One product, built properly.

Why a research hub

AI made drafting easy and verification hard. A model will produce a confident paragraph about anything, and the cost of checking it now routinely exceeds the cost of writing it yourself. For anyone whose name goes on the work, that is a bad trade — and nobody else was fixing it.

The Hub is built on a sharper premise: useful research AI is defined by what it is not allowed to read. Gather the sources, keep them in a project, let the assistant work strictly inside that boundary — then record everything it did, so the work can be checked a year later by someone who was not there. Every other feature follows from that one decision.

How we work

• Invitation only, on purpose. A small number of researchers, close feedback, and features built for real work rather than a launch.
• Observability first. The platform measures its own cost and behaviour to the token — because a research tool that cannot account for itself has no business asking you to trust it.
• Documentation to the same standard as the product. A full help centre, plus an assistant that answers from it, so the answer is one question away rather than one support ticket.
• Evidence over adjectives. Every claim on this site describes something the Hub does today, in production, for people using it.