Lovable adds EU inference to route AI requests to European endpoints

Lovable has introduced an EU inference setting for Enterprise customers that routes supported AI model requests to designated European endpoints. According to the company, it applies both while teams build in Lovable and when people use the built-in Lovable AI service in published apps.
- Lovable's EU inference setting routes supported AI model requests to designated European endpoints, while building and in published apps using Lovable AI.
- If no permitted endpoint can serve a covered request, it fails instead of leaving the configured boundary.
- It is not a data residency commitment: hosting and storage are controlled separately.
- Claude cannot be used through Lovable AI in published apps under this setting, and web tools still process data in the US.
- Available on Enterprise, enabled by a workspace owner or admin, with changes taking about 20 minutes.
What EU inference does
Inference is the step in which an AI model processes a request and produces a response. Lovable argues that an application can keep its records in Europe while prompts and project context are sent elsewhere for model processing. Knowing where a database lives answers one question in an enterprise security review, the company says. Knowing where the AI runs answers another.
With EU inference enabled, Lovable routes supported model requests to designated European endpoints. For requests covered by the setting, if no permitted endpoint can serve the request, it fails instead of being sent outside the configured boundary.
The feature is available on Enterprise and can be enabled by a workspace owner or admin. Changes may take about 20 minutes to apply to new requests.
How it differs from data residency
Lovable stresses that EU inference is not a data residency commitment. Data residency usually concerns where specified data is stored at rest, and the exact commitment depends on the provider's terms. EU inference addresses model processing only. It does not move all customer data into EU storage or establish a platform-wide data residency commitment. App hosting and storage are controlled separately.
In practice, a requirement about where application records are stored needs a storage answer, while a requirement about where supported model requests are processed needs an inference answer. Some organizations may need both, according to Lovable.
Limits teams should check first
Lovable lists several caveats to review before switching the setting on:
- Claude: It can be used while building in Lovable with EU inference enabled. Published apps using the built-in Lovable AI service cannot use Claude models under this setting. Those models are removed from the app's available model list, and requests to them are rejected.
- Direct provider connections: If an app calls OpenAI, Anthropic or another provider directly with its own code and credentials, those requests bypass Lovable AI. Processing location must be managed separately with that provider.
- European countries outside the EU: Models hosted by Google and AWS run in EU countries. Microsoft-hosted models can also run in countries such as Norway and Switzerland. Organizations that require processing strictly within EU member countries should review this with their account team.
- Web tools: When Lovable searches the web, fetches a webpage or uses browser automation, the data sent to those tools is processed in the US.
- Unavailable features: Video generation, voice conversations and certain image-editing tools cannot be used while EU inference is enabled.
- Credits: AI usage consumes more credits, both while building and when people use Lovable AI features in published apps.
Lovable also says EU inference does not guarantee that network traffic stays within the EU as it travels between the user, Lovable and its services.
What it means for existing apps and evaluations
Existing apps need a check before the switch. If an app uses Lovable AI with a model that is not available under EU inference, including Claude, those AI requests will fail once the setting takes effect. Lovable advises updating the app to an available model and publishing it again, and points to its EU inference documentation for model availability.
For organizations where regional AI processing has been a blocker, the company suggests revisiting the discussion. It recommends starting from the workflow to be supported: what will be built, which AI features it needs, and what internal policies require for storage and processing. Lovable's account team can help compare the documented coverage with those requirements. The company says the setting gives builders, security teams and procurement a clearer basis for deciding which projects can move forward and which extra controls they need.
Lovable: EU inference: control where AI processes your requests


