chb-learner/paperpilot

PaperPilot is a command line research agent built for reviewing scientific literature in fields such as artificial intelligence, biomedicine, and AI for science. A user gives it one research request in plain language, and it builds a full workflow around that request: it writes a search plan with rules for what papers to include or exclude, queries many literature databases at once, removes duplicates, and sorts papers into core, adjacent, and excluded groups. It checks whether each paper's link, PDF, or code is actually reachable, without bypassing any paywalls. Once the screening is done, PaperPilot pulls together the evidence into a report, written in both Chinese and English, with numbered citations tied back to specific sources. Reports come out as Markdown, HTML, and PDF files, and can also be exported to Obsidian as linked notes covering papers, methods, topics, and claims. Every run is saved in its own folder with logs and intermediate files, so the process can be inspected or resumed later. The default free sources include arXiv, Semantic Scholar, OpenAlex, Crossref, OpenReview, PubMed, Europe PMC, bioRxiv, medRxiv, DBLP, ACL Anthology, and Papers.cool. Paid sources with their own API keys, such as IEEE Xplore, Springer Nature, Elsevier, Dimensions, and Lens.org, can be added for wider coverage. The tool also has an interactive shell with commands like /model to switch between language model profiles, /sources to check which databases are working, and /doctor to run a quick health check. There is also a lighter online demo running on Cloudflare Workers, which lets someone try a smaller version of the search and download a basic Markdown or HTML report from public paper metadata, without installing anything. The full command line version is the complete tool, meant for screened collections of papers, full text handling, and generating the bilingual PDF reports. To use PaperPilot, you install it with pip and connect it to any OpenAI-compatible language model by editing a configuration file that is created automatically the first time it runs. The project is written in Python and is aimed at researchers who want a repeatable, traceable way to build literature reviews instead of doing it by hand.