Security & trust
AI with authority limits, not blind autonomy.
North acts on commercial data, so control is part of the product design. Organizations decide what AI may prepare, what it may execute, and what always requires a human decision.
Tenant isolation
Every record is scoped to an organization and enforced with row-level security in the database, not in the interface.
Role-based authority
Admins, managers, representatives, finance approvers, and viewers each see and approve only what their role permits.
Server-side rules
Pricing, discounts, margins, and approval thresholds are calculated and enforced on the server. The UI never grants authority.
Complete audit history
Every AI and human action records the inputs used, the decision, the reason, confidence, the rule applied, and the outcome.
Pause and take over
Any user can pause AI for a customer or opportunity, take over a conversation, or report an incorrect action.
No invented facts
AI never invents prices, margins, inventory, products, or legal terms. Low confidence becomes a clarification or a human action.
Start in recommendation mode
New organizations default to Prepare: AI drafts documents and messages, and your team approves before anything reaches a customer.