docling_chunk() gains a doc_items list-column with per-item provenance:
self_ref, label, page and a 1-based inclusive character span
(char_start/char_end) locating each occurrence within the originating
item's text (populated by layout analysis; NA for formats without
provenance, such as Markdown sources).docling_write_chunks() persists chunk tibbles — including embedding
and other list-columns — to Parquet or Feather via the arrow package (new
optional dependency in Suggests), for hand-off to Python, DuckDB and other
Arrow-aware RAG tooling.vignette("options") gains a "Model cache and offline use" section covering
HF_HOME and HF_HUB_OFFLINE for air-gapped machines.First release: an R interface to Docling for document intelligence and RAG.
docling_convert() converts a file path or URL (or a vector of them) into a
docling_document, with pipeline options for OCR, table mode, accelerator
device/threads and image generation.as_markdown(), as_text(), as_html(), as_json() and
as_doctags().docling_tables() returns detected tables as a list of tibbles.docling_figures() returns figure captions/pages and can save images;
docling_n_pages() reports the page count.docling_chunk() produces RAG-ready chunks as a tidy tibble, with hybrid and
hierarchical chunkers, tokenizer/max_tokens control, and pages/
n_doc_items metadata.docling_embed() attaches embeddings from any user-supplied embedding
function, with batching.install_docling() / docling_available() manage the Python backend.