Skip to content

Changelog

All notable changes to EdgeParse are documented here.

The format follows Keep a Changelog and this project adheres to Semantic Versioning.


  • Type 3 / Skia font widths and metadata; header/footer swallowing; raster OCR routing; WASM compact JSON
  • ActualText, TJ word gaps, CID/Identity-H, encrypted PDFs, OCG visibility
  • OCR on/off across Web/WASM/Python/Node/CLI + demo toolbar toggle
  • PDF format helpers (Type3 OCR seam, image region classifier, codecs, OCG, pdf_string)
  • OCR Models guide — tiers, OPFS/Cache, Hugging Face CSP, self-host / offline
  • Site and Web SDK docs: OPFS+Cache (not IndexedDB); CSP includes huggingface.co

  • Skia/Chrome Pattern fills, filled-path decoration, OCR routing for image-only pages
  • Overpaint text dedupe, heading false positives, title/checkbox markdown
  • Pattern-image fixtures and WASM/Playwright e2e coverage

  • Two-phase ParseSession and @edgeparse/web observable SDK with PP-OCRv6
  • Dual labeled benchmark boards (odl-bench + EdgeParse harness)
  • Web SDK quick-start and API docs; accessible demo consent / quality UI
  • Optional in-browser OCR for image tables; born-digital path still needs no ML stack
  • Site and README numbers now sourced from benchmark.ts claims

  • Zero Clippy warnings across all crates
  • Corrected .gitignore to exclude build artifacts cleanly
  • Comprehensive Rust doc comments on public API surface
  • Step-by-step tutorials for CLI, Python SDK, Node.js SDK, Rust library, and output formats
  • CI/CD publishing guide
  • Bumped version to 0.1.1 in all crates and SDK manifests

  • Core extraction engine (edgeparse-core): Rust-native PDF-to-structured-data pipeline — no ML, no Java, no GPU
  • Python SDK (edgeparse): PyO3-based bindings, available on PyPI
  • Node.js SDK (edgeparse): NAPI-RS bindings, available on npm
  • CLI binary (edgeparse-cli): Zero-dependency binary for all major platforms
  • Rust library (edgeparse-core): First-class crate published to crates.io
  • Reading-order reconstruction for multi-column and sidebar layouts
  • Ruling-line and borderless table detection with cell-span merging
  • Heading and paragraph classification
  • AI safety filters (PII scrubbing, content flags)
  • Tagged PDF support (PDF/UA accessibility structure)
  • Output formats: JSON (full schema), Markdown, HTML, plain text
  • Benchmark suite comparing EdgeParse against Docling, Marker, pymupdf4llm, MinerU, MarkItDown, and LiteParse
  • Docker image for containerised deployment
  • GitHub Actions CI workflows for Rust, Python, Node.js, and Docker releases
  • Renamed Node.js package from @edgeparse/pdf → edgeparse