The 24/7 reception desk that lives on your existing PBX.

Why hotel phone systems are broken in 2026.

Three pieces, one install.

A short-let operator with 12 apartments.

12 apartments across two cities, three of them on Booking.com, the rest on direct bookings. The owner answers the master phone, but bookings drift to OTAs whenever calls go unanswered. Existing PBX is a single FRITZ!Box at the office.

• One DID per property → one AI persona per property, with the right city, language and house rules pre-loaded.
• Caller asks “is the apartment near the metro?” — the AI knows the address and the closest stop, and answers in seconds.
• Caller wants to book direct — AI captures the dates and emails the owner before they hang up.
• Edge case: caller needs the owner urgently — AI transfers to the owner’s mobile during defined hours.

Direct-booking ratio shifts. Booking.com commission falls. The PBX did not change.

Hostaway voice-AI plugin — one-click install for vacation-rental operators.

If your properties are already in Hostaway, you do not need to build the integration above by hand. Paste your Hostaway Account ID + API key once, link each listing to a CodeB virtual number, and every guest call lands with the live reservation context: which property, who is arriving today, the door code, the WiFi password, the house rules. BYOC + BYOG — bring your own SIP trunk and your own Google Gemini API key; CodeB charges only the plugin licence.

• Inbound: AI overlay per call with live reservation lookup by guest phone number.
• Outbound: 24h-pre-arrival reminder calls scheduled automatically on each new booking webhook.
• Encrypted: SIP-over-TLS signalling + SRTP-encrypted audio end to end (mandatory, not optional).
• EU-sovereign: Malta-hosted, GDPR-clean, transcripts stay on operator-controlled disk.

Two products solve different problems.

Two paths, both inexpensive.

Smoke or noise spike at 02:14? The apartment’s phone rings before housekeeping wakes.

Short-let operators increasingly fit each apartment with IoT noise meters (Minut, NoiseAware, Roomonitor) and smart smoke detectors. They all do the same thing well — threshold tripped → webhook fires. What they don’t do is actually reach the guest in time. A push notification to an app the guest never installed isn’t a phone ringing next to the sofa.

CodeB plugs into that webhook. One HTTP POST from your sensor platform → an AI voice agent dials the guest’s mobile in seconds, in the language you set per booking, with the context the AI needs to be useful: which apartment, which sensor, what time, house-rule reference. If the guest doesn’t answer, the second attempt rings the host. Whole flow, zero human-in-the-loop, audit trail per call.

Your webhook receiver does the heavy lifting in maybe ten lines. The CodeB side is one POST:

curl -X POST https://phone.yourtenant.com/api.ashx/v1/calls \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "phone":        "+4915157610183",
    "displayName":  "Apt 14 noise alert",
    "email":        "ops@yourcompany.example",
    "systemPrompt": "You are calling the guest in Apartment 14 (Vienna, Mariahilf). The in-unit noise sensor measured 78 dB sustained for 12 minutes between 23:50 and 00:02. House quiet hours are 22:00-08:00. Politely ask them to bring it down. If they apologise and agree, say goodbye. If they push back, say a member of the team will follow up by message. Do not threaten eviction. Speak in the booking language: German.",
    "apiKey":       "YOUR_AI_API_KEY",
    "language":     "de-DE",
    "maxSeconds":   90,
    "retries":      1
  }'

When the call ends, CodeB POSTs outbound-ai.finished back to your receiver with status, duration and the transcript filename. Your PMS / channel manager records the contact attempt on the booking and the audit log writes itself.

Full step-by-step with curl, Python and Node.js examples lives at schedule-ai-call.html. The same pattern powers payment-reminder calls, check-in confirmations, review-request follow-ups — anything where a system event should produce a human-feeling phone call within seconds.

Recipient consent. Outbound AI calling is regulated in most jurisdictions (GDPR / ePrivacy in the EU, TCPA in the US, plus local rules). Capture consent at booking (it is almost always already in your T&Cs as “emergency or safety contact”), honour opt-outs, and document the legitimate-interest basis for the call. The platform places the call you ask it to — consent capture is on you.