AN AI STUDIO FOR SMALL AND TINY LANGUAGE MODELS

Small is not little. Small enough to own.

AI that lives inside your walls, fitted to your work, and tested on it before you rely on it.

The Key 4 AI transforms organisations end to end, with large language models, small language models, or a hybrid of both. Our distinctive edge is the small model you own rather than rent: expert at your work, inside your walls, and yours to keep.

Talk to us See what we have made
  • We understand the work, choose the AI it truly needs, prove it on a slice of your data, build it, check it against a bar you set, release it on hardware you already own, and run it for as long as you rely on it. Where the job allows, and it usually does, the model is small or tiny: small enough to own, to check and to keep.

WHY WE EXIST

Let intelligence grow with trust, and all of us rise with it.

The company in full
  • We build them so that whoever holds one can do more, own more, and answer for more than they could before. Everything else we leave to the people it belongs to.
An arch made of six short phrases, with the words trust, the rise and us lit inside it.

TRUST

  • With it, never ahead of it: we grow intelligence no faster than the trust it grows alongside; speed is never a reason.
  • Honest enough to test: we claim nothing you cannot check for yourself, and we never grade our own work; the person relying on it holds the test.

THE RISE

  • The last hand first: we build for the person with the least, on what they already own; if it cannot live there, it is not finished.
  • No one smaller, everyone larger: we take no work whose purpose is that a person ends up less.

US

  • Theirs to command: what a person holds is theirs; no door locks from our side.
  • Small first, and grown by the work: we start with the smallest model and the smallest team that does the job, add only what the work demands, and what we need not keep, we give.

THE EDGE

The right model for the job: small first and owned, large when the job needs it.

  • We chose small because small is where ownership becomes possible. A model light enough to run on your own machine is light enough to own, to check and to keep. When a job needs more, we add more: a larger model for the rare case, or no language model at all when plain algorithms do it better.

THE MODEL YOU RENT

Rented, large and metered. It runs in someone else’s data centre, costs more the more you use it, knows nothing of your work, and can be repriced without asking you. Right for the rare, open-ended question, and we use it there.

THE MODEL YOU OWN

Owned, small and fitted. A small language model made expert at your work, running on hardware you already have, checked on your own test, answering to no one but you. Small done well is a craft, and it is ours.

COST THAT DOES NOT GROW WITH USE
No charge per question on your own hardware.
DATA THAT STAYS
Your records never leave your building, or your phone.
ACCURACY CHECKED ON YOUR WORK
On a test you hold, before you rely on it.
NO DEPENDENCY ON ANYONE
Nobody can switch it off, change it or reprice it.

We are AI people, experienced in AI transformation with large language models and small language models. We use what the job needs. Our distinctive edge is the small language model, and the tiny one, that you own.

WHY NOW

Three things changed at once.

THE METER IS RUNNING

Rented AI is priced by the question, so the bill grows most where the value is.

THE REGULATORS ARE ASKING

For an inventory of every model, a gate before any agent acts, and a record that can be audited.

SMALL BECAME GOOD ENOUGH

On narrow, high-volume work, a fitted small model now does the job on hardware you already own.

$13.0bn · 2025$28.3bn · 2026$51.6bn · 2027, forecastSpending on generative AI models, worldwide

Each task an AI agent runs: 5 to 30 times the tokens

Less per token, more per task. Gartner, September 2026 and March 2026.
Inventory. A gate before action. A record. Everywhere. Bank of England, European Commission, Central Bank of the UAE, Qatar Central Bank, Federal Reserve, Reserve Bank of India; read October 2026.
20222024142 times smaller, same score

By 2027, small task-specific models used at least three times more than general-purpose ones

Good enough on narrow tasks, not for everything. Stanford AI Index 2025; Gartner, April 2025.

LIGHT BY DESIGN

No data centre behind every answer.

  1. 485 TWh to about 950 TWh

    Electricity used by data centres, 2025 to 2030.

    IEA, April 2026.

  2. 15 to 50 times less

    Energy a small task-specific model used for the same job as a large general one, in one study of three tasks.

    UCL and UNESCO, July 2025.

  3. 5.8 billion

    People with a mobile subscription: machines intelligence can already live on.

    GSMA, March 2026.

One data centre15 to 50 times less energy, small against large

Power is still used. The point is a lighter load, carried by the right-sized model, spread across machines already in use. Intelligence that runs on what people already own reaches more people.

2.2 billion people are still offline (ITU, 2025). Small models reach the devices people already own, and that is where we start.

ONE STUDIO, TWO ARMS

From the first idea to the last day it runs.

The Key 4 AI is an AI studio for small and tiny language models. It has two arms, and one way of working for both.

PRODUCTS

Products

SERVICES

Services
  • PRODUCTS Key4MB, Key4KB and Key4EDU, built on the Key4 platform. Each is ready to deploy, and each shortens the work for everyone who picks it up.
  • SERVICES AI transformation for your organisation: from the first assessment to the last day the system runs, built on what you already have, to the KPIs you set.

  1. DECIDE

  2. PROVE

  3. BUILD

  4. RUN

  • DISCOVER AND ASSESS Consultancy first. A small team learns your business, the work worth changing, and what you already hold in people, data and systems, then ranks the opportunities with you. You leave with a plan and a clear go or no-go.
  • ARCHITECT We choose the technique the job needs: plain rules, machine learning, a small model you own, a large one, or a mix of them, and we record why each other option was set aside. Where it runs, how data moves and what it costs to own are settled here. This stage can be bought on its own.
  • PROVE Before anything is built, we show it can be done. A working version on a slice of your data, measured against a pass mark you set in advance. Pass, and you decide to build. Fail, and you have spent weeks, not a year.
  • BUILD AND INTEGRATE This is our craft. We fit the model, build the software around it, connect it to your systems, and put a check before any agent acts. Independent testing before release.
  • LAUNCH AND ADOPT Going live with your people, not around them. A phased release, the workflow redesigned with those who do the work, training, and a hand-over of the model, the tests and the documentation.
  • RUN AND IMPROVE The Keep. One governance layer over every model, monitoring to the KPIs you set, and no change reaching production until your test has run again. Improvement is constant and systematic, for as long as the system serves you.
  • SCALE What worked is repeated across more workflows, sites and languages, reusing the method, the tests and the models. What did not earn its keep is retired.
  • RETIRE AND HAND BACK When a system ends, nothing is lost. Its data, models, tests, decisions and lessons are preserved and handed to you, or carried into what replaces it. No copy stays with us.

Every stage can be taken on its own.

The Studio Services

KEY4 · THE PLATFORM

One platform. Three products cut from it.

All ready to deploy. Each built on the Key4 platform.

The platform they are built on.

Our method for making small models expert and checking them; our engineering that runs them on ordinary hardware, phones included; our own small language model at its heart.

Key4
A phone screen, filmed in one take: an assistant reads the screen, opens apps, taps and types, narrating each step in a strip at the bottom.

KEY4MB · THE FLAGSHIP

An AI that uses your phone the way you do.

The whole small language model sits inside the phone. Fast, on the phone itself. Private, and yours.

Zero tokens. Zero cloud. Zero data leaving your phone.

Partners build on it: phone makers preload it, carriers and retailers bundle it, enterprises deploy it to field teams, chipmakers fit it to their hardware.

Ready to deploy · Open to partners

Key4MB

KEY4KB

Answers from your own records, with the page.

For public and regulated bodies. Every answer with its page, checked word for word, in your languages, on your server.

Key4KB

KEY4EDU

Learning that teaches by asking.

For schools. Knows the learner · Teaches by asking · Shows the teacher.

Key4EDU

Our AI skill makes both: the transformations we run for you, and the products we build.

The range

PROOF

Owned. No tokens. No cloud. Private. In your control.

No cut inside a take, no cloud behind the phone. The strongest proof we have is to let you watch.

Seven films so far. What they show is not the task; it is how each one runs: a small language model, on a phone, fast, private and secure.

  1. Still from the film “Ask about any screen”: the phone screen mid-task, with the step strip at the bottom.Ask about any screen
  2. Still from the film “Two apps, one ask”: the phone screen mid-task, with the step strip at the bottom.Two apps, one ask
  3. Still from the film “Shop with sources”: the phone screen mid-task, with the step strip at the bottom.Shop with sources
  4. Still from the film “It sees”: the phone screen mid-task, with the step strip at the bottom.It sees
  5. Still from the film “It remembers you”: the phone screen mid-task, with the step strip at the bottom.It remembers you
  6. Still from the film “Teach it once”: the phone screen mid-task, with the step strip at the bottom.Teach it once
  7. Still from the film “Movie night”: the phone screen mid-task, with the step strip at the bottom.Movie night

TWO FACTS

THE STOP YOU HOLD

Volume down stops it, anywhere, mid-step. Every send, payment and sign-in waits for your tap.

NOTHING LEAVES THE PHONE

The model thinks on the phone’s own chip. What it reads stays there, and every step is kept on the phone for you to see.

See the proof

OWNERSHIP OF THE MODEL, THE TEST AND THE RECORD

No lock. Just the key.

Nothing we build ties you to us. You hold the model, the test and the record, and you can run them without us.

Talk to us

Let intelligence grow with trust, and all of us rise with it.