Two jobs, done end-to-end.
Build the bibliography of your next paper. Write a review of the literature across multiple papers. Every citation in either output is backed by a verbatim quote from a real source: never invented, never hallucinated.
Build the bibliography of your own paper.
A bibliography is the structured set of citations behind an experimental paper, a thesis chapter, or a grant introduction. Each reference plays a specific role: setting the big-picture context, anchoring established science, framing the gap you fill, justifying your methods. The Research Bibliographer agent turns your draft abstract into that structured set, and validates every entry against the source.
Paste the abstract, project aims, or thesis introduction you've drafted. Your own words define what the paper claims.
We identify 6–14 atomic claims your introduction makes: big-picture context, established science, the gap you fill, the methods you use.
For each claim we search PubMed, propose 3–5 candidate papers, retrieve a verbatim quote from each, and validate the quote against the source.
We check coverage: no missing methodological references, no over-cited subtopic. You confirm picks; we never silently add.
BibTeX, RIS, APA, or Vancouver. Every entry carries the verified quote and the role it plays in your paper.
Write a review across multiple papers.
A scientific review surveys what is known about a question across the existing literature, draws conclusions, and surfaces gaps. The Literature Reviewer agent does this end-to-end: from your seed papers and your research question, it builds a thematic bibliography, extracts verified quotes from every paper, and drafts a Markdown review you can hand to a co-author.
You phrase the research question and pick 3–5 anchor papers you trust. The corpus we'll build is the citation neighbourhood of those seeds.
We walk references and cited-by from each seed until the working corpus hits 40–60 papers, or 60–100+ for a thesis chapter.
Confidence scoring plus obsolescence detection flag retracted, low-confidence, and superseded papers before they reach your review.
We retrieve the full text, search inside each paper for passages relevant to your question, and extract verbatim quotes. Every quote runs through a three-method verification before it lands in your bibliography.
Verified quotes get clustered into thematic groups. We write a gap report, then draft the Markdown review with inline citations to every claim.
No quote, no claim.
Both workflows share one rule: every citation that lands in your output is paired with a verbatim quote from the source paper, and every quote runs through a three-method verifier before it gets persisted. When the literature doesn't support a claim, the agent says so. It never invents a PMID, an author, or a finding. You can audit any entry back to its source in two clicks.
Run an example on a real paper.
Paste a paragraph from a paper you know well into the Paragraph Grounder demo. Watch the agent decompose, search, retrieve, and verify. No signup. No fake PMIDs at the end.