA256).Hash
# Compare with the hash above — must match exactly

Command prompt:

certutil -hashfile GlyphPDF-1.0.0-x64.msi SHA256

Build From Source

# MSYS2 ucrt64 toolchain (GCC 16.1.0, Qt 6.11.0)
pacman -S mingw-w64-ucrt-x86_64-qt6-base mingw-w64-ucrt-x86_64-podofo \
          mingw-w64-ucrt-x86_64-tesseract-ocr mingw-w64-ucrt-x86_64-openjpeg2

mkdir build && cd build
cmake .. -G Ninja -DCMAKE_BUILD_TYPE=Release
cmake --build . --parallel 8
QT_QPA_PLATFORM=offscreen ctest --output-on-failure -j4

Full package list in README.md → "Build Instructions".

Acknowledgements

GlyphPDF would not exist without these open-source projects:

Library License Use
Qt 6.11 LGPL-3.0 Application framework
PoDoFo 1.1.0 LGPL-2.0 PDF editing + annotations
PDFium BSD-3-Clause PDF rendering
qpdf 12.3.2 Apache-2.0 PDF linearization + repair
Tesseract 5.5.2 Apache-2.0 OCR engine
OpenJPEG 2.5.4 BSD-2-Clause JPEG2000 encoding
agl/jbig2enc Apache-2.0 JBIG2 lossless compression
Lua 5.4 MIT Djot reference parser host
ONNX Runtime MIT ML inference (layout + OCR)
OpenSSL Apache-2.0 Cryptographic operations

Full dependency audit: docs/MRC-ENCODER-LICENSE-AUDIT.md

Contributions welcome — see CONTRIBUTING.md for the contribution guide and SECURITY.md for responsible disclosure.

Assets 6
Loading