Live demo
Admin assistant: rules through conversation
The manager below describes a business rule in plain language. @de_canter/apogee-rules parses it against the dimensions the rental company declared (event, equipment category, rental tier, customer type), returns it as a kernel Assertion with a confidence and its assumptions, and waits. Confirming on the card records who decided; from then on the rental desk on the chat demo carries the rule in its prompt and its booking tool runs the rule's structured check as a gate.
Package
defineDimensions generates the parser prompt, the validator, and the evaluator. Evaluation is two layers: structured checks first, deterministic; the model second, only when the checks pass.
Composition
src/demo/rental/admin.ts: an @de_canter/apogee-agent session with ruleTools(...) and bounded memory. The card's Confirm button becomes the next message with a model-only annotation, and the assistant calls confirm_rule.
Effect
src/demo/rental/rules.ts: one in-memory rule world per visitor, keyed by a cookie both demos share, so your rules never reach another visitor's desk. The desk reads it through a prompt contributor and create_rental runs evaluateGate on every booking, recording each firing in the audit log.