Why the default is still abstract-only
The default extract reads (head, relation, tail) from titles and abstracts. OpenAlex is a catalog: display_name, DOI, inverted abstract, and sometimes an open PDF link. Most journal PDFs are not licensed for a web app to download, parse, store, and send to a model. That is why the older posts said we do not scrape publisher files, and why that sentence is still true.
Abstracts are also faster. A run already talks to OpenAlex many times. Parsing every hop as a PDF would fight the 300-second cap. Methods and limitations that never made the abstract are the only good reason to opt in. See what DOI, title, and abstract mean.
What the checkbox actually does
You stay on conscRAG. The server reads OpenAlex open-access fields, fetches those URLs, and tries to extract text. Failures are silent: we keep the abstract. We cap the job to the target and base papers, not every reference hop. Figure 2 then marks abstract versus open PDF so you can see which works got the extra slice.
This is not a library. We will not upload your PDF for you, and we will not buy access. If a work matters and OpenAlex has no open file, paste a better abstract or open the PDF yourself in another tab. Why conscRAG uses OpenAlex is still the index story.
How to decide
Leave it off for a first pass. Turn it on when the target is a methods-heavy preprint (arXiv, PMC) and the abstracts in Figure 2 look too thin to support a limitation or a dataset name. Expect a slower run. Do not treat “open PDF” as a complete methods section. Extraction still compresses to short triples.
Questions
Will this open publisher paywalls?
No. We do not guess PDF URLs and we do not scrape publisher HTML. If OpenAlex has no oa_url or best_oa_location we can use, that paper stays abstract-only.
Do you store the PDF?
No. We do not keep the file. A saved run may remember that a work used an open PDF (viaPdf). Neighbor abstracts are still stripped on share.
Does Find a paper fetch PDFs?
No. Find only fills title, DOI, and abstract. PDF fetch happens on Run, and only with the checkbox on.