Research that can show its sources

Our mission

Generative models made it trivial to produce plausible scientific text and impossible to trust it. We build the opposite: a place where the tedious parts of research — searching nine databases, de-duplicating, screening, extracting, checking reporting guidelines, formatting for a journal — are done by software, and where every shortcut leaves a trail a reader can follow back to a page in a paper. Researchers should spend their time on judgement, not on plumbing, and never on checking whether a citation exists.

What we will not compromise

• Evidence before prose

  Every claim the assistant makes points at a quotation on a page. Every number points at a verified extraction or an analysis result. If it cannot find evidence, it says so.

• Authors stay authors

  The assistant proposes as tracked changes; people accept, reject and sign their work. Screening suggestions are votes, not decisions. Exports carry an honest AI-use statement.

• One place for a study

  Team, tasks, sources, evidence, screening, analyses, manuscripts and submissions share one project, one permission model and one activity log — from question to decision letter.

• Boring, strong security

  Two-factor codes, passkeys, institution single sign-on, encrypted provider keys held by administrators, audited access, and data you can export or delete.

Who it is for

Clinician-scientists writing between clinics, methodologists running systematic reviews with two screeners, statisticians who want their numbers bound to their analyses, students learning evidence-based methods, and research offices that need one secure workspace instead of a dozen subscriptions. It works in the browser today; desktop and mobile applications follow.