AI agents
Not a chatbot on your website. Agents that do real work inside your ledger, PMS or CRM: they triage, gather evidence and draft, within the permissions you give them, with every step logged and a person approving wherever a mistake would cost money or trust.
Where agents earn their keep
An agent triages KYC and screening alerts, gathers the evidence behind reconciliation exceptions and drafts payment investigations, so your team starts each case already prepared. A person makes every compliance decision and signs off anything that moves money.
An agent sorts guest messages, answers routine questions from your PMS data (Mews, Apaleo and others) and turns requests into housekeeping and maintenance tasks. Refunds, complaints and anything unusual go to staff.
An agent finds and sorts incoming leads, looks each one up and drafts a reply in your CRM. Nothing is sent until a person approves it. We run the same set-up for our own sales.
Safe by design
Air Canada had to honour a refund policy its website chatbot made up, and Cursor's support bot told users about a login rule that didn't exist. Both bots could say anything to customers with nobody checking. Ours can only do what their role allows, and a person approves what matters.
Proof
We don't have a client AI project we can publish yet, and we won't dress one up. We run AI inside our own products and tools every day: drafting replies to leads and emails that a person approves before anything is sent, and sorting incoming requests. The rules on this page are the ones we work under ourselves.
The systems these agents work inside are the ones we build and run: banking orchestration for a European payment platform and Mycarro's car-market data platform.
Questions
By default, Anthropic's Claude, under business terms that don't use your inputs for training. If your company has already chosen a provider, such as OpenAI, Google or models hosted in your own AWS, Azure or Google Cloud account, we build on that. Where it holds up in testing, we use a smaller, cheaper model for routine steps and a larger one only where it changes the result.
Before we build, we run the task on a sample of your real cases and measure what each one uses, then multiply by your volume at the provider's current prices. You see a cost per task and per month before you commit. In production, daily spending limits and alerts keep it there.
You do, from the first commit. We work in your repository and your accounts, and the contract assigns the IP to you in the present tense, not at some later date.
The engineers named in the proposal, and you meet them before you sign. A key-person clause means nobody is swapped without your written consent.
We work on Lisbon time, which is the same clock as London: Mon–Fri, 08:00–18:00 UK time. That leaves about four working hours a day in common with the US East Coast. You get a demo of working software every week and a short written update every working day.
A master services agreement with a statement of work for each project, or your own paper if you prefer. A mutual NDA before we see anything, and a GDPR data processing agreement. US clients get a W-8BEN-E from us. Which law governs the contract is answered on Contracts & security.
The fixed-price first step is paid upfront. Builds are billed by milestone, after you have seen the work running. Monthly care, if you want it, is billed at the start of each month. We invoice in euros, pounds or dollars.
Free 30-minute call
You talk to an engineer who would work on it, not a salesperson. You leave the call knowing what it would take and what the first step costs, or we tell you that you don't need us.