version: 1

parsers:
  pdf-docling:
    parser: pdf
    enabled: true
    settings:
      min_content_chars: 20
    engines:
      - backend: docling
        settings:
          do_ocr: true
          ocr_engine: rapidocr
          ocr_lang: [en]
          force_full_page_ocr: false
          bitmap_area_threshold: 0.05
          do_table_structure: true
          table_do_cell_matching: true
          accelerator_device: auto
          accelerator_threads: 4
          pdf_backend: pypdfium2
          max_file_size: 104857600
          raises_on_error: true
          export_format: markdown
          strict_text: false
          include_raw_document: false
          extra_convert_options: {}

  # Active image parser: use the same local RapidOCR/ONNX stack installed by
  # the PDF extra. The engine is memoized and reused across image parses.
  image-rapidocr:
    parser: image
    enabled: true
    settings:
      ocr_engine: rapidocr
      max_pixels: 100000000

  # Inactive alternatives. Keep exactly one definition enabled for each stable
  # parser type (pdf or image); the runtime rejects ambiguous replacements.

  # Fast embedded-text PDF extraction with no OCR or layout reconstruction.
  # pdf-pymupdf:
  #   parser: pdf
  #   enabled: true
  #   settings:
  #     min_content_chars: 10
  #   engines:
  #     - backend: pymupdf

  # Fast PDF profile: PyMuPDF first, then LiteParse.
  # pdf-fast:
  #   parser: pdf
  #   enabled: true
  #   settings:
  #     profile: fast
  #     min_content_chars: 20

  # Mixed text/layout/OCR PDF fallback chain.
  # pdf-balanced:
  #   parser: pdf
  #   enabled: true
  #   settings:
  #     profile: balanced
  #     min_content_chars: 20

  # OCR-oriented PDF fallback chain for predominantly scanned documents.
  # pdf-ocr:
  #   parser: pdf
  #   enabled: true
  #   settings:
  #     profile: ocr
  #     min_content_chars: 20

  # Highest-quality built-in PDF chain; requires more models and compute.
  # pdf-quality:
  #   parser: pdf
  #   enabled: true
  #   settings:
  #     profile: quality
  #     min_content_chars: 50

  # Local LiteParse with an optional external OCR service.
  # pdf-liteparse:
  #   parser: pdf
  #   enabled: true
  #   settings:
  #     min_content_chars: 20
  #   engines:
  #     - backend: liteparse
  #       settings:
  #         output_format: markdown
  #         image_mode: placeholder
  #         extract_links: true
  #         ocr_enabled: true
  #         ocr_language: eng
  #         ocr_server_url: http://127.0.0.1:8080
  #         max_pages: 1000
  #         dpi: 200
  #         preserve_very_small_text: false
  #         quiet: true
  #         num_workers: 4
  #         input_mode: path
  #         include_raw: false
  #         extra_options: {}
  #       secrets:
  #         password_env: PDF_DOCUMENT_PASSWORD

  # Local MinerU pipeline; the mineru executable must be installed on PATH.
  # pdf-mineru:
  #   parser: pdf
  #   enabled: true
  #   settings:
  #     min_content_chars: 20
  #   engines:
  #     - backend: mineru
  #       settings:
  #         backend: pipeline
  #         executable: mineru
  #         timeout_seconds: 600
  #         method: auto
  #         language: null
  #         effort: medium
  #         keep_output: false
  #         extra_args: []

  # OCR-heavy local PDF analysis with CPU-safe PaddleOCR settings.
  # pdf-paddleocr:
  #   parser: pdf
  #   enabled: true
  #   settings:
  #     min_content_chars: 20
  #   engines:
  #     - backend: paddleocr
  #       settings:
  #         pipeline_class: PPStructureV3
  #         fallback_pipeline_classes: [PaddleOCRVL, PPStructure]
  #         lang: en
  #         device: cpu
  #         enable_hpi: false
  #         use_tensorrt: false
  #         precision: fp32
  #         enable_mkldnn: true
  #         cpu_threads: 4
  #         use_doc_orientation_classify: true
  #         use_doc_unwarping: true
  #         use_textline_orientation: true
  #         use_table_recognition: true
  #         use_formula_recognition: false
  #         use_chart_recognition: false
  #         use_region_detection: true
  #         format_block_content: true
  #         markdown_ignore_labels: [header, footer]
  #         extra_options: {}
  #         legacy_ocr_options: {}

  # Legacy image OCR using the external Tesseract executable.
  # image-pytesseract:
  #   parser: image
  #   enabled: true
  #   settings:
  #     ocr_engine: pytesseract
  #     lang: eng
  #     config: ""
  #     timeout: 60
  #     max_pixels: 100000000