We query an index. We do not host “all papers.”
There are hundreds of millions of scholarly works. Storing titles, DOIs, and abstracts ourselves would be a different product: storage, licenses, updates, and a crawler. conscRAG is a run over one target paper. For Find a paper and for retrieve-on-Run, we ask a live index and then forget the rest of the library.
OpenAlex is that index today. Their own docs call it “a fully open catalog of the global research system.” It has a documented works API, and it covers Crossref- and PubMed-class literature well enough to resolve a DOI, search a title, and often return an abstract. We send a mailto so they can contact us if traffic misbehaves. The browser never talks to OpenAlex directly; a small route on conscRAG does.
Why OpenAlex instead of starting with three vendors
Every extra provider is another key, another rate limit, another way abstracts can disagree. Semantic Scholar, Crossref, and PubMed each have strengths (citation intent, DOI registration, biomedical MeSH). They also overlap heavily with what OpenAlex already aggregates. Until OpenAlex misses abstracts we care about in daily use, more vendors are cost without a new capability.
Find a paper already understands DOI, arXiv id, and PMID, but it still resolves them through OpenAlex (for example pmid: on the filter, or the DataCite-style 10.48550/arXiv.… DOI). We did not add a second bibliographic API for those identifiers. If OpenAlex has no record, you see an empty result rather than a guess from another index.
What OpenAlex does not give us
Coverage is “almost all Crossref/PubMed-class papers,” not every PDF on the web. Some hits have no abstract; generate skips those. Some arXiv DOIs never landed in OpenAlex, especially older preprints whose canonical record is the journal version. We do not scrape publisher PDFs to fill the gap. That is a rights and product choice, not a missing import.
OpenAlex is also not a full-text index of every methods section. Retrieval quality follows titles and abstracts. If your field lives in workshop PDFs with weak metadata, the corpus will be thin. That is the main reason we might add another source later.
Will conscRAG add more sources?
Possibly, if OpenAlex consistently misses abstracts or identifiers that users actually drop. Candidates would be the obvious ones: Crossref for DOI metadata, PubMed for PMIDs, Semantic Scholar for extra abstracts. There is no scheduled second vendor and no promise of a date. Search will not start consuming a generate run just because we add a provider.
Until then, one index keeps the story simple: DOI, title, and abstract come from OpenAlex; the graph and findings are computed on conscRAG. If you need a paper OpenAlex cannot see, paste the title and abstract yourself. That path has always worked and does not depend on a vendor roadmap.
Questions
Will you add Semantic Scholar or PubMed’s own API?
Only if OpenAlex consistently misses abstracts or identifiers people actually drop. There is no scheduled second vendor.
What if OpenAlex has no record?
Find returns no hits rather than guessing. Paste the title and abstract yourself. That path has always worked.