An omnichannel AI front desk for hotels, built on Telnyx

Hospitality is changing fast. Guests book from their phones, at any hour, often from another country, and they expect the hotel to answer in their language, on whatever channel they picked. A front desk that works office hours in one language can't keep up.

So we built one that can. One Telnyx AI agent takes a guest from first hello to a confirmed booking: it talks, shows the rooms on screen, switches language mid-conversation and confirms by SMS and email.

Watch the demo

One call, two languages, from hello to booked. TelnyxHotel is fictional; the infrastructure behind it is not.

YouTube link.

Why it matters for hotels

Guests reach a hotel by phone, chat, text and email, in their own language and at any hour. Few front-desk teams can cover all of that, so some calls go unanswered and some guests book elsewhere, often through an online travel agency that charges commission.

The agent in the demo closes those gaps:

• Six languages. It speaks English, French, German, Italian, Spanish and Dutch, and switches the moment the guest does. It uses the formal form of address in each (vous, Sie, Lei, usted, u), and one knowledge base serves all six.
• Any hour. It answers straight away, day or night, and takes the booking from start to finish.
• The whole booking. It collects hotel, dates, guests, room, contact details and special requests, then confirms in writing.
• The hotel's own facts. It answers questions on rooms, dining, parking or accessibility from the hotel's knowledge base, and offers a follow-up when the answer isn't there.
• A handover for the rest. For a change, a cancellation or a complaint, it takes the guest's details for a colleague to follow up by email.

Every booking it takes is a direct booking, and staff keep the conversations that need a person.

Why the visual element matters

Hospitality is personal and visual: guests choose a room by what they see and trust a booking they can read. So the agent talks and shows at the same time, filling in the booking on the page while the guest speaks.

• Rooms on screen. Once the hotel, dates and guests are known, the available rooms appear with the nightly price and the total for the stay. The chosen room is highlighted.
• Details the guest can check. Dates, names and email addresses are easy to mishear on a call. Seeing them written out lets the guest catch a mistake before confirming.
• A clear path to done. A "Your stay" tracker moves from hotel to dates, guests, room, details and confirmation, so the guest always knows what's left.
• A face at the front desk. An optional video avatar, in beta, gives the receptionist a face, with the voice lip-synced and the video streamed over WebRTC.

The guest can type instead of talk at any point, and the page switches language with the conversation. Without the avatar, the same flow runs voice-only.

How it works

A small Node.js server sits between the guest's browser and one Telnyx AI agent, so the page stays live and the API key stays on the server.

The agent works through five tools, three of which run in the page so the guest sees each detail land:

The server re-validates every booking before it sends an email or text.

Inference on Telnyx GPUs

The language model behind the receptionist, Kimi K2.6, runs on GPUs that Telnyx owns, on the same platform that handles the speech. That keeps every turn of the conversation quick.

• Cost that fits a small chain. Open-source models on owned GPUs cost less per token than proprietary alternatives. That puts an AI front desk within reach of an independent hotel, not only a global brand. See inference pricing for current rates.
• Data that isn't kept. On the inference side, prompts and completions aren't stored. That comes from the architecture, not from a setting someone has to remember to switch on.

Build it yourself

The full demo is open source under the MIT license: team-telnyx/telnyx-demo-hotel on GitHub. It runs as one Node.js process, and a setup script creates the agent and its knowledge base on your Telnyx account.

What's in the repo

• Booking site: the hotel website with the voice receptionist, in six languages.
• How it works page: an animated architecture diagram that follows a live call step by step.
• Call history: each call's conversation with per-turn latency, tool calls, events and the messages it sent.
• Back office: users and access; the agent's languages, voices, prompt, knowledge base and video avatar; email, SMS and WhatsApp settings.
• Ten fictional hotels: in Chicago, Rome, Paris, Munich, Madrid and Amsterdam, each with a knowledge base file.

The repo also ships end-to-end tests: npm test boots the real server on a throwaway database and checks access rules, isolation, cookies, validation and rate limits.

Run it locally

You need Node.js 22.5 or later and a Telnyx API key from the Telnyx Portal.

git clone https://github.com/team-telnyx/telnyx-demo-hotel.git
cd telnyx-demo-hotel
npm install
cp .env.example .env    # add TELNYX_API_KEY, or let setup ask for it
npm run setup           # creates the knowledge base and the AI agent
npm start               # http://localhost:3000