Building a tax research assistant that's honest about what it knows — and what it doesn't.

Auryth was born from frustration — not from a pitch deck.

For over a decade, our team has lived inside Belgian tax. We've sat in the offices of law firms, Big Four advisory teams, accounting practices, family offices, and corporate tax departments across Belgium, Luxembourg, and Switzerland. We've done the work: navigating the WIB, cross-referencing circulars, chasing down rulings across Fisconetplus, Jura, and Monkey — often for hours, just to be confident in a single answer.

We also happen to be deeply technical — from both sides. One of us has been programming in Python for years and was working with natural language processing long before anyone called it "AI." Algorithmic trading, early OpenAI models like DaVinci, custom-built RAG systems — we were implementing this technology for our own work well before it had a name that trended on LinkedIn. The other spent years as a technology consultant advising major financial institutions — designing systems, managing complex transformations, and understanding firsthand what it takes to build tools that professionals in regulated industries actually trust and adopt.

That combination — deep tax expertise, hands-on AI experience, and enterprise technology consulting at the highest level — is what makes Auryth different. When we tested general-purpose AI models on real Belgian tax questions, we saw exactly what any experienced practitioner would see: confident answers that were subtly, dangerously wrong. Outdated rates. Blended jurisdictions. Missing exceptions. No sources.

We knew it could be done better. Not by making another chatbot that guesses, but by building a research platform that thinks the way a tax professional thinks — methodically, with sources, with awareness of what changed and when.

That's Auryth. Every answer is scored for confidence. Every source is ranked by legal authority. Every claim can be verified.