Live · pilot phase
For a small business, every missed call is a customer who rings the next search result. We built an AI assistant that texts the caller back within 10 seconds, answers their questions and books a free slot.
Try it at bevetel.ai01 · The problem
The hairdresser is mid-colour, the dentist is drilling, the masseur is with a client. Meanwhile the phone rings and nobody picks up. The caller leaves no message: they ring the next result on Google.
Calling back is no fix, because by the evening the caller has booked somewhere else. And hiring a receptionist doesn't pay for a business of 2–5 people.
02 · What we built
A Claude-based SMS assistant that writes natural Hungarian, answers questions and arranges appointments. Professional limits (for medical questions, for example) are set per industry.
The business signs up, gets a number, sets up call forwarding on its own phone (mobile, landline or VoIP) and checks it with a live test call.
Caught calls, conversations, bookings and recovered revenue in real time, with a pause switch.
Stripe subscriptions with a customer portal, transactional e-mails, and an in-house sales CRM with its own lead finder.
03 · How it works
01
Call
not answered → forwarded
02
Telnyx
Hungarian voice number, hangs up in ~1 s
03
Engine
Cloud Run, EU region
04
Claude
writes the reply
05
Infobip
two-way Hungarian SMS
06
Calendar
only slots that are really free
07
Firestore
live dashboard, EU region
04 · The hard decisions
We split the channel in two: Telnyx takes the call, Infobip sends the SMS. The providers sit behind a shared layer, so an outage is a matter of configuration, not a rewrite. Trying a US number was a dead end: Hungarian networks rewrite the sender, and the conversation splits in two.
The model can only offer what the calendar shows as free. If no calendar is connected, it doesn't book; it records a request, and the owner approves it.
We measured it: an AI reply costs ~1.8 Ft, an SMS segment ~22.6 Ft. Hungarian text with accents fits only half as many characters into a segment, yet we kept the accents, because without them the message reads as machine-made. A per-conversation message limit keeps the cost in check.
We moved the industry rules (what it may say and what it may not) out into a table. A new industry is therefore a row, not a new product, and the switch happened without a single database migration.
We looked at a “pay when the customer shows up” model and rejected it: we would have had to process health-related attendance data (GDPR Article 9). Instead we have conversation-based plans that follow the real cost.
05 · Build log
Decision: the missed call is the most measurable problem. First target: clinics.
First commit, and the first real SMS arrives on a phone.
The dashboard shows live data: conversations, bookings, metrics.
Self-service setup: the business reserves a number and tests it on its own.
Stripe subscription and a live demo on the website, with a cost cap.
A cloud-deployable engine with more than one provider (Telnyx, Infobip).
It outgrows dentistry: now called Bevétel, for any small business.
Our own cloud infrastructure in the company's name, deployed without touching a console.
The assistant can be tried before entering a bank card.
06 · Tech stack
07 · Where it stands now
The system runs in production at bevetel.ai, and the first businesses are coming on board now. The next step is voice replies, which we will only offer once they are good enough in Hungarian too. What we can't do well, we don't sell.