Rich content trust policy
Structured cards and carousels in Chatevo follow a strict trust policy: interactive offers, prices, and media appear only when they come from a successful tool execution mapped by result_transform. The assistant cannot fabricate commerce UI from documents or imagination.
| Allowed | Not allowed |
|---|---|
| Cards from HTTP 2xx tool JSON + valid transform | Cards from failed or timed-out tools |
| Prices/images from your API fields | LLM-invented product lists |
| Actions you define in transform config | Fake “Buy now” links to unrelated URLs |
Why it matters
Section titled “Why it matters”Visitors treat carousel cards like real inventory. Showing unverified offers would mislead customers and create compliance risk. Chatevo builds RICH_CONTENT on the server after tool success — the model never authors card payloads directly.
Failure behavior
Section titled “Failure behavior”| Tool outcome | Chat shows |
|---|---|
| Success + transform match | Cards + assistant message |
| Success + transform mismatch | Text only (check transform reference) |
| 4xx/5xx or network error | Text apology; no cards |
| Tool not called | No cards for that data |
Configuration implications
Section titled “Configuration implications”- Always attach tools to real APIs for catalog/offer UX.
- Use Test a tool with production-like JSON before go-live.
- Do not rely on the master prompt to “show products” without a tool.
Related security doc
Section titled “Related security doc”See also Rich content trust (security) for org-wide policy context.