How we pick the companies

A directory is only useful if you trust the list. We'd rather have 30 sharp entries than 300 noisy ones. Here's how we decide what makes the cut.

The bar

Three things, roughly in order:

1. AI is the product, not the topping. The company would be meaningfully different without the model at the core.
2. There's something to look at. A live website, a real product, customers — or at minimum, a public team and a clear thesis.
3. Croatian footprint. Founded here, headquartered here, or has a substantive engineering presence here.

What we don't include (yet)

• Pure consulting shops that occasionally ship AI projects.
• Stealth companies with no public surface.
• Crypto-adjacent projects that wrapped a model and called themselves AI.

Edge cases

A few that come up:

• Acquired companies stay in the directory if the team continues to operate out of Croatia (or if the work was clearly Croatian-built before the exit). Photomath is a good example.
• Multi-country companies with a Croatian co-founder but no real local presence — usually a no.
• Tooling for AI (vector DBs, dev environments, eval tools) — yes. The infrastructure layer is part of the scene.

This list will get refined. If you think we got something wrong, tell us — we'd genuinely rather hear it.