Current status
All AI models come from Mistral AI in Paris, connected directly with no intermediary: EU Recap transcribes with Voxtral and summarises with Mistral, and the AI features in EU Notes use models from Mistral. We don't operate our own GPU capacity.
Why we say this openly
It would be easy to apply "data stays in Germany" only to the database, and quietly send transcription to a US provider. For a meeting's minutes, that would mean the audio track of the conversation leaves the EU, even if the recording is then stored in Germany. That's why transcription also runs in the EU, and every provider is on the subprocessor list with their country.
Where we want to get to
- AI calls routed through our own gateway, which only allows providers outside the EU with the organisation's permission
- We're evaluating self-hosted models; nothing has been decided
- No training of our own models on customer data
The honest part
For AI and transcription, we only use Mistral AI in France. If a provider changes, it appears on the subprocessor list first.
