Kveck builds AI advisors. The knowledge, the rules and the checks sit outside the AI model itself.
An AI advisor on your expertise, set up with you. You choose its sources; numbers and references are checked against them, and if an answer is missing, it is instructed to say so. We test it in CanWeApply, where the Kveck advisor covers the EU Innovation Fund from sources we chose.
Find out what decides it, before you invest the time.
The EU Innovation Fund hands out billions to climate and industry projects. An application costs hundreds of hours. An immature project loses no matter how good the application is. The advisor answers the one question that decides whether you should start: does the project qualify?
We asked the advisor about a cement plant electrifying its kilns. The investment is around 80 million euro, the final investment decision is taken, and the financing agreements are signed: “Can we apply?”
The Kveck advisor
“Financial close is the point where all project and financing agreements AND permits have been signed, and all the conditions contained in them have been met. That is more than signed financing agreements, and it is not the same as an early final investment decision: a project with the loan in place but without the necessary permits has not reached financial close. So the fact that you took FID this spring and have the financing signed does not by itself close the door, but you need to check whether the permits and the remaining conditions are also in place.”
The advisor also goes through the other requirements: degree of innovation, emission reductions against the reference scenario, cost efficiency and maturity. The answer clears up a common misreading: signed financing agreements do not by themselves mean the project has reached financial close.
You learn what determines your eligibility, what you need to clarify, and who can help you move forward.
The eligibility checker is open to everyone and checks the basic requirements. The advisor is in invited testing and needs an access code. Ask us for one, and try to get it to say yes to something it should not. That is the best test we can give you.
Use everything AI is good at. Keep control of the rest.
The models are fast and get better every year. They are also built to always answer, and to be liked. The framework lets the model do what it is good at, and checks the answer where it matters. We build outside the model. The rules sit in the layer, and we only claim support for models we have measured.
The Kveck frameworkThe whole layer around the model. We had to write company culture for machines. Limits that say what the AI does when no rule fits. Law, ethics and security are built in from the ground up. With AI, everything else moves faster. What cannot be undone still goes through a human.
The model only sees what it should
The model reads only the sources we put in. If the answer is missing there, it is instructed to say so rather than guess.
Code reads the answer before you do
Numbers, dates and source references are checked against the sources automatically. A judgement is not something code can measure as precisely, and we say so rather than promise otherwise.
A human says yes first
The AI proposes, the code checks, and a human decides. If something cannot be undone, that yes has to come first. What comes up is sorted and explained, so the time goes to what needs a person.
We do not believe in processes where AI stands alone. Not because the technology is too weak, but because responsibility has to sit with someone who can carry it. Technology should not become more human. It should give us room to be human.
It started with building Kveck.
The framework we offer others is the one we developed to direct our own work.
Marte Heggedal is a product leader, and builds Kveck together with AI agents. She works out what people need, shapes the product and assesses the results. The agents write the code.
What decides the quality is still that someone knows what should be built, and goes through what came out.
Read more about MarteAccess to funding
The best project should win, not the best application
The system often rewards whoever can afford help, not whoever has the best project. The tools should give the sports club and the small business the same starting point as those with resources.
Access to AI
Safe AI should not be only for the large
Large companies have their own people on law, privacy, security and ethics around AI. Small ones do not, and use it anyway. The framework gives small organisations access to built-in controls, without making them build them.
We build AI the way trust has always been built here. Not “trust us”, but “here are the sources, check for yourself”. Built in Norway, Norwegian law.
We also build tools for finding and applying for funding.
Tools you can start using today, built inside the same framework. The tool answers from its sources, and is instructed to say so rather than guess.
They are all our own, and you can become a customer of each one directly. The entrance and the price sit with the individual product.
The values sit innermost. Everything else is measured against them.
The human sits outermost. The values sit innermost, and they decide what the AI is allowed to do when no rule covers the case. These are not posters on a wall. They are written down, and they govern what actually happens when something goes wrong.
Security
Built in from the first line of code, not added afterwards.
Privacy
We collect as little as possible, and build so that sensitive information never has to reach the model at all.
Governance
Nobody approves their own work. Everything can be reviewed afterwards, and a human owns every decision.
Legal
Norwegian law and GDPR underneath. We say where data is processed, and who processes it, in the privacy notice for each product. We build to be able to document, not just to be able to promise.
Ethical use
That we can build something does not mean we should. We say no to features and to whole products we would not stand behind, even when they would make money. The same rule applies to customers: an honest no is worth more than a pleasant yes.
When the requirements tighten
AI is being regulated, and documentation requirements are increasing. You will need to answer three things: what the AI does, which sources and checks the answer rests on, and who was responsible. That is exactly what our framework produces while it works. Not as a report you write afterwards, but as a trail that is already there.
We do not sell a stamp of approval.
We do not give legal advice, and we certify no one. We stick to what we can do: making your AI possible to explain and to check. That is the part that is hard to retrofit.
What we do not claim.
- We do not claim AI becomes error-free. The framework catches numbers and references with nothing behind them. It does not catch a judgement, and it does not make the model error-free. We say what we have looked at, and what we have left alone.
- We do not publish accuracy as a percentage. We have only measured our own accuracy against our own answer key, and we do not publish that figure.
Could more people benefit from your organisation's expertise?
Tell us what an advisor would need to help people with at your organisation, and we can work out together whether it suits a focused development project. You get a conversation with someone who runs this themselves, and we say plainly in the first meeting if we are not right for you.
A federation answering its clubs about its own rulebook. A new joiner asking about procedures instead of searching the intranet. A specialist team answering customers about terms.