Build the literature vault
Search open sources, read with line anchors, run the method, build the matrix.
In short
The vault finds papers from open scholarly sources, lets you accept or reject each one, and then makes you read them properly: every claim is anchored to the exact line it came from. A method run writes cited notes and a matrix, and drops any claim it cannot quote.
1Let it suggest papers
After a verdict the vault can suggest up to 20 ranked candidates, each with a reason and the question it addresses. Accept the ones you want; only accepted papers are downloaded, and only from a legal open-access copy.
2Upload what it cannot fetch
A paper with no open-access copy is listed under "Needs manual upload", with link-outs to ResearchGate and Google Scholar and an upload button.
3Read and anchor
Open a paper to read it continuously. Select text to cite it, ask the AI about it, or make a note. Citations point back to the exact line.
4Run the literature method
The method plans researchers (facet × viewpoint), searches and fetches, then writes notes per key question. Each finding must quote the paper word for word; a quantitative or major claim without a quote is dropped with the reason.
5Fill the matrix
The matrix has one row per paper and a theme × source pivot. Your edits win and keep their history; you can export CSV.
For technical users
Sources: OpenAlex, Crossref, Semantic Scholar, arXiv, Europe PMC (including PubMed), Zenodo and OpenAIRE Graph; CORE with a key; Unpaywall with an email. dblp and bioRxiv/medRxiv are link-out or DOI lookup only. Requests are rate-limited per provider with a hashed disk cache and back-off; an offline or bot-check reason never carries a URL. De-duplication is by DOI, then title + year + first-author surname. Metadata is backfilled by DOI, else by title search; an ambiguous match is shown to you to pick or reject. A literature verdict reports per-question coverage and can send you back to the vault.