Help

The road, step by step

1. Config Write the review question, the research questions and the eligibility criteria (one inclusion or exclusion criterion per line), and set how many reviewers screen each record.
2. Import Write the search string once and let the app search the open databases (OpenAlex, Semantic Scholar, arXiv, DBLP and more, IEEE Xplore with your free key); search the others with the query it writes and upload their exports (RIS, BibTeX, CSV or PubMed format). Every search is logged with its query and date, so it can be reported and run again. Citation searching from the included studies is here too.
3. Duplicates Exact duplicates are merged automatically; check the likely ones, such as a preprint and its published version.
4. Title/abstract Each reviewer votes without seeing the others. Keyword highlights and the AI suggestion, when switched on, help but do not decide.
5. Conflicts Records on which reviewers disagree are discussed and resolved by an owner, or by the rules of the autopilot.
6. Full text Open-access PDFs are found automatically; upload the others. Each report is then assessed against every criterion, and exclusions need a reason.
7. Extraction Set up the form (or let the AI propose one from your research questions), then extract the data of each included study. Values the AI extracts come with the passage they rest on.
8. Related work Prior surveys are searched for, rated from their full texts and compared with your review in a table.
9. Manuscript Fill in the paper details, write each section with AI or by hand, and build the PDF. Every page of the build is checked.
10. PRISMA The flow diagram and the methods text are generated from the records; they always match the data.
11. Checklist The PRISMA 2020 checklist, filled in with the pages where each item is reported.
12. Export The paper, the LaTeX project, the data and the decision log, for the journal and for your records.

Good to know

• Autopilot

  The autopilot on the Overview tab runs the whole review with two AI reviewers and no human step, from the imported records to the PDF. It can be stopped and resumed at any time and never repeats finished work. What it decided, and by which rule, is reported in the paper as AI work that no person has verified yet.

• AI models

  Each review, task and AI reviewer can use its own model. Two AI reviewers on two different models read each record independently, which makes their agreement meaningful. The AI tab of a review shows every job with its model, tokens and cost.

• Who can do what

  Owners set up the review, manage the team and resolve conflicts. Co-authors screen, extract and write. Reviewers screen and extract. A locked review can be read but not changed.

• Good to know

  The question mark next to a setting explains it. Counts refresh while a job runs. Nothing is lost when a job stops: starting it again continues where it ended.