How we use AI on calls

Louneh is an AI receptionist. When a caller dials a phone number our Customer has connected to the Services, an AI agent answers, listens, asks clarifying questions, and can collect an appointment request, take a message, or attempt a configured human handoff. Live calendar booking and messaging depend on the Customer’s verified integrations. A saved appointment request is pending until confirmed; a queued message is not proof of delivery.

The AI identifies itself as an automated assistant when it answers. If a caller asks “am I talking to a person?” — or anything substantively similar — the AI must confirm it is an automated assistant. This instruction is part of the agent prompt and must be reviewed and tested before live use.

Recording and transcript storage depend on the agent configuration. Before live use, the Customer must review its disclosure script and recording consent setup. Synthetic test agents use restricted storage and may have no retained recording or transcript. Louneh does not automatically determine a caller’s jurisdiction or verify recording consent from the phone number.

The current operational configuration uses a 30-day retention period for both recordings and transcripts. Provider deletion failures are retried and can delay remote deletion. Callers can contact the Customer about access or deletion requests.

• Greet the caller and identify itself as an automated assistant.
• Answer factual questions about the business (hours, location, parking, services list, policies) from a knowledge base the Customer controls.
• Check live availability when a verified calendar integration supports it; otherwise collect a preferred appointment time for staff review.
• Save appointment requests, and report a confirmed booking only when the booking tool verifies it. Changes or cancellations can be recorded as messages for staff.
• Queue supported SMS messages when the Customer has an approved sender and the required consent. Delivery depends on the messaging provider.
• Attempt a configured human handoff when the destination and hours have been verified.
• Take a clear message and notify the Customer through the dashboard.

• Give medical, dental, veterinary, legal, financial, or other professional advice.
• Diagnose a condition, recommend a treatment, or quote a dosage.
• Tell a caller they would be a good candidate for any procedure.
• Discuss anyone else’s appointment, payment, or chart.
• Process a payment over the phone (we use SMS deposit links, never card-over-voice).
• Negotiate price, discount, or settle disputes — those go to a human.
• Claim to be a human when asked directly.

These restrictions guide the agent through its prompt and the configured vertical guardrails. Prompt instructions alone do not guarantee an AI response; live use requires script review, testing, and appropriate staff oversight.

The AI can attempt a handoff to the Customer’s configured destination when the caller asks for one, when the situation is outside the AI’s safe scope (clinical questions, complaints, suspected emergency), or when the AI’s confidence is low. When transfer is unavailable before an attempt, the AI can offer to save a message in the Customer dashboard. An attempted transfer does not guarantee that a person answers or that a busy or unanswered call returns to the AI.

Louneh staff sample call recordings and transcripts to monitor agent quality, detect regressions, and tune prompts. Reviewers operate under access controls and audit logging; PHI is redacted from any internal analytics or alerting pipeline before it reaches a reviewer’s screen where appropriate.

We do not use Customer call audio, transcripts, or PHI to train any public foundation model. We do use anonymized, aggregated quality metrics (response time, transfer rate, booking conversion) to improve the Services. Anything more specific — for example, fine-tuning on a Customer’s own scripts — happens only with that Customer’s written consent and remains in their tenant.

We track and comply with state-level AI disclosure requirements that apply to automated voice agents (for example, California SB 1001 for bots, Utah’s AI Policy Act). If a Customer’s jurisdiction requires a specific disclosure script, our agent prompt incorporates it.

Email trust@louneh.ai. If you are a caller and you want a Customer’s call recording deleted, contact that business directly — they own the data and we will assist them in honoring your request.