Cento exists for one reason: in medical research, a citation has to be true, and a general-purpose AI cannot promise that. We built a co-writer where that promise is the whole product.
A language model predicts the next plausible token, and a reference is just more tokens to predict. So it will produce a perfectly formatted author list, journal, year and DOI for a paper that was never written. In most settings that's a nuisance. In medicine, a single fabricated citation is a retraction risk — and the person whose name is on the paper is the one who pays for it.
Most AI writing tools optimize for polish and speed, then bolt on a citation check. We inverted that. For Cento, citation truthfulness isn't a feature — it's the entire point. The model can only cite papers Cento actually retrieved, every reference opens, every claim is graded against its source, and drift is caught as you revise. The writing serves the evidence, never the other way around.
"The model cannot emit a citation that didn't come from retrieval."
We're beginning in ophthalmology — a field with a deep, well-structured literature and clinicians who read methods sections for a living. It's a proving ground, not a limit. Getting verification right for a demanding specialty first is how we earn the right to expand. Early access is open to clinician-researchers now; the people who publish under their own name will decide whether we've earned theirs.
Early access for clinician-researchers, starting in ophthalmology.