paradox-pdf
Copyright (c) 2024-2026 CreAI

This product includes software developed by CreAI
(https://github.com/CreAI-mx/DocumentExtractor).

Proprietary software. Copyright (c) 2024-2026 CreAI. All rights reserved.
See the LICENSE file for terms. For commercial use, contact
feliperodriguez@creai.mx. The third-party components listed below remain under
their own licenses.

================================================================================
Third-party components bundled or downloaded at runtime
================================================================================

This distribution depends on, bundles, or downloads at runtime the following
third-party models and libraries. Each component remains the property of its
respective copyright holder and is governed by its own license. The summaries
below are informational only; consult each project's upstream LICENSE for the
authoritative terms.

--------------------------------------------------------------------------------
1. Microsoft Table Transformer (TATR)
   - Repository: https://github.com/microsoft/table-transformer
   - Copyright: (c) Microsoft Corporation
   - License: MIT License
   - Used for: Table structure recognition in scanned/photographed pages
     (`paradox_pdf/pipeline/table_vision.py`).

--------------------------------------------------------------------------------
2. PP-DocLayoutV2 (PaddleX / PaddleOCR)
   - Repository: https://github.com/PaddlePaddle/PaddleX
   - Copyright: (c) PaddlePaddle Authors
   - License: Apache License, Version 2.0
   - Used for: Document layout detection and pointer-network reading order
     in the vision pipeline.

--------------------------------------------------------------------------------
3. TexTAR (Text Attribute Recognition)
   - Repository: https://github.com/RohanSaw/TexTAR
   - Copyright: (c) TexTAR Authors
   - License: Verify upstream LICENSE before redistribution.
              Weights are bundled in `paradox_pdf/pipeline/textar_model/`; downstream
              users must comply with the upstream terms when redistributing.
   - Used for: Per-word mark detection (bold / italic / underline /
     strikethrough) in `paradox_pdf/pipeline/marks_vision.py`.

--------------------------------------------------------------------------------
4. RapidOCR (ONNX Runtime)
   - Repository: https://github.com/RapidAI/RapidOCR
   - Copyright: (c) RapidAI
   - License: Apache License, Version 2.0
   - Used for: Optical Character Recognition in the CPU vision backend.

--------------------------------------------------------------------------------
5. DocLayout-YOLO (legacy fallback)
   - Repository: https://github.com/opendatalab/DocLayout-YOLO
   - License: AGPL-3.0 (verify before redistributing weights)
   - Used only as a fallback when PP-DocLayoutV2 cannot be loaded.

--------------------------------------------------------------------------------
6. PaddleOCR-VL (optional, [gpu] profile only)
   - Repository: https://github.com/PaddlePaddle/PaddleOCR
   - License: Apache License, Version 2.0
   - Used for: 0.9B-parameter VLM for layout + OCR + tables in one pass.

--------------------------------------------------------------------------------
7. Anthropic Python SDK (optional, [llm] profile only)
   - Repository: https://github.com/anthropics/anthropic-sdk-python
   - License: MIT License
   - Used for: Claude API fallback when local pipelines fail on a page.

--------------------------------------------------------------------------------
8. PyMuPDF
   - Repository: https://github.com/pymupdf/PyMuPDF
   - License: AGPL-3.0 (also available under a commercial license from
              Artifex Software for proprietary use).
   - Used for: Digital PDF parsing, font extraction, and rendering.

================================================================================

For the full list of Python-level dependencies and their individual licenses,
see `pyproject.toml` and the installed wheel metadata.
