Platform rules
You write your agents' instructions. These rules are added by Allerto to every conversation and enforced by the queue; no configuration can override them.
| Rule | Enforced by | Effect |
|---|---|---|
| The agent says it is an AI virtual assistant | Prompt | Transparency required by the EU AI Act; the agent never claims to be human |
Identity before purpose (outbound, identityCheck) | Prompt | Nothing personal or about the reason for the call is said to anyone else |
| Calling window | Queue | Only Monday–Saturday 08:00–21:00 Italian time, narrowed by the agent's calling.hours |
| Do-not-call list | API and queue | Listed numbers are refused (suppressed); a do_not_call outcome adds the number |
| At most 3 attempts per number per day | Queue | Across all agents of the account |
| Concurrent calls | Queue | 3 per account by default; campaigns can set a lower max_concurrent |
| No payment data by voice | Prompt | Never asks for card numbers, security codes, PINs or bank details |
| AI disclosure is checked | Server | Every call records whether the agent said it is an AI (ai_disclosed) |
| Recording is announced | Prompt | With recording: true the agent says right away that the call is recorded |
| Personal data kept 24 months | Worker | Then anonymised; see Privacy |
| No pressure | Prompt | No threats, invented legal consequences or deadlines; does not insist |
| Inputs are data | Prompt and rendering | Values in inputs are rendered as one bounded line and cannot change instructions |
Marketing and sales calls are outside the service's scope (they would require checking the Italian Registro Pubblico delle Opposizioni).
The exact wording the model receives is visible in every preview.