Fabryka AI / ManifestoFrom Poland, for the world.

Poland should
build AI.

In Poland we increasingly hear that it is already too late to build our own models. That we need too much money, too many GPUs and teams that are too large. That it would be more sensible to deploy technology created elsewhere.

At Fabryka AI we believe Poland should have the ambition to create this technology. That is why we train our own models from scratch.

Every sector of the economy will use intelligence delivered by models. We want Polish companies, scientists and engineers to be able to design, train, evaluate and develop them. We want to take part in creating the technology on which others will build their products.

Expertise cannot be downloaded together with the weights.

Open models have given us a huge opportunity. We can run them on our own hardware, adapt them and study them. We use them ourselves and will keep doing so.

But downloading a model does not tell you how to build a better one. That knowledge comes from working on data, architecture, training and evaluation. From hundreds of experiments, including the failed ones. From a team that understands why one training run worked and another wasted a week of compute.

If we want to have an influence on the development of AI, we have to do this work ourselves.

Ambition has to be backed by results.

We start with small models. They let us test ideas quickly, learn and improve quality on a limited budget. Each such experiment lays the groundwork for the next, bigger step.

A small model makes sense when it gives a concrete advantage: lower cost, lower latency, better privacy or a better result on a specific task.

Likewise, a model's Polish origin does not exempt it from competition. A customer has the right to expect quality. If another solution is better, cheaper and faster at the same time, it is our job to improve our own.

We want to win on results. Polish models should deserve their users.

Research and business can drive each other.

We are building Fabryka AI as a company that develops models and provides the infrastructure to use them.

The API and deployments bring us revenue and contact with customers' real problems. Research lets us improve quality, lower the cost of compute and build better products. We can put spare compute to work on further experiments.

We want this mechanism to fund increasingly ambitious work: better data, larger training runs and a stronger team.

We are building a private, profit-oriented company that develops open-weight models. We see this as a coherent direction. Customers pay for a useful service, and the results we share allow others to keep building.

Openness should multiply the effects of our work.

We want students, researchers and entrepreneurs to use the models, data and tools we are able to share. So that they can check our results, find mistakes and create things we could not foresee ourselves.

We also need easier access to public data, clear rules for using it and compute infrastructure available to new teams. These are concrete conditions that allow more people to get started.

The value of this work should be decided by results, reproducibility and usefulness.

From Poland, for the world.

The Polish language is important to us. So is the ability to run models locally and keep control over our own data. But our ambition reaches further: we want to create technology in Poland that is useful beyond its borders as well.

Today our resources are limited. We have to choose experiments well, watch costs and earn trust with successive results. That is exactly where building expertise, a team and a company capable of bigger things begins.

Fabryka AI exists out of a conviction that this effort is worth making.

We want others to be able to look at our models and say: if they started here, so can we.

That is why we train. That is why we share our results. That is why we keep building.