PrudAI

Agentic AI for regulated professional work. A Dutch AI company building EU-pinned, audit-grade assistants for law, the built environment, and healthcare.

🌐 prudai.com · 🔒 Trust Center · ⚖️ Legal Center · 💼 LinkedIn · 💻 GitHub

---

What we build

We ship three agentic AI assistants, each grounded in the source corpora of a regulated Dutch professional domain. Every answer is traceable to a verifiable citation — no hallucinated paraphrase, no "trust me, it's in there somewhere."

⚖️ LEO — Lex Explanata Oraculo

A legal-AI assistant for Dutch lawyers and in-house counsel. Multi-agent dossier analysis, deep case-law research across 86+ validated sources (wetten.overheid.nl, Rechtspraak.nl, EUR-Lex, Autoriteit Persoonsgegevens, Kifid, BAG, KVK, and more), document drafting in your house style, and word-for-word citation auditing that blocks any answer whose quote does not appear verbatim in the source. → leo.prudai.com

🏗️ VERA — Vergunning & Regelgeving Agent

A compliance assistant for the Dutch built environment. Covers the Omgevingswet, Bbl/Bal, and Wkb across the full permitting and construction lifecycle: developers and contractors get RvA risk signals on execution detail, inspection moments, stop-the-line criteria, and file-readiness checks for gereedmelding. → vera.prudai.com

🩺 ZIA — Zorg Informatie Agent

A healthcare-AI assistant for Dutch care providers. Supports quality officers and care professionals with methodical quality-management workflows, source-use and dossier checks, and structured preparation of quality reviews. → zia.prudai.com

---

How we run

The reason customers in regulated sectors can deploy us is operational, not marketing:

• Primary storage in the EU/EER. Customer documents and case files are stored on Hetzner Nürnberg / Falkenstein under a signed DPA. AES-256 at rest, TLS 1.2+ in transit.
• EU-first AI inference. Azure OpenAI pinned to Sweden Central with zero data retention; Vertex AI (Claude, Gemini) via the global endpoint with EU routing preferred and zero-retention configured. Prompts and outputs are not retained, not logged for human review, and never used to train models — neither ours nor the providers'.
• Customer data is never used for training. Not for our models. Not for our subprocessors'. Not for "service improvement."
• Built on open standards. ECLI, BWBR, CELEX, BAG-id — no proprietary citation lock-in. If you leave, your dossiers and exports keep working.
• Compliance posture. GDPR-conform. ISO/IEC 27001:2022 and NEN 7510 control framework in place; Stage 1 audit July 2026, Stage 2 certification audits September 2026. Annual independent white-box pentest.

Full subprocessor list, security controls, and compliance documents: trust.prudai.com.

---

On Hugging Face

This org is where we plan to publish the open artifacts that emerge from our work — evaluation sets for Dutch legal-QA and Bbl/Wkb compliance, small adapters and distilled models where we can do so without touching customer IP, and public demos of capabilities that don't need to be paywalled.

Nothing is published yet. When it is, you'll see it appear here. We don't intend to fork-and-rebrand other people's models; if we open-source something, it'll be because we made it and it's genuinely useful.

If you're building in EU-regulated AI and want to compare notes — or you've found something we should know about — reach out.

---

Contact

• General: hello@prudai.com · prudai.com
• Security & privacy: security@prudai.com · trust.prudai.com
• Legal: legal.prudai.com
• Docs: docs.prudai.com

PrudAI B.V. · Enschede, Netherlands · KVK 99876868 · Wikidata Q139906958