Will compliance break AI agents?
With Kyvvu, we are building a runtime compliance layer for AI agents. What does that mean? Well, it simply means that we are creating a middleware that allows you to seamlessly log the behavior of AI agents (immutable and auditable) and enforce specific agent policies (at the level of the agent, the task, or individual task steps and their relations). As such, we allow users to — out of the 100+ articles of the AI Act, of which just over 40 are technically enforceable — comply with 100% of the logging requirements, and as is, with roughly 95% of other "agent behavior" requirements.
We are building Kyvvu to be hyper-easy to integrate (using Python decorators, LangChain integration, etc.) and non-intrusive. Effectively, we aim to provide compliance "out-of-the-box" so you can move forward with agent development and let us worry about the shifting regulatory landscape.
Right now, regretfully, all of our code is closed. We want to run several large scale evaluations before opening up. However, we will open up our SDK(s) soon; planned Q2 2026. If you are interested in early testing once we gradually open up, email us at alice@kyvvu.com.
Co-Development Pilot Program
At this point in time, we are actively signing up potential design partners for a co-development pilot. As we are a small team, spaces are limited. Here are the details:
- We are looking for companies building their own AI agents (either in house, or with technology partners), using custom Python frameworks and/or LangChain.
- We are looking for companies in highly regulated spaces: healthcare, banking, insurance, etc. This is where the AI Act will first come into full force.
- We are looking to pilot our platform: that means that — together with one of our implementation partners — we will install the Kyvvu back-end on your infrastructure and decorate one or two of your agents with Kyvvu logging and policy checking.
- We will likely need about 1 week to get things started, and we will actively monitor for roughly 3 weeks. After those 3 weeks you will get a compliance report, insight into your gaps concerning the AI Act and your agent deployment, and a first say into our future platform development.
- And, obviously, the go-to book on how to integrate agents into your organization (see theaiagentbook.com).
Are you interested in co-creating the future of automated AI compliance? Reach out at alice@kyvvu.com.
About Kyvvu
Kyvvu is early stage, but we intend to fully automate your AI Act compliance. Why? Because we think AI agents can make our lives better, but proper oversight — whether enforced by the AI Act or not — is essential. As is, we have a feature mature API, SDK for Python (with LangChain integration) and our first management dashboard for both developers and compliance professionals. After a lot of internal testing, we are now ready to move out.
Kyvvu is founded by Maurits Kaptein (maurits@kyvvu.com) and our first round of pre-seed funding allows us to keep building; we are expanding the team. Reach out iff you are interested in building this.