Reserve Voice 2026: how the voice AI takes phone reservations in 45 seconds

What the guest actually hears on the other end

The guest calls your regular number. When you do not pick up in the first two rings, the call is automatically forwarded to Reserve Voice. The AI opens with: good evening, restaurant [name], may I help you with a reservation. Natural voice, normal intonation, relaxed pace — no language button, no robotic tone, no press 1 for English.

The guest says table for four on Friday at eight. The AI replies: table for four on Friday the 8th at 8pm, may I have your name and phone number for the confirmation. Total call takes 45-90 seconds. The reservation appears in your dashboard in real time, the guest gets SMS confirmation within 30 seconds. No manual entry, no double-booking.

Under the hood: Gemini 2.5 Flash Native Audio

Reserve Voice runs on Google Gemini 2.5 Flash Native Audio — the most advanced speech-to-speech model from Google. The difference from older generations: audio is no longer translated to text before processing. The model receives audio directly over WebSocket (16 kHz input) and sends audio directly back (24 kHz output). Without transcript-to-LLM-to-TTS steps, latency drops below 500 ms for language detection and precise reaction.

Function-calling access: the AI has in context your current floor plan, opening hours per day, special rules (groups over 8, allergies, special menus), and POS integration for sync. That means the AI never accepts a reservation for a non-existent table, nor for a slot when the restaurant is closed. All of that is read in real time, not from a cached snapshot.

Three languages in one call, no configuration

The biggest specialty for the Croatian market and coastal restaurants: the AI auto-detects the guest language in the first second and switches. A Croatian guest calls in Croatian, a German tourist in the same minute gets a German reply, an English-speaking guest English. No menu gating, no header prompts. Gemini 2.5 supports 70+ languages and 2000 language pairs for real-time translation.

This delivers concrete value to coastal restaurants in Istria, Dalmatia, and Kvarner where typically 60% of summer guests come from the DACH region. A restaurant with manual staff that is not fluent in German loses or poorly processes 30-40% of those calls. Reserve Voice has native understanding of Austrian and Bavarian German variants, as well as Bavarian-specific food terms.

Where the practical limits are and what the AI hands to humans

Reserve Voice works perfectly for reservation intake, cancellation, and basic queries about opening hours. What the AI does not do: detailed menu recommendations at sommelier level, wine pairings for a 12-course tasting, food complaints, agreements for special events with 50+ people. For all of that, the AI auto-forwards to staff or sends an SMS with call details and callback number.

Statistically: 90% of calls succeed fully automatically. 8% finish successfully but with one repeat (guest has a strong dialect or background noise from a car). 2% of calls require human takeover. When the AI judges a human is needed, it transfers the call in real time to your number — but only during business hours. Outside business hours, the AI schedules a callback for the next available slot.

Pricing tiers and 1-3 day implementation

Three tiers: Starter 49 EUR monthly for 200 calls (small restaurant, weekend peaks), Pro 99 EUR for 600 calls (medium restaurant, summer season), Enterprise 149 EUR for unlimited (chains, peak season). Prices include telephony, AI model, support, and unlimited covers — no per-cover fees, no service charges, no hidden costs.

Implementation in three days. Day 1: number setup and call forwarding from your existing landline (all major EU telcos work). Day 2: training the AI on your menu, opening hours, special rules — all via a single tabular form. Day 3: test calibrations with your staff, fine-tuning tone of voice and complaint handling. Day 4 morning: AI is live on your number.

Frequently asked questions