Software, apps, AI agents. We build tools that work for you.
Not just websites. We build custom software, mobile apps, and AI agents that talk to your customers, take bookings while you sleep, write seasonal menus and tell you what each dish actually costs. 2026 technology at the service of Italian hospitality.
- Built on Claude API + Anthropic SDK
- OpenAI · Voice agents
- Self-hosted EU
- GDPR-compliant
What we actually build
Three things, not forty. Software, apps, AI agents — all started from the same place: a real problem from a restaurateur we knew.
Custom software
Tailored back-office for your flow. Our Ordina was born this way, from one restaurant that wanted to know how much it actually made on each dish.
Mobile + web apps
iOS, Android, PWA. Punta Ala App: tourist marketplace, 9 months of build, 3 senior devs, PostePay in their portfolio.
AI Agents
Agents that talk to your customers, automate internal operations, write brand-coherent content. The newest part of our work.
8 real AI agent examples for hospitality
Things we can build for you — some already live in our products, some prototyped, some applicable to your venue. Those marked "Live" are running in production right now.
- On request
Voice agent for phone bookings
Answers the restaurant phone when staff is in the kitchen. Speaks Italian + English, takes the booking, checks live availability, confirms by SMS or email.
Stack: Twilio + Claude Sonnet 4.6 + existing Reservation API
- Live
Menu description generator
Give it the dish name and ingredients, get a contemporary Italian description, brand-coherent, max 80 words.
Stack: Claude API + few-shot brand voice training
- In development
Instagram event extractor
Post on IG as usual, the agent extracts date, location, price and title and populates the site's event calendar automatically.
Stack: Instagram Graph API + Claude Haiku/Sonnet with prompt caching
- Applicable
Multilingual menu chatbot
On the restaurant site, a chatbot answers 'Do you have gluten-free?', 'How much is dinner?', 'Can I book for 8?' — in IT, EN, FR, DE.
Stack: Claude Sonnet with live menu context
- In development
Lead scorer for outbound
Analyses prospects — site audit, geo, type, size — and ranks the most likely to convert. Suggests what to write to each one.
Stack: Claude API + Outreach module + Lighthouse data
- Applicable
Review responder with brand voice
Replies automatically to Google and TripAdvisor reviews keeping the restaurant's tone: calm, honest apologies, controlled wit when appropriate.
Stack: Claude Sonnet + brand voice training
- In development
Ingredient consumption predictor
Analyses 30 days of sales + weather + days of week, predicts consumption for the next 7 days. Suggests supplier orders.
Stack: Claude + cron + sales data + food cost module
- Applicable
Weekly social post generator
Every Sunday generates 3-5 brand-coherent Instagram posts from the week's menu, uploaded photos, upcoming events. You approve and publish.
Stack: Claude + Instagram Graph API + asset library
Software we've built
Real products, not slide decks. Everything below is built by us and either runs in our venues or is sold as SaaS.
Ordina
LiveOur B2B SaaS for Italian restaurateurs — table ordering via NFC, per-cent food cost, bookings and integrated POS.
Learn more →Punta Ala App
In developmentMarketplace for Punta Ala tourists — delivery, bookings, event tickets, beach club shops. 9 months of build.
Learn more →CRM Lead
LiveCustom multi-business lead management, plugged into all 5 business lines of Italy On Demand.
Outreach Bot
In developmentAutomated cold outreach for the web design service — audit + custom mockup + email.
Food Cost Engine
LivePer-cent dish cost with automatic supplier invoice import and real margin per dish sold.
How we work with AI Agents
- 1.
Discovery
A one-hour free chat to understand what you want to automate and whether an AI agent is the right tool.
1 hour · free
- 2.
Proof of concept
I build a working prototype you can hold. If it doesn't convince you, refund.
1 week · from €800
- 3.
Build & integrate
Production-ready build, integrated with your existing stack — back-office, site, phone, floor.
2-4 weeks
- 4.
Maintain
The model evolves, I keep the agent up to date. Marginal token runtime, depends on volume.
€2-30/month at scale
Transparent AI pricing
Same logic as our catering: declared prices, no surprises. Numbers below are indicative and confirmed at the site visit.
Single-agent PoC
A working prototype in a week. Proves the thing actually works before you spend more.
From
€800
Production-ready agent
Full build, integrated with your stack, monitoring and support. Most projects end here.
From
€3,500
Multi-agent system
Multiple agents working together — voice + content + analytics. Bespoke quote at the site visit.
From
€8,000
Anthropic token runtime is typically €2-50/month per agent, depending on volume. We give you the estimate at the PoC.
Indicative pricing — final quote comes after the free site visit. Nothing to sign upfront.
What we DON'T do
ChatGPT wrappers without value
If you just need a generic chat, use ChatGPT directly — cheaper, same job.
Generative art / spam content
No fake reviews, no AI photos of dishes that don't exist, no automation that fools customers.
Replace humans where it matters
The floor needs a real person. Creative kitchen too. AI does the repetitive, not the value.
Tech lock-in
The prompt and the code of your agent are yours. We give them to you. Switch vendor, switch.
Tech stack
Models
Claude Sonnet 4.6 (default quality), Claude Haiku 4.6 (volume). OpenAI fallback where it makes sense.
SDK
Official Anthropic SDK with native prompt caching to cut costs at volume.
Hosting
EU-only on Hetzner + Coolify. No customer data leaves Europe.
Compliance
GDPR + DPA with Anthropic — no training on our data, no fine-tuning with customer data.
Voice
Twilio for calls, ElevenLabs for natural Italian TTS, Whisper for speech-to-text.
FAQ
How much does runtime cost for an AI agent?
Anthropic tokens cost very little at single-restaurant volume: €2-50/month depending on usage. Prompt caching saves another 80% on recurring input tokens. We estimate it at the PoC stage.
Can I see a demo before paying?
Yes. The PoC is exactly that: one week, a working prototype. If it doesn't convince you, no invoice.
Does my customer data go to OpenAI or Anthropic?
No, and we have signed DPAs. Anthropic does not use our data for training. Data stays yours and in EU. For voice, ElevenLabs and Whisper are configured without retention.
What if Claude breaks or the API changes?
Your agent's code is yours, modular. If Anthropic changes the API, we update. If you want to switch to OpenAI or another model, the prompt is portable.
Will the agent replace my staff?
No. The agent does what your staff SHOULDN'T be doing: answering the phone while you're cooking, writing descriptions late at night, reading 200 reviews for sentiment. Your team makes the value.
Can I write the prompt myself or do you?
We write it together. You know how you talk to your customers, we know how to make a model talk. We need your voice, not a brief.
Does it work in Italian only?
Yes. In fact that's what we do best. Claude speaks contemporary Italian very well — better than GPT on tone and regional nuance.
Got an idea, an agent, a dream?
One free hour of chat. If a PoC comes out, we go. If nothing comes out, at least we've met.
Free site visit →