Talk to us
Company

AI a nation or a company
can call its own.

Our objective is to help countries and companies build AI of their own. We make small language models that run on your hardware, speak your people’s languages and do one job well, so you keep control of the AI you depend on.

Talk to usOur models
Why we exist

Most AI is rented from someone else’s cloud.We build AI that nations and companies own.

On your hardwareIn your languagesUnder your rules
Who we work with

Built for those who
cannot hand AI to a stranger.

Governments and public bodies

Language models for national languages and public services, deployed inside your own infrastructure and under your own rules.

Regulated enterprises

Organisations that cannot send customer data to a shared cloud, from financial services to healthcare.

Teams with one hard task

A small model tuned for one job in your business, measured on your own test set before it goes live.

What we build

Models, a product
and a service.

ProductSoftware you use

Methodal: a collage of prints rebuilt in new inks, standing for synthetic data
MethodalIn development

Real patterns.
No real people.

Synthetic data that keeps the patterns of your tables and belongs to no real person.

Learn about Methodal
stand-in.csvTest and demo data
idageareabalance
c7f2e946LS6£3,204.18
1a90d331BS3£388.70
e04b7c61CF24£26,511.09
9d3f2137YO31£1,902.44

ModelsTwo model lines, five models each

For the UKBalsa

British English, pounds, UK dates and postcodes.

  • Omni
  • Text
  • Vision
  • Voice
  • Decision
View Balsa models
For IndiaGati

Eight languages used in India, rupees and UPI, in each script.

  • Omni
  • Text
  • Vision
  • Voice
  • Decision
View Gati models

ServiceBuilt with you

Bespoke

A small model around one task in your business, built and tested with your team.

  1. Scope one job with your team
  2. Build and test on your data
  3. Run it on your own machines
Talk to us
How we work

Four things
we hold to.

  1. Small on purposeA model only as large as its job needs, so it fits the machine you already own.
  2. Local by defaultWhat people type stays on their own hardware. No cloud is needed.
  3. Evidence with every numberEvery figure says how it was measured and where the model falls short.
  4. A person decidesOur models recommend. A person on your team approves what happens next.
Fahrenheit Research

Deep Research.
Real Intelligence.

Tell us what your country or company needs a model to do. We will plan it with you.

Talk to us