EU sovereignty & residency

niil is designed so that AI processing happens in the EU (or on infrastructure you host yourself), and so that your organization stays in control of that choice.

EU-only models, per organization, on by default

Each organization has an EU-only models policy that is enabled by default. While it's on:

• Only models whose processing region is EU (or self-hosted) can be selected or routed to.
• This applies everywhere models are used — Secure Chat, the workflow assistant, agents, and workflows.
• Non-EU models are hidden from the picker and rejected if requested.

Admins control the policy under Organization → Security. Because it's per organization, each tenant sets its own posture — and the secure default means EU processing unless someone deliberately opts out.

Region-pinned inference

Models are reached through EU-pinned endpoints (for example Azure OpenAI in Sweden, Google Cloud in the EU, Mistral in France, or a self-hosted model server). Operators configure these region-locked endpoints so requests don't leave the EU. niil also supports native Vertex AI and AWS Bedrock connections for running Claude and Gemini within EU regions.

Processor transparency

Every model in the picker shows its processor and region (e.g. "Azure OpenAI (Sweden)", "GCP (EU)"), and an EU marker. You always know where a given model runs before you use it — see Choosing a model.

What "sovereign" means here

• Data location — processing stays in the EU (or your own infrastructure).
• Control — the residency policy is yours to set, per organization.
• Self-hostable — models can be self-hosted, removing external processing entirely for the most sensitive workloads.