One EU sub-processor
for every model

It can be, and what matters is where the data is processed, what the provider retains, and whether your DPA covers the flow. Opper gives you EU-hosted routes for major models, acts as your single sub-processor, and provides a standard DPA with SCCs, so you can use frontier models while keeping processing in the EU and your paperwork unchanged.

Zero data retention means the model provider does not store your prompts or completions after the request is served, and runs no abuse monitoring that could hold them. A large share of the EU routes carry it on all traffic by default, which is rarer across the category than the marketing suggests, and you can filter for it live in the model directory, either every model with zero data retention or only EU-hosted routes with zero data retention. Routes where content is kept only for abuse monitoring are marked Abuse monitoring, with their window; routes that keep content for other purposes are marked Retained. On the leading closed models, zero data retention without abuse monitoring is available under a signed Enterprise agreement.

Not when you route through Opper. You sign one DPA with Opper, and Opper maintains the sub-processor relationships with the model providers, so adding a new model does not require amending your DPA, notifying your customers, or waiting out a 30-day objection window.

The Opper platform runs in AWS Stockholm, Sweden, and inference requests are processed in the region of the route you select. EU routes keep data in the EU end to end, and data only leaves the EU if you explicitly choose a route hosted elsewhere. Opper can also be deployed onto other clouds and regions, European sovereign providers or elsewhere, as it already runs on evroc for high-privacy workloads.

Not by default. Opper stores metadata like model, token counts, latency, and cost for your analytics, while prompt and completion content is not retained. On top of that, many EU routes carry provider-side zero data retention by default, so prompts are not kept at either hop, and the model directory filters on it. You can enable full tracing with the Control Plane when you want observability, and the zero data retention rule in Controls sets trace retention to zero and restricts routing to providers that keep nothing, on every plan.

No, Opper never trains on your data. Provider-side training posture is documented per provider in the sub-processor list, which records that no listed model provider uses customer requests or responses to train its models by default.

Every call is logged with model, cost, latency, and policy decisions, and the Control Plane adds session replay, retention controls, budget caps, and model allowlists. That gives compliance teams the audit trail and documentation these reviews ask for.

The standard Data Processing Agreement, the sub-processor list, and the platform security overview are published on opper.ai, Standard Contractual Clauses are available on request, and Enterprise customers can request tailored terms including zero retention.