Where Litlas gets its papers — and what it does not cover
The bibliographic sources behind the Litlas index, the size of the corpus, how often it is rebuilt, the gaps we know about, and how to verify all of it yourself.
Litlas does not crawl the web. Every paper in the index arrives from a named bibliographic source, and every search result carries the slug of the source it came from. This page lists those sources, states the size of the corpus, explains how often it is rebuilt, and — the part that matters most if you are deciding whether to recommend Litlas — spells out what it does not cover.
Measured on 2026-08-21 against the live API. The release being served was built on 2026-08-11.
Evaluating Litlas for a library research guide? What a student needs in order to use it, what it costs, and a description you can paste into your guide are on a page of their own.
Where the papers come from
The backend declares 16 connectors, and that is the number the overview endpoint reports as “sources”. Be careful with it: it is the length of a list in the source code, not a measurement of the index. Measured against the live index, 10 of those connectors actually hold papers, and only 2 of them carry the corpus.
OpenAlex
openalex
Paper metadata and canonical identifiers
Papers in the index: Above the count cap
DBLP
dblp
Computer-science bibliography
Papers in the index: Above the count cap
ACL Anthology
acl_anthology
Conference proceedings
Papers in the index: 1,000
Crossref
crossref
Paper metadata and canonical identifiers
Papers in the index: 992
OpenAIRE Graph
openaire_graph
Open-access availability and aggregation
Papers in the index: 764
arXiv
arxiv
Preprints
Papers in the index: 178
PubMed
pubmed
Biomedical literature
Papers in the index: 100
PMC Open Access Subset
pmc_open_access
Biomedical literature
Papers in the index: 100
DataCite
datacite
Paper metadata and canonical identifiers
Papers in the index: 60
Common Crawl
common_crawl
Web-scale URL discovery
Papers in the index: 1
OpenCitations
opencitations
Citation edges (enrichment only)
Papers in the index: None
ROR / ORCID
ror_orcid
Author and organisation identifiers (enrichment only)
Papers in the index: None
OpenReview
openreview
Conference proceedings
Papers in the index: None
DOAJ
doaj
Open-access availability and aggregation
Papers in the index: None
CORE
core
Open-access availability and aggregation
Papers in the index: None
Unpaywall
unpaywall
Open-access availability and aggregation
Papers in the index: None
“Above the count cap” means the search API stops counting at 50,000, so no exact per-source figure can be obtained from outside. It does not mean the source is complete. Rows showing a small number are exact — and they are samples, not corpora: a connector that contributes 1,000 records per rebuild is a taste of that database, not a mirror of it.
The rows marked “enrichment only” are working as designed: they contribute identifiers, citation edges, or open-access links to papers that came from somewhere else, so they hold no papers of their own. The remaining empty rows are connectors that are declared but are contributing nothing to the release currently served.
One source is used but not listed above: Semantic Scholar is queried out of band to raise citation counts and add citation edges. It contributes no records to the corpus, so it has no slug you can filter on.
How big it is
19,379,997
papers
413,996,658
citation links
Both figures come from the same rebuild, so they describe the same corpus. The paper count is live — the API reports it on every request. The citation-link count is not exposed by any endpoint; it is the edge count recorded by the build that produced the release now being served, which is why it is dated rather than live.
The home page rounds the citation figure down to 390,000,000. That is deliberate: the rounded claim is checked against the measured value, so it can never overstate it.
How often it updates
We do not publish a refresh schedule, because there is not one. The index is rebuilt by running the ingest pipeline in full — there is no incremental or delta path — and each rebuild is started by hand rather than by a scheduler. A full rebuild takes on the order of a day to complete.
The practical consequence for you: a paper published or cited after the build date above will not be in the index, and citation counts are as of that date, not today. If you need same-week coverage, this is not the tool for that job.
What Litlas does not cover
This is the section we would want to read first if we were evaluating someone else’s tool, so it is deliberately blunt. Each item below is a limitation we can point at in our own code, not a hedge.
Search does not read abstracts, and there is no full text
Free-text queries match titles, author names, DOIs, and venue names — nothing else. Abstracts are shown on a result but are not searchable, and no full text is stored or indexed anywhere in the system. A phrase that appears only in the body of a paper will not find that paper.
Coverage outside computer science and biomedicine is citation-gated
Papers in computer science and the biomedical fields are kept from 10 citations upward. Everything else has to clear 200 citations to enter the index at all. Recent, specialised, or low-citation work outside those two areas is therefore largely absent — not because it was ranked low, but because it was never ingested.
Most of the listed sources are samples, not mirrors
Apart from the two at-scale sources, each connector is capped at roughly 1,000 records per rebuild and is seeded from a fixed query. PubMed appearing in the table above does not mean Litlas has PubMed coverage; it means a few dozen to a few hundred PubMed records are present. Treat those rows as provenance, not as reach.
Japanese-language literature has no dedicated source
There is no connector for CiNii, J-STAGE, or any other Japanese bibliographic database. Japanese papers appear only when an international source such as OpenAlex or Crossref already carries them, which skews coverage toward work that was published or indexed in English.
Citation counts are not comparable with Scopus or Web of Science
A displayed citation count is the largest of three values: the count OpenAlex reports, the number of citing papers inside this corpus, and a floor taken from external assertions. Because the second term can only ever see papers that are themselves in the index, these numbers are not a bibliometric measurement and should not be used as one.
Result totals stop counting at the cap
The search API stops counting matches at 50,000. A result total showing that number means “at least that many”, never “exactly that many”. Do not read it as a corpus statistic.
How to check this yourself
None of the above needs to be taken on trust. The endpoints are public and answer without a login, so you can reproduce the table on this page in a few minutes.
See the source of any result
Every object returned by the search API carries a source_slug field naming the database it came from.
GET /api/v1/search?query=attention — 各 result の source_slug
List the declared connectors
The sources endpoint returns every connector the backend declares, with its slug, role, and description. Its count is the 16 figure discussed above.
GET /api/v1/sources
Read the live corpus size
The overview endpoint reports the current paper count straight from the search index.
GET /api/v1/overview
Count the papers behind any one source
Filtering search by a source slug with an empty query returns that source’s total, which is how the coverage column above was measured. The response also tells you whether the total hit the count cap.
GET /api/v1/search?query=&source_slug=dblp&limit=1 — total と total_capped
One number on this page cannot be checked this way: the citation-link count is not served by any endpoint, so it can only be quoted from the build record.