Prompt rules

Prompt rules are small pieces of policy or persona that steer the AI. Use them when a behavior cannot be captured in a knowledge item — tone, refusals, escalation triggers, safe defaults.

Rule kinds

Rules come in four kinds: persona (voice and style), policy (must / must-not), hint (soft preference), and escalation (route to a human). Each kind renders into a distinct section of the system prompt.

Priority

Rules have integer priorities. Higher wins on conflict. Use gaps of ten so you can insert rules later without renumbering.

Activation

A rule can be always-on, restricted to a channel, or gated by tags on the incoming conversation. Deactivate a rule to A/B test the difference without deleting it.

kind: policy
priority: 100
activation: whatsapp
body: Never quote prices without checking the live catalog.

Related docs