Tenant hotline

Tenants text one number. The policy decides who gets woken up — the model only reads the message; time and weather come from your systems.

  • A low bar for danger: 0.3 pages on-call
  • Time and temperature never reach the model
  • Same leak, different hour, different person

A request comes in

Pick a scenario or write your own. The facts on the right come from your systems.

From your systems · never sent to the model

The model answers the policy's questions

You play the model: drag an answer and watch the decision change.

issueWhat is message reporting?

danger?Could anyone be in physical danger right now?

The policy decides

Plain code picks the branch, and says which clause and which readings did it.

Pick a scenario to see the decision.

Call it from your code

The same route, the same answer. Drop the answers to let the model reply.

shell
curl https://jevlang.sh/api/maintenance \
  -H 'content-type: application/json' \
  -d '{"input":{"message":"Its 3am and the hallway outside 4B smells like rotten eggs. Is that normal??","hour":3,"outsideTempC":2},"answers":{"issue":{"choice":"other","confidence":0.6},"danger?":{"noul":0.92}}}'