llama-index-readers-docling-rs

DoclingRsReader turns PDF, DOCX, PPTX, XLSX, HTML, EPUB, e-mail, scans and 40-odd other formats into LlamaIndex Documents with docling.rs, a Rust port of IBM's docling: layout analysis, table structure and local OCR from a single native wheel, no PyTorch. It is the API of llama-index-readers-docling — export as Markdown (the default) or as docling's native JSON for a Docling node parser.

Installation

pip install llama-index-readers-docling-rs

PDFs and images use ONNX models (~720 MB), fetched once with python -c "import docling_rs; docling_rs.download_models()"; Office, HTML and e-mail need none.

Usage

Basic usage looks like this:

from llama_index.readers.docling_rs import DoclingRsReader

reader = DoclingRsReader()
docs = reader.load_data(file_path="https://arxiv.org/pdf/2408.09869")
print(f"{docs[0].text[:500]}...")

To export to docling's native JSON (parse it downstream with a Docling node parser):

reader = DoclingRsReader(export_type=DoclingRsReader.ExportType.JSON)

With SimpleDirectoryReader:

from llama_index.core import SimpleDirectoryReader

dir_reader = SimpleDirectoryReader(
    input_dir="/path/to/docs",
    file_extractor={".pdf": reader, ".docx": reader},
)
docs = dir_reader.load_data()

The converter is docling.rs's DocumentConverter; configure it as usual and pass it as doc_converter (DoclingRsReader(doc_converter=DocumentConverter(do_ocr=False))). DoclingReader is the same class under the upstream name.

The package has no code of its own: it pins the docling-rs wheel (with its llamaindex extra) and re-exports docling_rs.llamaindex, where the parameters are documented; pip install "docling-rs[llamaindex]" and from docling_rs.llamaindex import DoclingReader is the same thing under the engine's own name. Published from the docling.rs repository, not from LlamaIndex's monorepo (which no longer takes new integration packages).

License

MIT. docling.rs is an independent project of the docling community (LF AI & Data); not a LlamaIndex package.