Every security team asks the same thing before letting an agent near live work: what will it do when nobody is watching? A slide can't answer that. Ninety days of your own agent activity, measured against your own policy boundaries, can.
How the pilot works
We deploy a read-only version of the Glass Wall Framework that we call the Shadow Proxy. It sits in parallel with your existing agent workflows and watches the token stream. It calculates how far each agent's reasoning has moved toward a policy boundary, and it records the result. It does not correct, block, freeze, or delay anything. Your agents run exactly as they do today.
It runs inside your walls
The Shadow Proxy is deployed in your own environment, either in your virtual private cloud or on your own on-premises gateway, whichever your compliance requirements call for. All of the analysis happens there. Raw agent data stays inside your environment, and the deployment is single-tenant, so nothing is shared with any other customer.
It works on the token stream of any large language model you route through your own orchestration layer. We don't need access to your model provider's infrastructure.
From watching to acting
What we ask in return
Honest feedback during the pilot, and a case study of what we learned together. That is the whole price. Pilot spots are limited to a small group of design partners across financial services, insurance, legal, and banking.
What it isn't
The pilot is an evaluation, not a safeguard. While the Shadow Proxy is watch-only, it isn't a production control or a compliance measure, and neither of us should present it as one. It also won't catch every instance of drift, and we won't claim it does.
Enterprise pricing for active containment is scoped to your environment and agent volume, and it's discussed on a call.