Mission
Make AI agents safe to run at scale.
Autonomous agents will change how organisations work, but only if every action they take stays inside boundaries you can verify. Kyvvu turns those boundaries into runtime guarantees.
Why we started Kyvvu
Kyvvu was founded in late 2025 on one observation: organisations were deploying AI agents faster than they could govern them. The infrastructure to make agents follow organisational policy at runtime, not in a report afterwards, didn't exist. Prompting isn't enough. Permissions aren't enough. The missing layer is runtime enforcement.
The risk is in the path
Governance violations rarely live in a single action. They live in the sequence. An agent reading a customer record is fine. Reading it, contacting an external system and writing data back may or may not be, depending on the context. You only see that by looking at the whole task. We call this approach policies on paths, and published it as a research paper: Runtime Governance for AI Agents: Policies on Paths.
Security first, compliance follows
Serious organisations worry about security before they worry about compliance, and Kyvvu is built that way round. The EU AI Act codifies much of the need, but the more important outcome is deploying agents with confidence: whatever path an agent takes, your rules are enforced before the action, not after it. The audit evidence comes with the enforcement.
Where we are
Kyvvu runs in production at regulated enterprises in the Netherlands, in financial services, insurance and healthcare, delivered through authorized partners. In 2026 we raised a €1M pre-seed round from Volta Ventures and the Brabant Development Agency (BOM).