Features
Search and import
Search the databases. Log every search.
Database search
OpenAlex, Semantic Scholar, arXiv, DBLP, Crossref, Europe PMC, PubMed, CORE and Springer Nature through their official APIs, and IEEE Xplore with your own free key. Your search string is written in each database's syntax and counted before anything is retrieved.
A search log for PRISMA-S
Every search, run here or imported, keeps its database, interface, date and time, exact query, limits and counts. Run it again later and only what is new is added.
Six import formats
RIS, PubMed/MEDLINE, Web of Science, BibTeX (citation keys kept), CSV and EndNote XML, for the databases you search yourself, with the query and the export steps.
Citation searching
Backward and forward from your included studies, screened and counted under other methods in the flow diagram.
Duplicates
Found by DOI, PMID and normalised titles; exact groups resolve in one click.
Screening
Independent votes, then a discussion.
Blinded dual screening
Reviewers vote without seeing each other, with keyboard shortcuts, highlighted keywords, reasons, notes and labels.
Conflict discussions
Each disagreement has a thread. AI co-authors argue from the record or the PDF and may change their vote.
Cohen’s κ
Agreement for human pairs and AI pairs separately; a vote changed in a discussion counts with its value before it.
AI screening modes
Off, advisory, second reviewer, or auto-exclude above a confidence threshold. Sensitivity, specificity and κ against the human decisions are reported.
Full text
Find the PDFs, assess every report.
Open-access retrieval
Every legal open-access copy that OpenAlex, Unpaywall, Semantic Scholar, arXiv, Europe PMC, CORE, IEEE Xplore and the ACM Digital Library know of, and a link that opens the rest through your library. Never shadow libraries.
Assessment with reasons
Each report is judged against every criterion; exclusions need a reason, and reports not retrieved are counted.
Extraction and related work
Data with its evidence. Surveys with a verdict.
Extraction forms
From templates, by hand, or proposed by the AI from your research questions. Manual, AI pre-fill with quoted evidence, or automatic.
Consensus
With several extractors, a compare view produces the final values that feed the charts, tables and sections.
Related-work table
Prior surveys rated on your review’s dimensions; the AI can rate them and an author’s edit marks a rating as verified.
Manuscript
A paper written from the review’s own data.
Sections from your data and notes
Each section asks for the authors’ points and receives the extracted data, never the AI’s memory. Lock what you like; restore earlier versions.
Your writing style
Rules, banned words and example passages the AI matches; a style check flags what to fix.
IEEE Access, IEEE, ACM, APA
The LaTeX project with the PRISMA diagram, charts, taxonomy, tables, appendices and references, built to PDF.
Every page checked
Nothing outside the margins, no table too wide, no unresolved reference; the AI can look at each page image as well.
Reporting
Numbers that add up, and a record of who did what.
PRISMA 2020
The flow diagram is computed from the records, so its counts always reconcile, with drafted methods text and the 27-item checklist.
Audit trail
Every decision is stored with who made it and why; the AI-use statement counts only work a model actually did.
What the AI used
For each part of the workflow: jobs, AI calls, tokens read and written, and the cost.
Exports
CSV, RIS and BibTeX including extraction data, the LaTeX project as a ZIP, and the decision log.
AI that shows its work
AI in the workflow
AI can screen as an independent reviewer, propose the extraction form, extract data from full texts, rate prior surveys, draft sections and look at every page of the built PDF. Each AI output is recorded with its model and prompt, and the paper states exactly what AI did and what people verified. AI tools are never listed as authors.
- Every AI judgement is stored with its model, prompt, confidence and rationale.
- Agreement between AI and people is reported as separate AI pairs, never mixed in.
- Each review can pin the model its protocol names, so a new default never changes what it reports.
- The AI-use statement counts only work a model actually did.
Straight talk
What it does not do
Search databases against their terms. Google Scholar, the ACM Digital Library, Web of Science and ScienceDirect do not allow searching by software, so you search them with the query written for you and import the export. It never scrapes.
Fetch papers from shadow libraries. Only legal open-access copies, and a link that opens the rest through your own library. It never downloads through a subscription.
List AI as an author. AI co-authors are tools with personas; people are the authors.
Promise acceptance. It gets the reporting right; the science and the revision are yours.