The future here lies in the past—
and in how well we understand it.

In 1543, in Padua, a young Flemish anatomist named Andreas Vesalius opened a body and drew what he saw. For thirteen centuries, the study of the human body had run on the authority of Galen—a 2nd-century physician who, forbidden from dissecting humans, had built his anatomy from Barbary apes. Vesalius read Galen. Then he looked. The discrepancies he found, he rendered as plates—public, beautiful, peer-reviewable artifacts anyone with a copy of De humani corporis fabrica could verify. The book did not argue. It simply showed. That gesture is the modern scientific method, compressed.

What is humanovo? It is the next entry in that conversation.

Five hundred years on, the body of knowledge has changed scale. There are thirty-six million papers indexed in PubMed. There are 200,000 new ones every month. A working scientist, in a working life, can read perhaps 10,000 of them. The remaining 99.97% sit in a library no one will ever finish. Some of those papers contain the answer to the question you are asking right now. You will not find them by searching for them, because you do not know to search for them. They were written in a different decade, in a different field, in a different vocabulary. They are the papers Vesalius would have wanted— and they are unreadable at human bandwidth.

Read everything. Look yourself. Render the discrepancies. That is what humanovo is for. We read every paper that has ever been published in your field, and the adjacent fields, and the fields you do not yet know are adjacent. We hold them as one corpus in one model, alongside your own notebooks and your lab’s data, and we surface the thread of thought you were already pulling on—weighted by evidence, grounded in citations that exist, contradicted where the literature contradicts itself. We do not give you the answer. We give you the next experiment.

We are aware of the company we keep. There are LLMs that will write you confident-sounding paragraphs about biology with citations to papers that do not exist. There are literature-search tools that return the same twenty-five papers everyone else has already read. There are notes apps that let you organise what you have already written down. We are not any of these. The pipeline behind every humanovo hypothesis is twelve adversarial stages: generation, evidence grounding, mechanism extraction, contradiction-search, counter-argument, revision, and so on—each stage scored, every claim round-tripped through CrossRef and NCBI before it lands on your screen. If a citation does not resolve to a real paper with the claim we attached to it, you do not see it. The literature-of-record is our peer reviewer.

And we are sceptics by training. We do not train models on your work. We do not share your data with the LLM vendors we use. We acknowledge, in the colophon, that the foundational anatomical plates we love were possible only because of dissections often performed without the consent of the families involved. Renaissance science had ethical failures we have inherited a duty to repair; our own version of that duty is that the work we ship be auditable, that our methodology be inspectable, and that the thread of provenance from claim to citation be something a graduate student in 2126 could still follow.

What we are reaching for, then, is not “AI for biology.” It is the next chapter of the same impulse Vesalius started: to know the body of living knowledge, render its discrepancies, and ask— in the company of every researcher who has come before—what does this make possible now?

If that is the question your work is trying to answer, humanovo was made for you.