Integration / Model provider
Mistral AI, routed by jurisdiction.
Connect Mistral's La Plateforme API and self-deployed Mistral open-weight models to YTT Secunet, giving European tenants an EU-headquartered model provider with policy enforcement and signed receipts.
Mistral AI is a Paris-headquartered model developer. Its hosted API, La Plateforme at api.mistral.ai, stores customer data in the European Union by default according to Mistral's help centre; a separate US endpoint exists and is used only if you call it explicitly. Mistral publishes a subprocessor list and notes that some features can involve transfers outside the EU, so the default is EU hosting rather than an absolute EU-only guarantee for every feature. Several Mistral models are also released with open weights, which means you can run them yourself in a location you control.
For procurement teams the attraction is the parent company's jurisdiction: a provider headquartered in the EU removes the most direct route to US compelled disclosure that applies to US-parented providers. It does not remove every risk, because the infrastructure and subprocessors underneath still matter, and YTT Secunet's policy packs evaluate the endpoint, not the brand. A Mistral route is treated as EU when it resolves to the EU API host or to your own EU deployment, and the receipt says which.
In practice teams use YTT Secunet to blend three Mistral paths behind one base URL: La Plateforme for general workloads, a self-deployed open-weight model on vLLM in their own VPC for special category data under Article 9, and Mistral models on a European sovereign cloud as a failover target. The application calls one model alias; the policy pack decides which path a given tenant and sensitivity class may use, and PII guardrails mask identifiers before any hosted call.
| Region or deployment | Location | Jurisdiction |
|---|---|---|
| La Plateforme (api.mistral.ai), default EU hosting | European Union | EU / European Union |
| Self-deployed open-weight model | Customer VPC, Frankfurt | EU / European Union |
| Self-deployed open-weight model | Customer data centre, Zurich | CH / Switzerland |
| La Plateforme US endpoint, explicit opt-in | United States | US / United States |
- 01
Create a La Plateforme workspace and API key, and review Mistral's data processing agreement and current subprocessor list against your RoPA.
- 02
Ask Mistral about retention options for your plan and record the agreed setting in the YTT Secunet endpoint record.
- 03
Register api.mistral.ai as an EU endpoint in YTT Secunet; register any self-deployed Mistral weights as Self-hosted runtime endpoints with their location.
- 04
Map a model alias (for example mistral-large) to the hosted and self-hosted endpoints, ordered by policy preference.
- 05
Point your OpenAI-compatible or Mistral SDK at https://eu.yttsecunet.com/v1 and pass the policy and tenant headers.
endpoints:
- id: mistral-plateforme-eu
provider: mistral
upstream: https://api.mistral.ai/v1
jurisdiction: EU
credentials: vault://mistral/prod
- id: mistral-selfhosted-fra
provider: openai-compatible
upstream: https://llm.internal.acme.de/v1
jurisdiction: EU
facility: customer-vpc-fra
aliases:
mistral-large:
- endpoint: mistral-plateforme-eu
model: mistral-large-latest
- endpoint: mistral-selfhosted-fra
model: mistralai/Mistral-Small-3.2-24B-Instruct-2506
routes:
- policy: eu-gdpr-strict
sensitivity: art9
allow: [mistral-selfhosted-fra]- The US endpoint of La Plateforme is registered as a US endpoint, so an EU pack can never reach it through an alias.
- Article 9 traffic can be restricted to self-hosted endpoints while ordinary traffic uses the hosted API.
- Features Mistral flags as involving subprocessors outside the EU are disabled on EU routes unless the pack explicitly allows them.
Does using an EU-headquartered provider mean no transfer impact assessment is needed?
If processing stays in the EU and no subprocessor outside the EEA receives personal data, Chapter V transfer rules are not triggered for that call. You still need a DPIA where Article 35 applies, and you should check Mistral's subprocessor list for the features you use.
Can I run Mistral models without calling Mistral at all?
Yes, for models released with open weights. Serve them on vLLM or another OpenAI-compatible runtime in your own infrastructure and register that endpoint in YTT Secunet. Prompts then never leave your environment.
Will my application code change when I switch between hosted and self-hosted Mistral?
No. The application calls the alias; YTT Secunet resolves it per policy and signs a receipt naming the endpoint actually used.
Sovereign route ready
Stop risking 4% global turnover GDPR fines on AI inference.
Deploy sovereign, cryptographically proven AI routing in 15 minutes.