Playground
Test your assistant with draft settings, simulated pages and customers, and read the debug panel for every tool call.
The playground is a chat with your assistant inside the dashboard. Conversations are never billed and never shown to shoppers.
Send a test messageControls
| Control | Use |
|---|---|
| Use draft settings | Test persona, instructions and guardrails before publishing (on when a draft exists) |
| Language | Any of the 11 widget languages |
| Page | Simulate a page type and a product |
| Consent | Simulate the shopper's consent choices |
| Simulate a signed-in customer | Enter a platform customer id to test order questions (admins; logged in the audit log) |
What you see
The playground renders replies like the widget: product cards, quick replies and forms. Cart actions show "Would add 1 × … to the cart" instead of changing a real cart, and confirmation cards run as dry runs ("Dry run: would cancel order #1042").
Debug panel
Per reply: the prompt version and model, the reply language, the tools offered, every tool call with its arguments, result and latency, grounding corrections, token usage, cost and time to first token. Billing always shows 0: playground conversations are never billed.
Limits
With managed AI, test replies are capped per day; at the cap you see "You've used today's test replies with managed AI" and when they reset. Playground use with your own key isn't capped by ShopAssist (your provider bills you).
Last updated on