Rehearse your Customer Agent before shoppers do
A personalized agent for Fernway Outfitters, a fictional DTC brand whose customers and orders live in Klaviyo. Simulated shoppers stress-test it, a judge scores every conversation, and information gaps tell you what to fix.
Klaviyo profiles, orders and catalog arrive through the Klaviyo MCP server. Add policy docs, a store site or a CSV next to them.
Simulated shoppers, each built from a real profile and its order history, chat with the agent in parallel. A judge scores every conversation.
Failures cluster into information gaps. Add the missing source or edit the guidance, then rerun only what failed. Calibrate the judge against your own labels.
Latest rehearsal
Checking…
Not connected on this server. Rehearsals still run and are scored; connect Braintrust to get traces, experiments and score comparisons.
The agent can only read the signed-in shopper’s data, coupon limits are enforced in code, every reply is validated for citations and invented figures, and spend is capped per run and per day.