We explained in an earlier article why the list of technologies is not the important part of the answer. It is still owed, and here it is, as it stands today. You will find most of it in our privacy policy, which names our subcontractors: that is a legal obligation, and it is also a good place to check what an agency really uses.

What serves the site and the applications

  • Cloudflare for page hosting and server side execution, as close as possible to visitors.
  • Supabase (PostgreSQL) as the database and identity system, with access rules applied row by row.
  • Vite, React and Tailwind on the interface side. Nothing exotic: any developer can pick these up.

What makes the AI work

  • Models from Anthropic and OpenAI, chosen task by task. The choice stays reversible: our agents do not depend on a single vendor.
  • Our knowledge bases, built from the client's documents, so answers come out of their material rather than a model's memory.
  • Our deterministic rules, written in code and executed outside the model, which decide what an agent is allowed to do.

What touches money and messages

  • Stripe for payments and subscriptions. We store no card data.
  • Brevo and Resend for sending and receiving email, including agent mailboxes.
  • Twilio for telephony and text messages.
  • The Google Ads programming interface to read and modify the advertising campaigns we pilot.

Where the data sits

Our main databases are hosted in the European Union, in Paris for recent projects. Some of the services we use are American, with the corresponding contractual safeguards: it is written in our privacy policy, with the list and role of each. If strictly European hosting is a requirement on your side, say so during framing: it changes some choices, and it is feasible.

An agency that cannot say where your data lives cannot tell you who has access to it either.

What we do not use, and why

We did not build our production on a visual automation platform. Those tools are excellent to start with, and we recommend them in simple cases. For what we operate daily, we needed fine grained error recovery, checks before action and usable logs, and that dictated the choice.

We also took back in house, in July 2026, everything that depended on a third party platform, for the reason set out in the article on ownership: what lives at a third party is only half yours.