Voice agents that run on the ship, so they don't strain your satellite link, or your guests'.
From the team that's installed shipboard servers on some of the world's largest cruise ships.
We're actively pursuing partnerships with the teams behind these systems.
Cruise procurement cycles are long. Here's exactly what you can buy now, and what's coming, so your team can plan the full deployment, not guess at it.
A shoreside reservations agent that answers your central booking number. New cabin bookings, modifications, cancellations, pre-cruise upsells. No ship install, no satellite touch. Runs entirely in our cloud, forwarded from your existing PBX. 30-day free pilot. Live in two weeks.
The same agent, running on a redundant server pair in your ship's equipment room. Calls land on the ship and never leave. Sub-300ms voice loop, indifferent to the satellite link. Survives outages. Eliminates satellite cost on every guest call.
Drag the slider. Every line of the equation updates live. Conservative defaults, hardware bundled. All inputs sourced from public filings.
Loaded cost per crew · $30,000 / yr
payroll $15,300 · food / cabin $1,800 · berthing $3,000 · medical & insurance $1,500 · training $2,000 · repatriation $2,000 · uniforms & misc $4,400
Phone seats removed (per cabin)
Conservative · 1 per 200 cabins ·
Midpoint · 1 per 150 cabins ·
Aggressive · 1 per 120 cabins
Multiplied by 3 shifts to reflect 24 / 7 reservations & guest-service coverage.
FlowStay pricing (v1)
Software only · $9 / cabin / mo · agent stack, voice, memory, FlowCast, FlowVue
Software + on-prem hardware · $12 / cabin / mo · inference appliance amortized over 5 yr, install & commissioning included
Sources: Royal Caribbean Group 10-K (crew loaded cost), MLC 2006 minimum standards, U.S. Bureau of Labor Statistics occupational data, CLIA industry reports.
Vapi. Retell. Bland. ElevenLabs Conversational. Every one of them assumes a fat, always-on internet pipe sitting between the caller and a US datacenter. On a cruise ship that pipe is satellite, it's shared with thousands of guests, and it's the most contested resource on board. Putting voice agents on top of it isn't an architecture choice. It's a tax.
Cloud voice agents push every word, every model token, every tool call out over your satellite connection. The same pipe your guests are streaming on, the same pipe your POS, PMS, and bridge systems share. FlowStay's inference runs onboard, so the call never leaves the ship.
Satellite round-trips add 500-800ms before the model even starts thinking. Layer a cloud LLM on top and the guest hears dead air every turn. FlowStay's voice loop closes on the ship, sub-300ms end to end, indifferent to the link.
Satellite outages aren't rare, they're scheduled. Weather. Beam handoffs. Capacity caps. A cloud-hosted voice agent dies with the link. FlowStay keeps taking calls through outages, because the brain, the catalog, and the booking state all live onboard.
The brain, the catalog, the booking state, the voice models, all of it runs in a redundant server pair in your ship's equipment room. Calls land on the ship and never leave. The satellite link only carries the things that can wait: nightly catalog deltas and audit logs back to shore.
Sized for a 5,600-passenger ship: a redundant active/standby pair running 20-30 concurrent voice agents, a multi-gigabyte catalog, and all booking state. Standard enterprise hardware, rack-mounted in your climate-controlled equipment room. No exotic marine-spec gear required.
REFERENCE FOR ICON-CLASS · SMALLER SHIPS SCALE TO ~$80K/PAIR · L40S · NOT H100 · POWER-BUDGET SANE
Before FlowStay, our team designed, built, and installed on-prem compute on some of the world's largest cruise ships. Racks shipped, networked, commissioned, and kept running at sea. Not a slide. Hardware bolted into a server room above the engine deck.
On-prem voice for a cruise line isn't a research project for us. It's the same playbook, with newer hardware and a smarter brain on top of it.
When the brain runs onboard, the call gets shorter, smarter, and never depends on the link. Guest memory is already there. Catalog is already there. Booking state is already there. The agent picks up the phone the way a senior reservations specialist would on their tenth year, with everything they need already in front of them.
Past sailings, cabin preferences, dining tier, beverage package, loyalty status, shore-excursion history. FlowStay pulls the full guest profile before "hello" and keeps it warm for the entire call, no re-asking, no re-routing.
Deck plans, cabin photos, excursion previews, dining menus, beverage tiers, sent to the guest's phone in real time as the agent speaks. The call does what voice alone can't: let the guest see what they're choosing.
Multiple ships. Twenty cabin categories per ship. Hundreds of excursions per itinerary. Dynamic dining, beverage, spa, and Wi-Fi inventory. FlowStay holds the whole catalog in working context and reasons across it without losing thread, even at minute forty.
When the protocol calls, the brain answers locally. Spark and ChatGPT can plan an entire sailing from shore, FlowStay responds with live cabins, dining, excursions, and spa, and the agent on the ship handles the conversion the same way it would for a guest calling the bridge phone in cabin 7042. No satellite link required to think.
Spark, ChatGPT, Gemini reach your line directly through UCP and MCP. Your brand is the destination, not a row on an OTA.
Same agent runtime on the ship handles in-cabin requests offline. Protocol intent in, full sailing booked out.
A guest asks about the balcony stateroom. It appears on their screen, rendered, as the agent describes it. The deck plan. The specialty restaurant. The shore excursion they'll book for Tortola. Each on cue, no app, no download. Visual pitches close at 2–3× voice alone.
Your website becomes part of the call. The guest asks about cabin categories, dining, shore excursions, beverage packages, the spa. Each renders on screen as the agent speaks, choreographed to what they're hearing. Closer to a private at-sea preview than a webpage. Direct booking just got compelling.
Some sailings sell on a phone screen. A seven-night Caribbean cabin, a transatlantic suite, an Alaska family booking, do not. We turn the family TV into a private screening room. The agent narrates. The ship reveals itself at theater scale: drone reveals of the lido deck, the suite category in 4K, the shore excursion in Tortola, all choreographed to her voice.
Nine specialists, one shared memory. Cabins. Dining. Shore. Beverage. Spa. Photo. Wi-Fi. Transfers. Loyalty. When you're the easiest call to make, you're the only call they make. Onboard spend stays onboard.
FlowStay loads every guest's name, prior sailings, cabin preferences, dining choices, beverage packages, and open reservations before the call connects. Then resolves any request across cabins, dining, excursions, spa, or onboard packages in under 60 seconds. No hold. No transfers.
When repeat cruisers stop fighting friction, they call more and book more. Every sailing.
Test-drive our reservations agent on your central booking line. New cabin bookings. Modifications. Cancellations. The same agent you'd run the day you sign, sitting in your shoreside call center.
Pilot lives shoreside, not at sea.
Forward your booking line to a number we give you, or drop one script tag on your site. Catalog and tone tuning takes about two weeks.
<script src="flowvue.js"></script> *72 → forwarded. You're live. SIP/2.0 → sip.flowstay.ai We will build a working version for your line before you decide. No commitment. No contract.
Book a demo Or reach us at hello@flowstay.ai ↗