Integration / Model provider
Google Cloud Vertex AI, routed by jurisdiction.
Send Gemini and partner models on Vertex AI to regional European endpoints through YTT Secunet, and keep the global endpoint out of reach for tenants that need in-region processing.
Google Cloud Vertex AI
Model provider
- Regions documented
- 6
- Setup steps
- 5
Vertex AI's residency behaviour depends on the endpoint you call. Google documents that when you call a regional endpoint, such as europe-west3 in Frankfurt, machine learning processing for supported generative AI models happens in that region. The global endpoint offers higher availability but does not guarantee where processing happens. For Claude and some other partner models, Google lists specific European regions such as europe-west1 (Belgium) and europe-west4 (Netherlands) as well as multi-region endpoints for the EU. Google also documents in-memory caching of inputs and outputs and abuse monitoring behaviour, both of which customers can review and, where Google allows, disable for zero data retention.
YTT Secunet's role is to make the regional choice impossible to get wrong. Each tenant's policy resolves to an allowed set of Vertex AI locations, and YTT Secunet builds the request against that location's endpoint using your service account. A developer who hard-codes the global location in a notebook still gets routed through policy once the call goes through YTT Secunet. Failover stays inside the jurisdiction: Frankfurt to Netherlands or Paris for an EU tenant, never to us-central1.
Because Vertex AI is a US-parented service, a regional endpoint is a technical measure, not a legal one. YTT Secunet's receipt names the location id, the publisher and model, and whether your project has data caching disabled according to your declared configuration. Keep the Google Cloud project's Assured Workloads or organisation policy resource location constraint aligned with the same regions, so the policy holds at two layers.
| Region or deployment | Location | Jurisdiction |
|---|---|---|
| europe-west3 | Frankfurt am Main, Germany | EU / European Union |
| europe-west4 | Eemshaven, Netherlands | EU / European Union |
| europe-west1 | St. Ghislain, Belgium | EU / European Union |
| europe-west9 | Paris, France | EU / European Union |
| europe-west6 | Zurich, Switzerland | CH / Switzerland |
| europe-west2 | London, United Kingdom | UK / United Kingdom |
- 01
Check Google's data residency documentation for the model you plan to use and confirm it is offered at a regional endpoint in your target location.
- 02
Create a service account with the Vertex AI User role in a project whose organisation policy restricts resource locations to your chosen regions.
- 03
Review Vertex AI data caching and abuse monitoring settings for the project and record your configuration in the YTT Secunet endpoint record.
- 04
Register each location as an endpoint in YTT Secunet and attach it to the matching policy packs; leave the global location unregistered for regulated tenants.
- 05
Point your OpenAI-compatible or Anthropic-compatible SDK at the YTT Secunet regional base URL and send the policy and tenant headers.
providers:
google-vertex-ai:
project: acme-ai-eu-prod
credentials: workload-identity://yttsecunet-vertex@acme-ai-eu-prod.iam.gserviceaccount.com
endpoints:
- id: vertex-fra-gemini
location: europe-west3
publisher: google
model: gemini-2.5-pro
- id: vertex-ams-claude
location: europe-west4
publisher: anthropic
model: claude-sonnet-4-6
routes:
- policy: eu-gdpr-strict
primary: vertex-fra-gemini
failover: [vertex-ams-claude]
deny_locations: [global, us, us-central1]- The global location is never a valid target for a residency pack, because Google does not guarantee the processing location for it.
- Multi-region EU endpoints are accepted by eu-gdpr-strict and refused by single-country rules, mirroring how Bedrock geography profiles are handled.
- Grounding and tool features that call other Google services may process data outside the selected region; YTT Secunet strips tool configurations that the policy has not approved.
- Model availability per region changes often; the registry rechecks availability daily and alerts before a route breaks.
Is Gemini available in Frankfurt?
Google lists several Gemini models at the europe-west3 regional endpoint, but the list changes by model version. YTT Secunet's registry checks the exact model and location before activating a route, and Google's data residency page is the authoritative source.
Can I use Claude on Vertex AI for EU tenants?
Yes, in the EU regions and multi-region endpoints where Google lists the Claude model you need. YTT Secunet routes the Anthropic Messages format to Vertex AI for you.
Does YTT Secunet need owner access to my Google Cloud project?
No. It needs a service account with permission to call Vertex AI prediction in the approved locations. It does not need to create or modify resources.
Sovereign route ready
Stop risking 4% global turnover GDPR fines on AI inference.
Deploy sovereign, cryptographically proven AI routing in 15 minutes.