Below, an actual screenshot of the desktop app: full-text screening, with the local AI's own suggestion, a leaning verdict and a literal quote from the PDF, offset-verified, right next to your decision.
A real screenshot of the app's own local AI reader, explore it yourself once you're in. Design continues to evolve.
Compendium guides you to a submission-ready forest plot and a PRISMA diagram, with every number traceable to its source. A finished synthesis, not a pile of tabs.
Compendium starts with systematic reviews and meta-analysis, built around gold-standard reporting guidelines. More study types are on the roadmap.
Randomized and interventional study designs
Cohort, case-control and cross-sectional designs
Structured reporting of individual clinical cases
Studies that evaluate tests and diagnostic tools
Planning and registering research before it starts
Development and validation of predictive models
Cost-effectiveness and health-economics studies
Interview, focus-group and mixed-methods studies
Each will be built around the gold-standard reporting guideline for its design, with the same guided, privacy-first, human-in-the-loop approach. None of these are available yet. Email us to help decide the order.
Early access and the launch price for the first researchers who reach out. Help shape the roadmap.
No. You stay the reviewer of record. The AI drafts, screens and shows reviewer agreement (κ), but every call is yours, with transparent AI-use disclosure.
By default, nothing leaves your machine: open local models only. The one optional exception is per-review sync to a Turso database you configure yourself, off until you turn it on for a specific review, ideal for sensitive data, ethics boards and LGPD/GDPR. Export to PDF/DOCX anytime, no lock-in.
Every AI suggestion is anchored to a verbatim quote from the source document, not a paraphrase, so you can check it against the original in one click. The AI abstains instead of guessing when the text doesn't support a confident answer. Nothing leaves your machine by default: no cloud AI, no cloud account, with the optional per-review sync above as the only exception. And the statistics are validated, not just executed: effect sizes and CIs match the metafor R package to the 4th decimal.
Today: systematic reviews and meta-analysis (PRISMA), end to end. More study types are on the roadmap.
We're upfront: early access, no paying customers yet. The core workflow is real and usable, but you'd be an early adopter, not buying a finished platform.
Yes. Built for your first study, with guided steps so PRISMA and RoB 2 come built into the flow, not something you assemble yourself.
For researchers in Brazil, the US and the UK, yes: a July 2026 legal review found no blocker for the current free early-access program in any of the three. The early-access program is local-first, runs without a backend and, by default, collects no personal data beyond an optional support email. That's not a compliance certificate, just no blocker found so far. Somewhere else? Email us and we'll look at your case.