Research papers

Paper.Root analyses the document from the PDF alone — typically in 0.2–0.5 s — and exposes a PaperState (bind:paper):

Field	Contents
`meta`	Title, authors, abstract, arXiv id, DOI
`sections` / `flatSections`	Section tree from the outline, or from headings in the text
`references`	Parsed bibliography entries (authors, year, title, venue, DOI, arXiv id)
`citations`	In-text citations (`[12]`, `[3–5]`, `(Vaswani et al., 2017)`) mapped to references
`figures`	Figures, tables and algorithms with caption and region
`crossRefs`	“Fig. 3”, “Table 2”, “Section 4.1”, “Eq. (5)” mapped to their target
`activeSection`, `activePath`, `readingProgress`	Where the reader is

<Paper.Root provider={defaultCitationProvider()}>
	<!-- in each Viewer.Page -->
	<Paper.Layer />
	<Paper.Headings />
	<!-- anywhere inside -->
	<Paper.CitationCard />
	<Paper.CrossRefPreview />
	<Paper.References />
	<Paper.Figures />
</Paper.Root>

<Paper.Root provider={defaultCitationProvider()}>
	<!-- in each Viewer.Page -->
	<Paper.Layer />
	<Paper.Headings />
	<!-- anywhere inside -->
	<Paper.CitationCard />
	<Paper.CrossRefPreview />
	<Paper.References />
	<Paper.Figures />
</Paper.Root>

Tables of contents

Part	Shows
`Toc.Tree`	Nested sections, active one highlighted
`Toc.Flat`	Numbered list with page numbers
`Toc.Breadcrumb`	Current section path, for a toolbar
`Toc.Progress`	Reading progress bar with one segment per section
`Toc.Rail`	Dots along the document height; hover for titles

Metadata providers

Everything works offline; a provider only enriches references (abstract, TL;DR, citation count, links). defaultCitationProvider(options) chains OpenAlex and Semantic Scholar behind an in-memory cache, with an optional persistent cache adapter. Also available: openAlex(), semanticScholar(), crossref(), arxiv({ proxy }) (arXiv’s API has no CORS) and chainProviders() to build your own.

Rate limits

Without keys, public limits apply: Semantic Scholar is then only used for arXiv-only references and the chain stops once an abstract is found. Pass semanticScholarKey / openAlexKey for production use.

Analysis quality

The analysis is tested on a corpus of 47 papers (recent Hugging Face daily papers plus ACL, ICML, ICLR, CVPR, IEEE, Springer and ACM templates): 99.6% of captioned figures and tables are found with a region that includes their body. Known weak spots are text-only figures and algorithm boxes (their region stops at the caption).