Our Story

I’m Blake Heron. Novyx started with a simple problem: agents lose context. Memory, rollback, and audit were the first primitives I built because agents needed durable state and recoverability.

But the more useful agents became, the sharper problem changed. The risk was no longer only that an agent forgot something. The risk was that it could change a database, push a deploy, alter a repo, email a customer, or touch billing before a human understood the blast radius.

The next agent infrastructure problem is not “does it remember?” It is “can it safely change production?”

That is where Novyx is going: change control for AI agents. The current wedge is the blast-radius gate for schema migrations. It reads the exact migration an agent proposes, joins it with live impact facts, and returns a verdict before the change can run.

Memory still matters. Rollback still matters. Audit still matters. They are the substrate underneath control: they explain why an action was requested, preserve what happened, and help teams recover. They are no longer the headline.

Novyx is still bootstrapped and built in public. The product is intentionally narrowing around the place where agents become operationally dangerous: the moment before they mutate real systems.

If you are building agents that can change production, start with one dangerous action. Put the gate in front of it. Make the verdict readable. Then expand from there.