Matter, on the record.
Every material is a hypothesis about atoms. 2,400 modules in build test those hypotheses: structures predicted, properties computed, candidates screened, and every number checked against the engines researchers already trust.
The working loop
Four passes, over and over.
Materials discovery is not a single tool but a discipline of loops. The corpus runs the loop; the model picks the next candidate; you keep the judgment.
Propose
Generative models draft structures the periodic table permits and intuition would miss. Crystals, surfaces, compositions.
Compute
DFT where it counts, learned potentials where it can. Formation energies, band gaps, elastic tensors, phonons: the numbers that decide.
Screen
Thousands of candidates against stability windows, property targets, and synthesis reality. The survivors are few and named.
Feed back
Every run writes its record. The next proposal starts smarter. status: coming soon, all of it
What the corpus knows how to compute
The properties that decide.
Every materials entry, packaged for ordinary machines. DFT post-processing that once requisitioned a cluster can wait for a laptop.
The same corpus, driven by Claude, Codex, or your own agents, running whole loops while you sleep.
Your lab's substrate, on your metal, behind your gate. The records stay where the work happens.
Draw the material first. Build it second.
2,400 modules in build behind 122 written records. The blueprint gets published before anything casts a shadow.