Package {doclingr}

doclingr: Document Intelligence via 'Docling'

Description

An interface to 'Docling', a document-understanding library that converts 'PDF', 'DOCX', 'PPTX', 'HTML' and image documents into structured, AI-ready data. The package wraps the 'Docling' 'Python' package through 'reticulate' to extract layout-aware text, tables and metadata, export to 'Markdown' or 'JSON', and split documents into context-rich chunks suitable for retrieval-augmented generation (RAG) and embedding pipelines.

Author(s)

Maintainer: Andre Leite leite@castlab.org

Authors:

• Andre Leite leite@castlab.org

• Marcos Wasilew marcos.wasilew@gmail.com

• Hugo Vasconcelos hugo.vasconcelos@ufpe.br

• Carlos Amorim carlos.agaf@ufpe.br

• Diogo Bezerra diogo.bezerra@ufpe.br

See Also

Useful links:

• https://github.com/StrategicProjects/doclingr

• https://strategicprojects.github.io/doclingr/

• Report bugs at https://github.com/StrategicProjects/doclingr/issues

Is the Docling backend available?

Description

Checks whether the docling Python package can be imported in the active 'reticulate' environment.

Usage

docling_available()

Value

A logical scalar.

See Also

install_docling()

Examples

## Not run: 
docling_available()

## End(Not run)

Split a document into RAG-ready chunks

Description

Apply a Docling chunker to a converted document and return the chunks as a tidy tibble. The default "hybrid" chunker produces tokenization-aware, context-enriched chunks well suited to embedding and retrieval pipelines; the "hierarchical" chunker follows the document's structural hierarchy without a token budget.

Usage

docling_chunk(
  x,
  chunker = c("hybrid", "hierarchical"),
  tokenizer = NULL,
  max_tokens = NULL,
  contextualize = TRUE,
  ...
)

Arguments

Details

The hybrid chunker is token-aware: it packs content up to a token budget and splits oversized passages. Control this with tokenizer (the model whose tokenizer defines "a token") and max_tokens (the budget). These are ignored by the hierarchical chunker.

Value

A tibble::tibble with one row per chunk and columns:

• chunk_id — 1-based index.

• text — contextualized text (or raw text if contextualize = FALSE).

• raw_text — the chunk's unmodified text.

• n_chars — number of characters in text.

• headings — list-column of heading paths for the chunk.

• pages — list-column of integer page numbers the chunk spans.

• n_doc_items — number of underlying document items in the chunk.

See Also

docling_convert(), docling_embed()

Examples

## Not run: 
doc <- docling_convert("paper.pdf")
chunks <- docling_chunk(doc, max_tokens = 512)
chunks$text[1]

# Match your embedding model's tokenizer
docling_chunk(doc, tokenizer = "BAAI/bge-small-en-v1.5", max_tokens = 512)

## End(Not run)

Convert one or more documents with Docling

Description

Runs Docling's document-understanding pipeline over local file paths or URLs and returns lightweight R handles around the resulting DoclingDocuments. Each handle can be exported to Markdown or JSON (as_markdown(), as_json()), mined for tables (docling_tables()), or split into chunks for RAG (docling_chunk()).

Usage

docling_convert(
  source,
  ocr = TRUE,
  table_mode = c("accurate", "fast"),
  device = c("auto", "cpu", "cuda", "mps"),
  num_threads = NULL,
  images = FALSE,
  images_scale = 1,
  ...
)

Arguments

Details

Supported inputs include PDF, DOCX, PPTX, XLSX, HTML, Markdown, AsciiDoc and common image formats, as determined by the installed Docling version.

Value

For a single source, an object of class docling_document (a list with the underlying Python document and the original source). For multiple sources, a docling_document_list: a list of docling_document objects named by source.

See Also

as_markdown(), docling_tables(), docling_chunk()

Examples

## Not run: 
doc <- docling_convert("https://arxiv.org/pdf/2408.09869")
as_markdown(doc)

# Batch, OCR off, fast tables
docs <- docling_convert(c("a.pdf", "b.pdf"), ocr = FALSE, table_mode = "fast")

## End(Not run)

Attach embeddings to chunks

Description

Embed a chunk tibble's text with a user-supplied embedding function and return the tibble with an embedding list-column. doclingr stays provider-agnostic: you bring the embedder (an OpenAI/Cohere/Ollama API call, a local sentence-transformers model via reticulate, anything), and this helper handles batching, validation and tidy assembly.

Usage

docling_embed(chunks, embedder, text_column = "text", batch_size = NULL)

Arguments

Value

chunks with an added embedding list-column of numeric vectors, and an n_dim integer column giving each embedding's length.

See Also

docling_chunk()

Examples

## Not run: 
chunks <- docling_chunk(docling_convert("paper.pdf"), max_tokens = 512)

# Any embedder: here a toy one
embed_fn <- function(txt) matrix(stats::runif(length(txt) * 8), nrow = length(txt))
docling_embed(chunks, embed_fn)

## End(Not run)

Export a converted document

Description

Render a docling_convert() result into a downstream-friendly format.

Usage

as_markdown(x, ...)

## S3 method for class 'docling_document'
as_markdown(x, ...)

as_text(x, ...)

## S3 method for class 'docling_document'
as_text(x, ...)

as_html(x, ...)

## S3 method for class 'docling_document'
as_html(x, ...)

as_json(x, ...)

## S3 method for class 'docling_document'
as_json(x, ...)

as_doctags(x, ...)

## S3 method for class 'docling_document'
as_doctags(x, ...)

Arguments

Value

• as_markdown(): a length-1 character string of Markdown.

• as_text(): a length-1 character string of plain text.

• as_html(): a length-1 character string of HTML.

• as_json(): an R list mirroring the DoclingDocument structure.

• as_doctags(): a length-1 character string in Docling's DocTags format.

Examples

## Not run: 
doc <- docling_convert("report.pdf")
as_markdown(doc)
str(as_json(doc), max.level = 1)

## End(Not run)

Extract figures (pictures) from a converted document

Description

Return a tidy tibble of the pictures Docling detected, with their captions and page numbers. When image_dir is supplied and the document was converted with images = TRUE, each picture is written to disk and its path returned.

Usage

docling_figures(x, image_dir = NULL, format = "png")

Arguments

Value

A tibble::tibble with one row per figure and columns:

• figure_id — 1-based index.

• caption — caption text (empty string if none).

• page — page number the figure appears on (NA if unknown).

• image_path — path to the saved image, or NA if not saved.

See Also

docling_convert()

Examples

## Not run: 
doc <- docling_convert("paper.pdf", images = TRUE)
figs <- docling_figures(doc, image_dir = "figures")
figs$image_path

## End(Not run)

Number of pages in a converted document

Description

Number of pages in a converted document

Usage

docling_n_pages(x)

Arguments

Value

An integer page count (0 for page-less formats such as Markdown).

Examples

## Not run: 
docling_n_pages(docling_convert("report.pdf"))

## End(Not run)

Extract tables as data frames

Description

Pull every table detected by Docling out of a converted document and return them as a list of tibbles, preserving the document order.

Usage

docling_tables(x)

Arguments

Value

A list of tibble::tibbles, one per detected table. The list is empty if no tables were found. Each element carries a page attribute when Docling reports the originating page.

See Also

docling_convert()

Examples

## Not run: 
doc <- docling_convert("financials.pdf")
tbls <- docling_tables(doc)
tbls[[1]]

## End(Not run)

Install the Docling Python backend

Description

Installs the docling Python package (and its dependencies) into a 'reticulate'-managed environment. This is a thin wrapper around reticulate::py_install() that you typically run once after installing doclingr.

Usage

install_docling(
  envname = "r-docling",
  method = c("auto", "virtualenv", "conda"),
  extra = NULL,
  ...
)

Arguments

Value

Invisibly NULL, called for its side effect.

See Also

docling_available(), docling_convert()