The chemistry build.
2,486 modules in build covering computational chemistry end to end: docking, molecular dynamics, reaction handling, formulation, and ADMET. Each module exposes a machine-readable interface for AI-agent operation and is verified against its reference engine.
Workflow
What a run looks like.
Every capability below is one entry in the catalog, one record in the build graph, and one call from the model you already use.
Bring the molecule
A SMILES string, a structure file, a vendor catalog number, or a sketch. The corpus reads the formats chemists actually keep.
State the question
Binding affinity, tautomers, route length. The answer is a chain of catalog entries, composed by the model that is driving, not a fixed form.
Run the engines
Docking, dynamics, descriptors, QC, ADMET. Each module is AI-native and checked against the reference implementation it stands in for; each has a public record page.
Read the result
Predictions with confidence ranges, figures ready for review, and the record that documents how each value was produced. Status: in development
Release standard, applied to chemistry
Verified before it releases.
A docking score that drifts is worse than no docking score. Where a reference engine exists, our implementation is tested against what that engine actually produces, under documented conditions. The record ships with the tool or the tool does not ship.
Every chemistry entry, packaged for desktop systems and distributed through standard app stores.
The same entries, operated by Claude, Codex, or your own agents.
Your cheminformatics stack installed within your perimeter, on your hardware, under your policies.
The chemistry catalog is public.
2,486 modules in build, each with a public record page. Release proceeds module by module as verification completes.