Testing changes safely

Try changes without risking real calls.

Test calls

Talk to an agent in your browser. Nothing is dialled or texted. More

Test site

A separate copy of the app with its own empty database, for trying new versions before they go live. It never dials leads automatically, and a yellow bar says you're on it.

Behaviour checks

A script runs scripted callers (asking if it's a robot, saying don't call again, asking for a price, being busy, small talk) against an agent and has a second AI judge each reply, plus rules like never speaking planning notes. Run it after changing prompts or playbooks: npx tsx scripts/eval-agent.ts <agent id>.