Three levels of French — the cheap version, the medium version, and the one Linkette ships. With the SIRET number, the sub-processors, and the gaps we still have.
The phrase made in France, applied to a piece of software, has been doing a lot of work lately. There is a generation of European SaaS companies, mine among them, that places the tricolore on the landing page and the fabriqué en France badge in the footer, and the meaning of those marks varies considerably depending on who is wearing them.
I want to lay out three levels of what made in France can mean for a SaaS product in 2026, because the consumer — the creator, the founder, the press desk at Sifted or Maddyness — has every right to ask which level a given company actually inhabits. I'll describe each, name where I think Linkette sits, and be honest about the gaps that remain.
Level one: the cheap version
The cheap version of made in France is: the founder is French, or the company is incorporated in France, and everything else is American.
This is the most common shape, and it is not without value — a French controller subject to French law, paying French taxes, employing people in France, is a real contribution to the local economy. But the data — the user accounts, the content, the AI inference, the analytics, the email — flows through US infrastructure. The database is on AWS or Vercel-managed Postgres. The AI calls go to OpenAI or Anthropic. The analytics is Google Analytics. The emails are Mailchimp or SendGrid. The CDN is Cloudflare.
The label made in France in this configuration describes the company's letterhead, not its data path. If a US sub-processor's terms change, if a US court issues a subpoena under the CLOUD Act, if a US foundation model decides to use customer prompts for training by default, the French letterhead does not protect the creator on the platform. The data is on US-controlled rails.
I am not in a position to throw stones. I have built products at this level before. The infrastructure was cheaper, the tooling more mature, the DX better. The decision to ship at level one is not lazy; it is rational, given the current cost curves. I am only saying that the label, in this configuration, is doing a lot of work it cannot fully back up.
Level two: the medium version
The medium version is: the founder is French, the company is incorporated in France, the infrastructure is European, but the AI is still American.
Several promising European SaaS companies sit here. Hosting on Scaleway or OVH. Database on a European Postgres provider. Email through Brevo (French) or Postmark EU. But when it comes time to call a foundation model — for a writing assistant, a summarization feature, a chat surface — the call goes out to OpenAI's API, with whatever data the creator has typed into the editor, and is processed on US-controlled GPUs under US jurisdiction.
This is a meaningful improvement over level one. The bulk of customer data lives in Europe under European jurisdiction. The CLOUD Act exposure on the primary database is gone. The cookies are honest. But the AI surface is increasingly the most sensitive surface in a modern SaaS — it is where the creator writes their voice, where they paste their drafts, where they ask questions they would not type into a search bar — and that surface, in level two, still routes through American infrastructure.
The phrase made in France at level two requires an asterisk that founders rarely place.
Level three: where Linkette sits
The Linkette configuration, as it stands today, is the third level. I will name every piece, with the legal details, because that is the only way the claim can be evaluated.
The legal entity is Parallactic AI SAS, French simplified joint-stock company, SIRET 94266255200010, registered at 60 rue François 1er, 75008 Paris. The controller is French. The bank account is at a French bank. The taxes are paid in France.
The infrastructure: Supabase eu-west-3 (Paris) for the primary database and authentication. Scaleway (France, Paris and Amsterdam regions) for hosting. BunnyCDN (Slovenia, EU-owned company) for the public-page CDN. Brevo (France) for transactional email. Plausible (Estonia, EU-hosted) for internal product analytics. Mollie (Netherlands) for payments. Every sub-processor lives in the European Union or the EEA. Every contract is governed by European data protection law. None of the day-to-day data path crosses the Atlantic.
The AI: Mistral La Plateforme (Paris) for every AI feature. Every generation in the editor, every onboarding turn, every link enrichment, every weekly brief. There is no path in the product to call an OpenAI or Anthropic model, even if I wanted to add one tomorrow without a release note. The lib/ai module that owns every model call routes exclusively to Mistral, by construction.
Why each step matters
Legal jurisdiction matters because the rights you have as a user of a SaaS are the rights conferred by the jurisdiction where the data lives. A French controller serving European users from European infrastructure subjects every interaction to GDPR, to the AI Act, to French civil law. There is no ambiguity about which court adjudicates a dispute. There is no CLOUD Act risk on the database. There is no need to negotiate Standard Contractual Clauses with every sub-processor — they are all already inside the EEA.
Currency stability matters in the small way that recurring SaaS bills denominated in dollars become unpredictable line items when the euro moves three percent in a month. All of Linkette's infrastructure is billed in euros. The budget is the budget.
The French Tech network matters less as a financial lever and more as a community one. Sub-processors that are themselves French or French-adjacent — Scaleway, Mistral, Brevo, Mollie — are companies whose support teams answer the phone in the same time zone, whose roadmaps I can read, whose technical leads are reachable at a Paris meetup. This is not a small thing for a solo founder.
AI Act compliance matters because the regulation came into force in 2025 and is increasingly being enforced. A foundation model provider that commits, by contract, not to train on customer data — which is Mistral La Plateforme's default position — is a meaningfully different counterparty from one whose default is to use customer data for training and offers opt-out as an enterprise upgrade. I will say more about this in a separate piece.
The Mistral choice, specifically
Choosing Mistral is not only a sovereignty decision. It is, in a small way, a vote.
There is a finite amount of European AI inference capacity in 2026. The European foundation-model layer is being built on a budget that is, by US standards, modest — a few billion euros against US AI capex in the tens of billions. Every API call routed to Mistral La Plateforme is a small amount of revenue that funds the next training run, the next GPU cluster, the next round of European AI talent kept in Europe rather than retained by a US lab at four times the salary.
A French SaaS that defaults to OpenAI for its AI surface is, by revealed preference, betting against the European AI layer. I do not think the founders making that bet are wrong on the engineering merits — OpenAI is a remarkable product. I am saying that the bet has consequences for the continental capacity over the next decade, and I would rather be on the side of the bet that wants Mistral to win.
The competitive frame
The link-in-bio category is, like most consumer SaaS categories, dominated by US incumbents. Linktree is Melbourne and San Francisco. Bento is Toronto. Beacons is New York. Bio.link is American. The two largest European link-in-bio platforms today, LinkDash and Wonderlink, do not publish a sub-processor list at the depth Linkette does, and the few details they do disclose suggest they sit at level one or level two.
This is not a criticism of those companies. It is an observation that the level three claim — French founder, French controller, EU-only sub-processors, French AI — is, as of May 2026, an empty niche. Linkette is, as far as I can determine after honest research, the first link-in-bio platform that can make it.
The gaps I will not pretend don't exist
I want to be honest about what is not yet at level three.
The hosting is on Scaleway, which is French, but the orchestration uses some open-source components that have US-origin maintenance. The eventual move is to a more deeply self-hosted setup on Scaleway bare metal, on the v2.5 roadmap.
The source code repository is GitHub, which is Microsoft-owned and US-controlled. I considered migrating to a self-hosted GitLab on Scaleway for the v1 launch. I decided not to, on the grounds that the source code is not customer data, and the engineering cost was significant for a marginal sovereignty gain. The repository is private; the CI runs on Scaleway. If GitHub's terms become untenable, the migration is a weekend.
The fonts are served from a CDN. The CDN is BunnyCDN, which is EU. The font foundry is independent and French. Good.
The error monitoring is Sentry, US-hosted in the EU region. This is a residual US dependency I want to retire by year-end. I have begun evaluating European alternatives.
I list these because level three is a moving target, not a finish line. The honest thing is to publish the gaps.
If you would like to read the full sub-processor disclosure with the legal details, it lives at linkette.eu/privacy. If you are building something else at level two and want to move toward level three, I would happily compare notes. The European AI and infrastructure layer is small enough that the people doing this work all know each other; you are welcome to write.