EU inference
EU inference routes all AI requests during building and inside your published apps to European Union servers. Any request that cannot run inside the EU will fail immediately instead of falling back to servers abroad.
Enterprise workspace owners and admins who are legally or contractually required to keep AI processing within the European Union.
Do this, in order
- 1
Open Workspace settings, go to Security, then Privacy & security, and turn on the EU inference switch under Data protection.
This starts the automated check to see if any apps in your workspace currently use AI models that are unavailable in the EU.
- 2
Select Review and update apps if Lovable flags apps using non-EU models.
Apps that continue calling non-EU models will crash with errors as soon as EU inference is turned on.
- 3
Turn on the option to publish updated apps again to the same audience before starting the update.
Live versions of published apps keep using their old non-EU models until the new code is published.
- 4
Start the update and wait approximately 20 minutes for EU routing to take effect.
Switching routing across models takes up to 20 minutes to apply to all incoming requests.
- 5
Open any app marked Needs attention and manually paste the model-switching prompt in its project chat.
Some apps store model names in configuration settings or data where automated bulk updates cannot safely edit them.
Paste this into your project
Switch the AI model requests in this app to a model that is available with EU inference. Keep everything else as it is and tell me which files you changed.
Words decoded
- Inference
- The process of sending your prompt to an AI model and receiving its generated response.
- Model endpoint
- The specific server and data center location where an AI model receives and answers requests.
- Data Processing Agreement (DPA)
- A formal legal contract specifying how user data must be handled, protected, and located.
Where people get stuck
- AI requests that have no available EU server will fail with an error rather than routing to other regions.
- Enabling EU inference increases workspace credit consumption for both building and live app usage.
- Audio, speech, video generation, and voice modes stop working because they do not have EU endpoints.
- Web search, web fetching, and browser automation tools still process their data in the United States.
- Workspace API keys used outside of Lovable cannot be updated automatically and will break if they call non-EU models.
- You can only change the EU inference switch once per hour.
The short version, steps, decoder and prompt on this page are written automatically from Lovable's own documentation and can lag or misread it. The official page is always the authority.