Lovable lets enterprise customers process AI requests in Europe

Lovable has launched EU inference, a setting for enterprise customers that routes requests to supported AI models through European endpoints. It works in two scenarios: when a team builds a product inside Lovable, and when a finished, published app calls a model from its own AI features. The announcement came out on 12 October 2026. It does not name specific models, countries or terms.

Two paths by which data reaches a model

A platform that generates apps has two distinct data flows. The first appears during development: prompts, code fragments, descriptions of business logic and uploaded files are sent to the model. The second appears after publication. If the app contains a chat assistant, a ticket classifier or document search, user input also passes through the model, and it is often customer data.

The second path is the more painful one for compliance. A developer can control what goes into a prompt at build time. They cannot control what thousands of end users type. Lovable says the feature covers both paths, and that is what sets this release apart from a simple region switch for the builder.

Why the question of where processing happens became mandatory

Lovable is based in Stockholm, and European customers are a natural market for the company. Their procurement questionnaires have long included a separate line: where is data processed? Storage and processing are not the same thing. A database may sit in Frankfurt while the model request goes to an endpoint across the ocean. Under GDPR, such a transfer requires a separate legal basis, and lawyers and security teams have to explain it.

Major cloud model providers already offer regional endpoints, so the technical foundation for this kind of routing exists. Lovable makes it available inside its own product, so a team does not have to build its own layer of keys and proxies.

What the announcement tells us:

• Audience: enterprise teams.
• Mechanism: routing requests to European endpoints.
• Coverage: development in Lovable and AI features of published apps.
• Limitation: supported models only.
• Not disclosed: the list of models, hosting countries, pricing, latency, SLA.

What the announcement leaves out

The word "supported" is the most important one in the announcement. Providers' European endpoints often receive new model versions later than their global ones. A team that enables EU inference may find that some models are unavailable or lag behind in version.

There is also the question of how the system behaves on failure. If the European endpoint is down, should the request fail or go to another region? For compliance, only the first option is correct; for uptime, the second is more convenient. The announcement does not say how this works.

Logging and retention deserve separate attention. Regional processing alone does not answer who stores the prompts, for how long, and which subprocessors are involved.

What to check before enabling it

A team planning to move its AI features to European processing should ask Lovable for several things:

• The current list of models and regions where EU inference is available.
• A description of the behavior when the European endpoint is unavailable.
• The terms for storing and logging prompts and responses, plus the list of subprocessors.
• The current data processing agreement (DPA), updated for the new option.

After enabling it, verify the result in practice. Run test requests from the published app and confirm, using the documentation and logs, that traffic actually goes to the intended region. If your record of processing of personal data relies on this feature, update it too: permission at the platform level does not replace internal documentation.