# ArcXiv

A fast, metadata-first research-paper search engine. ArcXiv indexes the arXiv
corpus, ranks papers with editorial quality assessments, and serves
agent-readable documents. PDFs are hosted by arXiv, not ArcXiv.

Human-readable documentation: https://arcxiv.org/docs/agents

## Search

- GET /api/agent/search?q={query}
  Ranked search across titles, authors, and abstracts. Parameters:
  q (query text or an arXiv ID/URL), category (an arXiv category such as
  cs.LG), scope (all|title|author|abstract), sort (relevance|newest|oldest),
  fromYear, page, pageSize (1-100, default 20). Returns hits with full
  abstracts, qualitySignals (editorial assessment metrics and category
  percentile), and links. Optional: restrict with category alone.

## Papers

- GET /api/agent/papers/{arxivId}
  Full metadata for one paper: abstract, authors, categories, version
  history, quality signals, related papers, and links.
- GET /abs/{arxivId}/markdown
  The paper as a markdown document: metadata, abstract, and canonical
  arXiv links. Converted full text is reserved for a paid tier and is
  not enabled by default.

## Notes

- Responses are JSON unless stated otherwise and are safe to cache.
- Public read traffic is limited to 300 requests per five minutes per IP.
- arxivId accepts new-style (2401.12345) and old-style (cs/0112017) IDs.
- Quality signals come from ArcXiv's independent editorial assessments;
  percentile is relative to the paper's primary category over 30 days.
- Creative Commons papers use arXiv HTML for best-effort full-text markdown;
  other licenses return metadata and abstract with canonical arXiv links.
- Search and paper responses may include sixteen independent research-signal
  dimensions. Definitions and limitations are documented at
  https://arcxiv.org/methodology/research-signals.