STOPHY_API_KEY. See the Quickstart.
1. Search for papers
This request asks for up to 3 papers from 2024 on, newest first.results holds papers with paperId, primaryId (the best outside ID), title, abstract, authors, year, venue, citedByCount and openAccessUrl. The index can return a paper with no abstract, so do not rely on it for claims. openAccessUrl may be a PDF or a landing page. Keep the papers whose link ends in .pdf or points at arxiv.org/pdf. See Papers for options.
2. Read a paper
Scrape reads a direct PDF link as Markdown. This request reads the first 2 pages and cuts the text at 2,000 characters.totalPages is the pages in the file. Send the same request with "offset": 1969 to read on, a larger maxPages for more pages, and pageMarkers: true to see where each page starts. A PDF that comes back empty may need "mode": "ocr" in the PDF options. See PDFs and documents.
3. Read every paper
Repeat step 2 for each result with a direct PDF link. Keep the paper’sprimaryId as id, with its title, year and Markdown as text. Hand each note to a model with the instruction to quote only from text and cite id. For a paper you hold as a file, Parse gives the same Markdown from an upload.
Check it
resultsmatches your topic. If not, change the query, or addauthors,fromandto.- Each note contains the paper’s title. A PDF can open with publication details before it.
- Read on with
nextOffsetwhen the answer needs the methods or results. - Every claim in your summary is a quote you can find in
text. Check three.