The substrate under serious science.
DarwinOmics is an AI-native working layer for biology, chemistry, and materials science: 7,666 tools, models, and pipelines in build, written so a machine can run them and a scientist can trust them. None of it has shipped yet. All of it is coming soon, on purpose.
substrate overview
in build
What the layer has to do
Four jobs. No substitutes.
Carry the right tools
Each one rebuilt AI-native in Rust: cheap for a model to call, cheap to read, and verified against what the original produced.
Know how to use them
Every entry ships with a machine-readable manifest. Claude, Codex, or a model you trained yesterday can pick one up and go.
Hold enough to wander
Corpus size is a feature. The tangential analysis, the run nobody thought to try, lives a few calls from the obvious one.
Stay grounded
A knowledge graph and a graph neural network sit under the corpus, for genomics, drug discovery, and repurposing first.
Three doors, one floor underneath
Pick your science. The substrate does not care.
Biology
It started here. Reads, variants, structure, cells: the working machinery of molecular biology, rebuilt for machines.
2,780 modules in build · 2,066 cataloged
Chemistry
The same logic, pointed at molecules. Docking to dynamics, reactions to formulations, spectra to verdicts.
2,486 modules in build · 1,147 cataloged
Materials science
The newest front. Crystals, batteries, and the properties that decide what gets built and what gets shelved.
2,400 modules in build · 122 cataloged
How a trading desk became a lab bench
The short version of a long build.
Power markets
The first substrate carried energy contracts, not proteins. Many tools, one layer, invisible when it works. The shape of the problem has not changed since.
One pipeline corpus
The plan was modest: rebuild nf-core so a researcher could run it conversationally. Then the map arrived: biology's tool universe made pipelines look like a harbor made to hold an ocean.
The wrapper collapse
Version one bolted AI onto containers and drowned in context. So we rewrote the tools themselves, in Rust, with interfaces drawn for the caller that never sleeps.
Three sciences
The chemistry and materials problems turned out to be the same problems, unchanged. One substrate logic, three domains, 7,666 modules and counting.
The one promise we make early
Nothing ships on a promise.
Checked against the original
Where a reference implementation exists, our output is verified against what it actually produces, under pinned, declared conditions. The claim is machine-checked, not marketing copy.
The record travels with the tool
Every entry carries a provenance record: what it was compared against, under which conditions, and what came out the other side. Released publicly, with everything else.
One bar for all 7,666
There is no fast lane and no showcase tier. A module releases when it clears the same gate every other module cleared, or it does not release.
Who does the driving
Any model can hold the wheel.
The tools speak a manifest, not a dialect. There is no DarwinOmics API you must adopt and no runtime you must marry. The model you already use is the interface.
Reads the manifest once, drives the corpus the same way it drives any instrument you hand it.
The same tools, the same surface, a different driver. The corpus does not care who is at the wheel.
Small or large, local or hosted. If it can call a function, it can run a aligner, fold a protein, or screen a battery material.
Raised on this corpus since birth. They will be open-sourced, and they are coming soon.
Three ways in
One of them will be yours.
The full corpus, packaged for a plain PC or Mac, including the builds that fit modest hardware. No AI attached: just very good, very light scientific tools.
Microsoft Store and Apple App Store. Rolling out over years, because 7,666 is a lot of packaging.
The same corpus with its AI turned on. Drive it with Claude or Codex, or download our own agents and run them on your machines.
Knowledge-graph reference access lands here too, as apps on the same stores.
The entire substrate installed inside your institution. Frontier-grade output on modest hardware, by design rather than by discount.
Your data never leaves your gates. Your token bill never leaves orbit.
Why we are building it at all
Software will not replace the people who carry science. It will hand them back their time. That is the whole idea, and the whole product.
We built this substrate for the researchers, the postdocs, the PhD students, and the curious with a laptop and a question. Where resources are thinnest, the tools will be free.
Come back when it ships. Or ask us while it is being built.
Every module on this site exists in a build graph, a contract, and a record. Ask which, ask why, ask when.