Policy and Governance decides what a person or an agent may send to a model, what a model may see, what comes back masked, and which model may hear it at all. Policies layer from the whole firm down to one person, and the Policy Resolver shows which rule decided.
Admins at every level write policy once. Cortx applies it to what is sent, what comes back, what is reached and which model hears it.
A question about what is allowed should have exactly one place that answers it, and the explanation an admin sees should come from that same place, not from a second copy of the logic.
Four steps take a policy from the admin who wrote it to the explanation of every decision. Select a step to jump to it.
Admins write guardrail policies and map them to seven tiers, from firm-wide down to one person. Data-access rules, routing policy and containment policy are written alongside.
For each protection category, the strictest mapped action wins. Where a policy has opted in to override, the most specific tier wins instead.
Prompts and replies are scanned before they move on. Replies are masked when the policy says so, and tagged content can be routed to a private model.
The Policy Resolver shows, for each category, which policy at which tier decided. Every scan's audit record lists every policy that applied.
"Policy" is really four questions. Each should have exactly one engine that answers it.
For guardrail categories, the Policy Resolver calls the same code as live enforcement, and a test proves the two agree.
Policies map to tiers, from the whole firm down to one person. For each category the strictest action wins, unless a policy has explicitly opted in to override, so loosening is always a deliberate choice.
Pick a person and see, for each category, the action that applies to them, the policy that decided it, and the tier it came from.
Replies are masked when the policy says so, and content tagged as sensitive can be routed to a model running inside your network.
How a decision is made, how it is explained, and where it applies today.
For guardrail categories, the Policy Resolver and live enforcement call the same function, and a test drives both over the same data to prove they agree.
A detection can raise severity, never lower it. Loosening takes an explicit override by an operator.
A control that changes verdicts ships in monitor mode first, with a counter you can read, so you see what it would have decided before you switch it on.
Guardrail policies decide what can be sent and what comes back masked. Private-model routing applies at the model gateway; we'll show you exactly which paths it covers in your deployment.
We'll map your firm's policies across tiers and walk one blocked prompt back to the rule that blocked it.
We'd like to use Google Analytics to see which pages are read, how visitors find us, and roughly where in the world they are. Nothing is stored on your device unless you turn it on, and you can change your choice at any time. See our Cookie Policy for details.
What we collect