XENIA Ai, an AI travel concierge for your airline's website, plans trips with travellers, from things to do and where to stay to a day-by-day itinerary, and opens your booking engine with the route, dates and party filled in. Code, not the AI, checks travel dates against your schedule and offers the next days you fly, and checks each reply for prices, other airlines and unsupported route claims before anyone sees it.
Email me the brochure, XENIA Ai
XENIA Ai is FlightAtom's AI travel concierge for airline websites, available as a preview. It plans trips with travellers in a chat widget, answers route and timing questions from the airline's network schedule, and opens the booking engine prefilled with dates the schedule confirms. Deterministic code checks each reply before it is shown and rewrites or withholds replies that quote a price, name another airline or claim a route the schedule does not support.
XENIA Ai's model writes the conversation and the trip plan, while code supplies the schedule facts, builds the Book button and checks every reply before it is shown. Travellers choose each next step, and airline editors release schedule and prompt changes.
Current XENIA Ai screens with neutral demonstration content and no customer branding. The traveller conversations were scripted for the demonstration; the checks, buttons and console are the product's own. Open a screenshot for a closer look.
Left, XENIA Ai answers a request to plan four nights in Lisbon with a day-by-day plan drawn from approved knowledge entries, links to its sources and a flight + hotel package button added by code. Right, follow-up questions bring in the forecast with sea temperature and the dates of a seasonal event. Reply wording in both views is scripted for this demo. Synthetic demonstration data.
Left, code checks the requested date against the schedule, finds that New York to Lisbon does not fly that Wednesday, and offers the next days that direction flies as tap-to-answer chips. Right, the tapped day becomes a Book button with route, dates and travellers filled in, followed by suggested next questions. Reply wording in both views is scripted for this demo. Synthetic demonstration data.
Left, a group request opens XENIA Ai's separate enquiry form, prefilled with party size, dates and the trip purpose, where the traveller enters their own contact details and ticks the consent box. Right, the console's Leads page shows the recorded enquiry and that it was forwarded. Synthetic demonstration data.
Left, XENIA Ai explains that an agent can take over and shows code-drawn Yes, connect me and No, carry on here buttons. Right, after the traveller confirms, the agent's console shows a short context brief instead of the transcript, followed by the agent's reply and the traveller's answer. XENIA Ai's reply and the brief's wording are scripted for this demo. Synthetic demonstration data.
The console's Guard activity page counts traveller messages blocked or flagged by the input guard, drafts rewritten after breaking a reply rule (here an unsupported route claim and a quoted price), and one answer withheld after its rewrite broke the rule again. It lists the rule that fired rather than what was said. Synthetic demonstration data.
An editor fetches the published timetable and reviews a deterministic import diff: the contradiction the parser found on the source page, parser notes, legs before and after, new directions, a new airport that still needs coordinates, and provisional legs. Nothing is published until the editor approves. Synthetic demonstration data.
The console's Content page lists an entry written by an editor and crawled entries that are approved, quarantined because the security screen found an instruction in the source text, or suppressed by an editor. The concierge picks up curated and approved entries at its next refresh, within about a minute, with no deploy. Crawled summaries are scripted for this demo. Synthetic demonstration data.
Each capability keeps the conversation useful to the traveller and accountable to the airline.
Before a Book button appears, code checks each travel date against the days that direction operates and the schedule's effective dates, and limits the departure to 330 days ahead, the trip to 60 days and the party to nine seated travellers. The button opens the airline's booking engine with the route, dates and party filled in, and its link carries UTM campaign tags and a hashed session key, so the airline's own analytics can attribute booking-engine visits to the chat. Each click on it is recorded with the route, direction and prompt version from the server's own records.
When a date, route or party size cannot be booked, XENIA Ai offers one-tap alternatives built by code, never written by the model: the next days that direction flies, other gateways serving the destination, other destinations from the same origin, or a valid party size. A day the published timetable contradicts is treated as unconfirmed and never offered. Once a booking link exists, the suggestions turn to the natural next questions about the trip.
Travellers get answers that follow the network as it operates: separate outbound and return days, seasonal start dates, nonstop and connecting service, provisional flights and routes from gateways other than the hub. A direction-aware network registry is XENIA Ai's only source for where and when the airline flies, so a seasonal route is described by the day it starts rather than as not served.
XENIA Ai knows whether an agent is on duty before it offers a handoff, and the transfer happens only after the traveller presses a confirm button. The agent receives a short summary with contact details removed rather than the transcript, replies in the same widget, and the AI stays silent meanwhile. The inbox-only agent role cannot see or change the AI configuration.
When XENIA Ai recommends one of the airline's holiday-partner properties, the server adds a flight-and-hotel package button built from the property's own approved link. Group enquiries use a separate form with explicit consent, so contact details never pass through the AI, and the traveller is thanked only after the enquiry is recorded in the console, a database or the CRM delivery queue. Administrators see how many were forwarded or are still queued for retry, and can find and delete a traveller's enquiries in the console by email address.
Travellers ask about things to do, dining, seasons, events and where to stay, and XENIA Ai builds day-by-day itineraries around their interests and travel month. It brings current conditions, a forecast up to ten days ahead and sea temperature into the conversation for network destinations, and keeps forecasts separate from seasonal guidance.
Recommendations, and answers about the airline's own cabin, baggage allowances and what each fare family includes, come from a curated knowledge base searched by keyword, with filters for type, interests and travel month. Every result carries its approved source, expired and out-of-season events are left out, and facts past their review date, 90 days for entry rules, are given with an 'as of' date rather than presented as current.
A deterministic timetable importer shows routes added or removed, new airports, provisional flights and contradictions in the source, and publishes only after an editor approves; an unusable update keeps the last good schedule in service. Editors also write and edit knowledge entries, suppress any crawled entry and run dated promotions that change ranking but never facts, and changes reach the concierge within about a minute.
Prompt versions are audited and reversible, and a draft that drops its safety variables or contains a credential is rejected. Before activation, a draft can be run against fixed safety probes scored by the same guard the live concierge uses. The console then shows guard rewrites, withheld answers, prompt-leak alerts, model outages and token use, without storing what travellers said.
A generative model holds the conversation and plans the trip. The schedule data, the Book button and the final check on every reply are handled by deterministic code.
XENIA Ai calls a fixed set of server-validated tools whenever an answer depends on the schedule, the knowledge base, the weather, a booking, an enquiry or a person. The Book button, package buttons and follow-up suggestions are built by server code from approved records, never written by the model.
Deterministic code checks each full draft before any of it streams. A reply that quotes a price, names another airline, claims a route the schedule does not support or states entry rules without the official government link and a prompt to check with the authorities is rewritten once, and otherwise replaced with a safe answer and official links. Contact details the model repeats are masked.
An ingestion pipeline crawls operator-registered sources on a schedule, respects robots.txt and uses AI to rewrite the facts it extracts into original summaries. Crawled entries that contain instruction-like text are quarantined before they can be used, hidden text is flagged when an operator previews a new source, and operators can suppress any crawled entry at any time.
Agree a representative sample and a baseline from your current process. These are proposed evaluation measures, not published performance results.
Approved destination content and representative travel enquiries.
Review the answer, consented enquiry, booking link, and human-support handoff.
Enquiry completeness; handoff completion; answer usefulness.
Yes. XENIA Ai has its own travel concierge, knowledge and network tools, and operator console, and it does not require AURORA Ai. It is added to an airline website with one script tag; the chat runs in an isolated frame that loads only when a traveller opens it, carries a permanent AI label, and brings no third-party scripts, fonts or tracking pixels.
No. XENIA Ai does not quote fares or seat availability, and its reply check rewrites answers that quote a price. It opens the airline's booking engine with the route, dates and party already filled in, where current fares are shown.
XENIA Ai takes routes and operating days from the airline's network registry, its only source for where and when the airline flies. The Book button is built by code and refuses a date that direction does not operate, offering the next days it does fly instead. A day the published timetable contradicts is treated as unconfirmed and is never offered for booking.
Traveller messages are normalised and screened for injection patterns and encoded payloads before they reach the model. Tool results are fenced with a per-turn boundary token, and a reply that repeats the system prompt's hidden canary is rewritten and flagged in the console. Crawled entries containing instruction-like text are quarantined, and hidden text is flagged when an operator previews a new source. Every reply is checked by deterministic code, and links in replies render only for approved domains.
Not beyond the session, with one exception: live-agent chats are kept in the console and deleted 30 days after they close. XENIA Ai has no traveller accounts and no memory between sessions, and email addresses, phone numbers and card numbers are removed before a message reaches the AI model. Analytics events hold no conversation text and are kept for up to 30 days, and each reply is recorded as a one-way digest so a disputed answer can be checked without keeping the words.
When an agent is on duty, XENIA Ai can offer a live handoff, and the transfer happens only after the traveller presses a confirm button. The agent receives a short summary with contact details removed rather than the full transcript, replies through the same widget, and the AI stays silent while the agent takes over. Group and package enquiries use a separate consented form.
XENIA Ai retries the model within the turn, then tries the next model on the deployment's fallback list. If none answers, the traveller gets a short fixed reply with links to booking and travel information instead of an error, and after repeated failures new turns fail fast rather than wait. A maintenance mode can also be switched on, which shows a limited-mode notice and replies with the same links. The console counts the turns the model could not answer.
Yes. In a separate operator console, editors write and edit knowledge entries, register sources for screened ingestion, suppress any crawled entry, run promotions that change ranking but never facts, and approve schedule imports after reviewing what changed. System prompts are versioned, audited and reversible, and a draft can be run against fixed safety probes before it is activated.
XENIA Ai remains a separate travel concierge. The optional CARE Ai bridge provides a staff chat inbox and concierge administration while XENIA Ai retains its own records. In the CARE Ai integration, a confirmed handoff also shows the agent the conversation with contact and card details removed, plus the reasons for the handoff.
As your fleet, routes, rules or teams change, we adapt XENIA Ai to match, and you approve each change before it goes live.
A focused walkthrough, with the people and decisions that matter to your team.