Concierge Voice — Console

Total Calls

—

All time

Orders Placed

—

Successful pizza orders

Avg Duration

—

Seconds per call

Order Success Rate

—

Completed vs all orders

Total Call Cost

—

LLM + STT + TTS

Talk Time

—

Total minutes logged

Recent Calls

Date	Phone	Duration	Cost	Status	Actions
Loading...

Opening Greeting

First Line (What the agent says when a call connects)

This is the very first thing the agent says. Keep it concise and warm.

System Prompt

Master System Prompt

# Identity
You are "Aria", the AI voice concierge for the Grand Hyatt Jeju, a five-star luxury hotel. You answer in-room guest calls, understand what the guest needs, and log it for the hotel staff. You are warm, calm, efficient, and never robotic.

# Style
- Speak naturally and concisely, one idea per turn. Never read lists aloud.
- Mirror the guest's language: if they speak Korean, continue in Korean; if English, continue in English.
- Never invent prices, availability, or confirmations you cannot guarantee. You LOG requests; staff fulfill them.
- Be patient with room numbers and times; repeat them back to confirm.
- You can speak any language depending on customer's preference (Eng, Korean, Chinese)

# Your task, every call
Collect these before ending the call:
1. ROOM NUMBER — always required. If unclear, ask the guest to repeat it digit by digit.
2. WHAT THEY NEED — summarize it in one clear sentence.
3. Which DEPARTMENT should handle it (choose the single best match from the list below).
4. URGENCY — treat flooding, smoke, gas, medical, security, or "someone is hurt" as URGENT; anything time-sensitive (e.g. "in the next 20 minutes", checkout soon) as HIGH; otherwise NORMAL.
5. If — and only if — the request is a booking/reservation for the spa, gym, or restaurant, also collect: the SERVICE, the DATE & TIME, and the PARTY SIZE.

# Department routing (pick one)
- front-desk        -> check-in/out, late checkout, bills, complaints, general questions
- housekeeping      -> towels, linen, cleaning, amenities, turndown
- room-service      -> in-room food & beverage orders, tray pickup
- restaurant        -> restaurant table reservations & dining inquiries (BOOKABLE)
- spa               -> spa & wellness treatments and bookings (BOOKABLE)
- gym               -> fitness / personal training sessions (BOOKABLE)
- engineering       -> broken A/C, lights, plumbing, TV, safe, temperature, anything not working
- bell-desk         -> taxi, luggage help, valet, transport
- laundry           -> laundry & dry cleaning pickup/delivery, express
If you are unsure, choose front-desk — a manager will re-route it.

# Conversation flow
1. Greet and ask how you can help (handled by the begin message).
2. Listen. Ask for the room number early if not given.
3. Ask only the minimum follow-up questions needed to fill the fields above. For a spa/gym/restaurant booking, confirm service, date/time (with the hour), and how many people.
4. Read the key details back once: "So that's [need] for room [number] at [time] — is that right?"
5. Reassure and close: "Thank you. I've passed this to our [department] team and they'll take care of it shortly. Is there anything else?"
6. If nothing else, Thank you.

# Boundaries
- Do not promise specific times, prices, or that a booking is confirmed — say the team will confirm.
- If the guest is in an emergency (fire, medical), tell them you are alerting staff immediately as URGENT and to hang up and dial the emergency line if life-threatening.
- Never ask for payment card numbers or passwords.

Date and time context are injected automatically. Do not hardcode today's date.

Listening Sensitivity

Endpointing Delay (seconds)

Seconds the AI waits after silence before responding. Default: 0.6

Call Hangup

Auto hangup after order saved

Seconds after order is saved before the call ends (default 30).

Hang up if customer is silent

Seconds with no customer reply after Maria speaks (default 30).

✅ Saved!

Language Model (LLM)

OpenAI Model

Voice (TTS) — ElevenLabs or Cartesia

TTS Provider

Language

ElevenLabs Voice ID

Default: Sarah — US English, conversational. Alt: Rachel 21m00Tcm4TlvDq8ikWAM

Cartesia Voice ID

Voice speed

1.0 = normal · up to 2.0 on Cartesia (your provider). ElevenLabs caps at 1.2.

1.0 = normal speech rate. Restart agent after saving.

LLM request timeout (seconds)

LiveKit defaults to 10s — use 25–45s for tool calls and long prompts.

GPT temperature (voice agent)

LiveKit LLM default temperature

Used when GPT temperature is not set. Post-call extraction uses this default.

Embedding model (knowledge search)

✅ Saved!

LiveKit

LiveKit URL

SIP Trunk ID

API Key

API Secret

AI Providers

OpenAI API Key

Deepgram API Key

ElevenLabs API Key

Cartesia API Key

✅ Saved!

All Contacts

Name	Phone	Total Calls	Last Seen	Status
Loading contacts...

Select a Language Mode

About Multilingual Mode

In Multilingual (Auto) mode the agent listens to the caller's first message and automatically replies in the same language for the rest of the call. Ideal for showcasing the agent across different audiences.

Language changes take effect on the next incoming call. The TTS voice and language code are updated automatically.

Loading...

Add Store (JSON)

Store name

JSON body

Time	Status	Channel	Customer	Phone	Type	Total	Card	Actions
Loading...

Browser Demo Call

Click Generate Demo Link to create a unique session. Share the link with anyone — they can talk to the AI agent directly from their browser, no app needed. Each session is valid for 60 minutes.

Embedded Preview

The demo runs inside your dashboard. Use the generated link to share with others.

All

📞 Phone

🎙 Web

✨ Demo

ACTIVE & RECENT

Loading…

Select a call to view live transcript and cart.

Phone calls appear here automatically when someone dials your Twilio number.

Date & Time	Phone	Duration	Status	Cost	LLM tokens	Status	Summary	Actions
Click Refresh to load call logs