About wovepaper

What it does

Every arXiv record is harvested through the official OAI-PMH metadata feed and stored as a paper: title, abstract, authors, categories, dates and identifiers. Titles and abstracts are then embedded as vectors, which is what makes the search semantic — you can describe an idea in your own words and find papers that never use them.

Author strings are resolved into disambiguated researcher profiles, so a name that belongs to four different people becomes four profiles rather than one confused list, and each profile carries its own publication history, co-authors, topics and citation counts. Institutions are linked the same way, through registry identifiers rather than affiliation strings, so a university is one page instead of forty spellings.

• Semantic search over the corpus, with filters for category, field, year and topic.
• Researcher profiles with co-authorship, topic and activity breakdowns.
• Institution pages with their researchers, collaborators and subject mix.
• Per-paper summaries, topics and key contributions, generated on first open.
• A signed-in assistant that answers across papers and cites the section it used.

Where the data comes from

Paper metadata and abstracts come from arXiv under its terms of use; wovepaper links to arXiv for the canonical abstract and PDF rather than rehosting them. Researcher records and citation counts come from Semantic Scholar (ODC-BY) and OpenAlex (CC0), which is also where institution identifiers come from.

Summaries, topic tags and assistant answers are generated by large language models from the paper's own text. They are a reading aid and can be wrong: the abstract, the PDF and the citation trail on each paper page are the record, and every paper page links to the arXiv original so a claim can be checked at the source.

What is free and what needs an account

Browsing, searching, paper pages, researcher profiles and institution pages are public and need no account — no paywall, no metering, and no sign-in wall in front of the corpus. That also holds for automated readers: the public read API is open, and the crawl rules welcome agents.

An account exists for the parts that are personal or that cost money to run: a library of saved papers, followed researchers and topics, and the assistant that reads papers and answers questions. Sign-in is by emailed magic link or a short code — there is no password to choose or lose.

Coverage and honest limits

The archive is harvested oldest-first and continuously, so the newest submissions can take a short while to appear, and a paper published minutes ago may not be searchable yet. Citation and institution coverage is inherited from the upstream sources: a large minority of arXiv works carry no institution in any registry, so an institution page shows the papers that could be linked, not every paper the institution ever produced.

Model-written summaries are produced lazily — the first signed-in reader to open an un-indexed paper triggers one — so a paper nobody has opened yet shows its abstract instead of a summary. That is deliberate: generating them for three million papers upfront would cost more than this project has.

Who runs it

wovepaper is built and operated by Mike van Egmond, an independent developer in the Netherlands. It is a personal project rather than a company, funded out of pocket, with no advertising and no data resale.

Corrections, disambiguation problems and takedown requests are read by a person: see Contact. Machine-readable pointers for agents live in llms.txt, and the public JSON API is at https://api.wovepaper.com.