Open-source AI paper tool checks citations against full texts
Individual developer Maoshu (GitHub user mmlong818) released Open Academic Paper Gen, an open-source multi-agent workflow that drafts papers in nine stages: scoping the research area, literature search, cleaning and filtering, trend analysis, finding an angle, outlining, writing, verification and export. It searches OpenAlex, Crossref, Semantic Scholar and arXiv, then expands the pool by following references and citing papers. Its distinguishing feature is three layers of citation checking: whether a citation exists in the collected pool, whether the paper's identity matches Crossref or doi.org records, and whether the full text actually supports the claim; when full text is available it locates the relevant passage and, if page data exists, flags the PDF page, falling back to abstracts or excerpts otherwise, and marking insufficient evidence as "undetermined" rather than passing it. The default mode pauses at major stages for human review, with a fully automated option also available; the project supports OpenAI and Zhipu GLM, though claim verification, uncited-statement checks and simulated review still require an OpenAI API key, and exports cover Markdown, LaTeX, RIS and BibTeX.
As many AI writing tools focus only on generation, pushing citation verification down to the source-passage level gives researchers a more traceable way to check claims.