MDIL gives generative AI its left brain

What MDIL Kide makes possible

Today’s AI chatbots are brilliant talkers. Ask them for a letter or a summary and they shine. Ask them to work through a long calculation, and they can be confidently wrong, with no way for anyone to check. And when the conversation ends, everything they worked out is gone.

MDIL gives AI what it is missing: the careful, logical half. With MDIL, the AI does the hard thinking once, step by step. Every answer is checked, and the work is saved. After that, anyone can reuse it: another person, another AI, even a small and inexpensive AI running on a company’s own computer.

Think of a spreadsheet. When you change one number, the sheet does not start the whole year’s accounts again. It updates only the cells that depend on that number. MDIL brings that kind of order to AI.

Why it matters

It costs less, and uses less energy. The expensive thinking is done once, not every time someone asks. Every change after that is quick and light. And MDIL itself is not a language model: it runs on ordinary processors, not on the scarce, power-hungry graphics chips that AI models need, so it does not compete with them for computing power.

It stays at home. A company can run the finished work on its own servers, with a small AI, instead of sending its most sensitive data to the largest AI companies abroad.

You can check it. Every answer comes with its working, so a person, an auditor or a regulator can see how it was reached. That is exactly what new rules such as the EU AI Act ask of AI that helps make important decisions.

It works with the AI you already have. MDIL does not replace AI models. It makes them reliable, and faster.

Put to the test: two examples

In one test, an AI was given a difficult business calculation, split into three parts. Working alone, one of the world’s leading AI models did not get a single part right. With MDIL, the same model got all three right.

The bigger point is what happens next. With MDIL, a problem solved once stays solved, and can be used again. Small AI models that run on a company’s own computers are where the market is heading, but this task would not even fit in a small model’s memory. With MDIL Kide, a small model took the reasoning the leading model had built and gave the right answer in eight seconds.

In another test, the AI had to keep a factory’s supply chain up to date as customer orders kept changing. On its own, it had to work the whole chain out again every time. With MDIL, it updated only what each change touched, and got every answer right in less than half the time.

“The race has been to build ever bigger models. We believe the next leap is different: do the hard reasoning once, keep it, and let even a small model use it.”

MDIL Kide launches on 17 November 2026 at 17:11 at Allas Pool in Helsinki. The name comes from the Finnish word for crystal. Three products built on MDIL launch with it: MDIL Portal, where developers and companies build with MDIL; REMAP, which turns a goal stated in one sentence into AI agents that run a company’s work; and Airut, an AI assistant for businesses and people, in the chat they already use.