Configure

AI guidelines

How Bob is allowed to behave for your firm. These rules wrap every agent, and every decision is logged against them.

In force

Bob always

Required

Enforced

routeModel(task, dataClass)

General questions, drafting, non-sensitive quoting

Fast, low cost. No customer PII.

Fast lane (non-sensitive)

Sensitive + health PII (life & health is in scope from day one)

Western in-region inference. Fails closed if unavailable.

In-region, DPA-backed

Firms see "in-region, DPA-backed inference", never a model brand name. Sensitive and health data never leaves the region.

Escalation rules

Adverse decision

Any decline, health load, or premium above the band routes to a licensed adviser before it reaches the customer.

Out of playbook

A question Bob is not confident on, or outside scope, hands off to a human with the full context.

Fail closed

If the in-region model is unavailable for sensitive data, Bob stops and asks a human rather than falling back.

Every response is checked against these guidelines and logged with reason codes. See the full trail in Compliance & audit.