Markitai Rust
Copyright (c) 2026 Ynewtime

This repository contains a new Rust implementation of Markitai's public
conversion interfaces. Project source is licensed under MIT; see LICENSE.
The standard license text and original project attribution are retained.

Document parsers and runtime dependencies are resolved in Cargo.lock. The
initial implementation uses anydoc, pdf-inspector, scraper/html5ever, htmd,
mail-parser, reqwest/rustls, and other Rust crates under their respective
licenses. It does not bundle or invoke the reference Python PDF engine.

Before distributing a release, generate the resolved dependency inventory
and include each dependency's applicable license/copyright notices. A crate's
metadata license identifier alone is not the complete notice bundle.
Reference test fixtures remain in the reference checkout; audit reports record
fixture hashes and never redistribute them as newly authored fixtures.

The embedded browser workspace includes Marked 18.0.14 (MIT), DOMPurify
3.4.16 (Apache-2.0), Preact 11.0.1 (MIT), KaTeX 0.19.0 (MIT) with the web fonts its package ships
(from the KaTeX/katex-fonts repository, also MIT),
Pele 1.9.0 (MIT, the Mermaid diagram renderer, with one Lucide icon under ISC),
icon path data from Phosphor Icons 2.1.10 (MIT) and the Inter 5.3.0 variable
font (SIL Open Font License 1.1).
Their complete notices are in vendor/web/marked-LICENSE,
vendor/web/DOMPurify-LICENSE, vendor/web/preact-LICENSE,
vendor/web/KaTeX-LICENSE, vendor/web/pele-LICENSE, vendor/web/phosphor-LICENSE
and vendor/web/Inter-OFL.txt. Download provenance,
upstream package integrity and embedded-file hashes are recorded in
vendor/web/provenance.json.

Windows and Linux builds render PDF pages with hayro 0.8.0 (MIT/Apache-2.0),
which embeds the Foxit standard-14 substitute fonts from PDFium (BSD-3-Clause),
Adobe's predefined CMaps (BSD-3-Clause) and a CC0 CMYK profile; their notices
are in licenses/hayro/.

Windows and Linux builds decode AVIF with rav1d (BSD-2-Clause; VideoLAN, the
dav1d authors and the Internet Security Research Group), vendored from upstream
commit d3d1cd67 in vendor/rav1d with the changes listed in
vendor/rav1d/MARKITAI-PATCH.md. Its Rust API derives from dav1d-rs (MIT,
Luca Barbato); both notices are in vendor/rav1d/COPYING and
vendor/rav1d/LICENSE-dav1d-rs.

`image.format: webp` is encoded with the lossy encoder of libwebp 1.6.0
(BSD-3-Clause, Google Inc.), built statically from the unchanged release
sources vendored in vendor/libwebp (file list and hashes in
vendor/libwebp/UPSTREAM.json). Its license and Google's patent grant are in
licenses/libwebp/COPYING and licenses/libwebp/PATENTS.

The HTML reader resolves Tailwind CSS colour classes and variables in the
inline SVG drawings it keeps with the default colour palette of Tailwind CSS
3.4.19 (MIT, Copyright (c) Tailwind Labs, Inc.), embedded as data in
crates/markitai-core/src/formats/html/svg.rs. The complete license, the exact
source identity and a digest of the embedded values are in licenses/tailwind/.

The Numbers reader uses iwork 0.2.1 (MIT), vendored in vendor/iwork with the
changes listed in vendor/iwork/MARKITAI-PATCH.md. Its independently
sourced regression documents are from numbers-parser (MIT), with the original
license and exact source identities in
crates/markitai-core/src/formats/numbers/fixtures/.

Two EPUB regression books are in crates/markitai-core/tests/fixtures/epub/: a
trimmed Project Gutenberg eBook (public domain in the USA) and markitdown's
test.epub (MIT, Copyright (c) Microsoft Corporation), whose license is in
LICENSE-markitdown there; provenance.json records their sources and hashes.

The bounded offline token-pricing catalog derives selected rows from LiteLLM
1.106.0.dev2's bundled model data. Its original BerriAI copyright/license notice,
exact selected source objects and official tariff/alias provenance are retained
in licenses/pricing/. This attribution does not assert universal price coverage
or complete legal review of the dependency bundle.

Reviewed supplemental upstream dependency notices and full terms are included in
licenses/upstream/. The unmodified MPL-covered selectors 0.38.0 source is available
offline at licenses/upstream/source-archives/selectors-0.38.0.crate; its exact
checksum, upstream download URL and license are recorded in that directory.
These paths are relative to the archive's attribution root (the Python wheel
uses its .dist-info/licenses/ directory). Three historical objc2-family MIT
notices remain explicitly review-needed. This collection does not complete a
legal review or resolve the upstream Apple SDK licensing discussion.
