Agent mentions feed

Per-day counts of AI agents checking and reading a paper through naulon, as open CC0 data.

How often AI agents checked or read a paper through naulon, per day, for any paper whose page is public. The data is CC0: use it for anything, no attribution required.

GET https://gate.naulon.app/_naulon/v1/mentions/{scheme}/{value}

scheme is doi, pmid, pmcid or arxiv, and value is the identifier with its own slashes, each path segment URL-encoded:

curl https://gate.naulon.app/_naulon/v1/mentions/doi/10.1371/journal.pone.0000000

{
  "work": "doi:10.1371/journal.pone.0000000",
  "licence": "CC0-1.0",
  "k": { "events": 5, "days": 2 },
  "days": [
    { "day": "2026-10-13", "checks": 3, "reads": 0 },
    { "day": "2026-10-14", "checks": 2, "reads": 1 }
  ]
}

Field	Meaning
`work`	The key the paper is known by. A DOI comes first when the paper has one, so a PMID request may answer with the DOI.
`days[].checks`	Times an agent checked a citation to this paper with naulon's free citation check.
`days[].reads`	Times an agent read the paper's text through naulon, with a receipt.
`k`	The threshold: a paper is public once it has 5 checks and reads on at least 2 UTC days.

What you get back

200 for a paper past the threshold.
404 {"error":"not_found"} for a paper we do not know, a paper below the threshold, and a paper whose page was taken down. The three look the same on purpose.

Responses are cached for five minutes, so a count can be up to five minutes behind.

They count agents that used naulon, not every agent on the web. No count names a caller or holds the text of a question. Each paper's human-readable page is at https://naulon.app/papers/{scheme}/{value}, and the page data as JSON is at https://gate.naulon.app/_naulon/v1/works/{scheme}/{value}.