How small can a model be and still do the job? We are a Nordic lab building efficient AI that carries out workflows.
Our story
Most of the work inside an organisation is process. A case is opened, documents are read, a decision is made, a system is updated.
Today that work is handed to the largest general models available. They were built to converse about anything, and they are powered and priced for it.
A workflow asks for careful reading, rules that are followed, and answers that can be checked.
Alexandra Sandberg
Co-founder
alexandra.sandberg@island.mlLinkedIn (Alexandra Sandberg, opens in a new tab)
We are a lawyer and an engineer from the Nordics. In 2026 we started a lab around one question: how small can a model be and still do the job?
We created the model and its inference engine from the ground up: self-funded, with support from RISE and time on Sweden's supercomputer. It is trained mainly on the Nordic languages and English.
Mats Gabriel Bosson
Co-founder and CTO
mats.bosson@island.mlLinkedIn (Mats Gabriel Bosson, opens in a new tab)
Every layer is ours, so we can change any of them for a customer.
Frequently asked questions.
It is our own.
We created the model and its inference engine from the ground up. No third-party model sits behind our API.
No.
It is a model for workflows. It reads the material, applies your rules and carries out the steps of a process.
Yes.
It carries multi-step workflows where documents are read, checked against each other and decided on, as well as simple sorting and routing.
The Nordic languages and English.
It is trained mainly on these, so it reads Nordic material in its own context.
On Genki, our platform, you sign up and build your own automations on our standard model.
In Applied Automation, we build a custom solution and a custom model with your team, for problems no off-the-shelf tool solves.
It stays in the Nordics, or with you.
Our platform runs on Nordic servers, and we can set the system up on your own servers.
On the platform, no.
We have zero data retention. In a custom solution, your model is trained on your material, and only as agreed with you.
You can keep the model and the right to run it.
On your own servers it keeps running without us. We agree the terms with you before we start.
The platform has a fixed monthly price per workflow, set by your volume band.
Custom solutions are priced on KPIs we set together.
Yes.
Our model runs your workflows, and it works alongside whatever you use for other things.
On the platform, test it with your own examples.
For a custom solution, book a call with us. We start with one workflow and agree the KPIs before anything is built.
Something else? Talk to us.