Questions, answered plainly

They read some sources and then answer in their own words, with citations bolted on at the end — the model is still the thing deciding what's true. QuarterMill flips that. The connections between sources, and the record of what's missing, live in a layer beneath the model, so every claim stays tied to the exact source it came from, and when the record has nothing, it says so instead of answering anyway. They're built to always respond. We're built to be checkable.

Most AI is good at sounding right but can't show its work. QuarterMill pairs the part that reads and understands language with a second part that checks every statement against the actual sources and rules. One reads; the other verifies. That's what makes the answers traceable.

When the record has no answer, most AI invents a plausible one. We don't. If the record is silent, we say so, instead of filling the gap.

Documents, recordings, and records in almost any format, printed text, handwriting, and audio, across many languages. It works on top of the collections you already hold.

Yes. The knowledge stays under the control of whoever holds it, and can run inside your own systems. Records never have to leave your control.

Anyone whose answer has to be right and has to be checkable, researchers, historians, legal and public-interest teams, scientists and clinicians, journalists, civic institutions, and the public.

V1 shipped in May 2026. You can request early access to work with us on what comes next.