A chemistry world model for formulation intelligence.

It learns how ingredients interact, and predicts how a formulation behaves before anyone touches a beaker.

350M+
Molecules pretrained
1,000s
Experiments we ran ourselves
Every run
Makes the model sharper
The model we built

It doesn't predict the next word. It predicts the chemical properties.

Formulation R&D still runs on slow, expensive iteration - because the failure modes are coupled physics, not a search space. That takes a model that understands the chemistry, and none existed. So we built all of it ourselves: the architecture, the pretraining, the experimental data underneath.

Inputs
Molecular structure Concentrations pH Temperature Process and shear
Predicted properties
Viscosity Stability Surface tension Foam volume Aggregation
01
Pretrained on 350M+ molecules

Arrives already knowing chemistry.

02
Grounded in our own experiments

Thousands of formulation runs we generated ourselves.

03
Adapts to your chemistry

Learns your ingredients from a little of your data.

World models power the Virtual Instruments that predict your formulation, and the Active Learning loop that sharpens with every experiment you run.

And no - it is not an LLM.

What it is not
What it is
A language model or a chatbot
A foundational model we architected and own
A wrapper on someone else's weights
Trained on our own experimentally collected data
A regression fitted to one dataset
Foundational chemistry knowledge that adapts to your data
Proven in industry

Already running in real R&D.

Not a research demo. Deployed today across formulation-intensive industries, and now extending into biologics.

Agrochemicals

Suspension stability and rheology

Personal care

Surfactants, shelf life

Cosmetics and CDMO

Faster reformulation at scale

Bio-pharma

Viscosity, aggregation, injectability

Partners and programs
BASF
NVIDIA Inception Program
Microsoft for Startups

Further programs under NDA. See our partners

The platform

The world model, in your lab.

Build a formulation, run it against digital twin instruments, see the result in seconds.

FastFormulator Workbench
Shear thinning analysis in the Workbench

Predict, model, optimize.

Screen formulations on your laptop

Thousands of candidates before the lab

001
Predict properties before you mix

Virtual experiments on digital twin instruments

002
Get more from your formulation history

Turn archived experiments into an asset

003
Tune the model to your chemistry

Adapt it to your ingredients and instruments

004
In your workflow

It joins the loop you already run.

Step 01
Set objectives
Step 02
Predict and screen
Step 03
Run the best experiments
Step 04
Feed results back
Step 05
Scale with confidence

Ready to put the model to work on your chemistry?

See a live prediction run, or talk to us about a pilot.

Partnership and investor enquiries welcome.