Cortex

Cortex shows the intelligence your brand's AI has built: how mature it is, how that grew over time, what it learned recently, and what it would cost to rebuild. It is a screen for reading. It does not resolve tickets or change your policy.

Open Cortex

Select Cortex in the sidebar of your dashboard. The top of the screen shows two figures:

• Proprietary Asset Valuation, with the estimated replacement cost below it.
• Model Maturity Level. Select it to open the full breakdown.

The accumulation chart

Knowledge Equity Accumulation shows how your AI's knowledge built up over time, in stacked layers: your ticket history, your policies, the corrections validated by people, the flywheel of operation, and your domain. Choose 1W, 1M, 6M, 1Y or ALL to change the period, and move over the chart to read any point.

What the AI learned

• Auto-Learned Insights lists what your AI learned recently, in plain language.
• Zero-Effort Training Loop shows how it learns: a ticket comes in, the AI resolves what it can, a Human Node covers the rest, and the AI absorbs the human resolution. The two percentages show how the work splits between them.

Model maturity

Maturity is scored from 0 to 100, from four dimensions. Each one appears with what it measures and why it weighs what it does:

• Policy Coverage (30%) — how much of your support has a decision rule in your Charter.
• Unsupervised Autonomy (30%) — how much the AI resolves on its own and stays resolved.
• Grounding (25%) — how much of what the AI says can be traced back to your information.
• Voice (15%) — how much it sounds like your brand.

Policy Coverage sets the ceiling for the others: a case without a policy always comes to your team, so the more of your support your policy covers, the more the AI can do. See Change your policy.

The valuation

The Financial Valuation tab breaks the valuation into its lines. Each line shows its own count and rate, and says where its number comes from: measured in your account, a Razviel cost model, or a verified external standard. Where a figure is an assumption, the screen says so. Below the lines, Cost to Rebuild Externally compares it with building the same system from scratch, and the methodology explains the approach.

How your AI is trained

Training happens at the intelligence layer, not the chip layer. Model weights are third-party; what grows on top of them is your policy matrix, your curated case corpus and your validated corrections.

No customer data crosses from one brand to another — not raw material, and not de-identified, aggregated or derived material either. Razviel operates no cross-customer improvement.

Before your operation starts, and while it is paused

Before your operation starts, Cortex shows zero and the chart is empty. While your account is frozen, it keeps your last figures, and the chart and the insights are shown at rest.