Documentation▾

Data residency — Porten and OpenAI

Where inference runs, who operates the machine and what gets stored are separate questions. An EU region alone does not answer all three. Start by defining which data and processing must stay within a particular boundary, then choose the service and configuration.

Compare the mechanisms

Question Porten OpenAI API
Regional inference The API key's allowed regions restrict which nodes can be selected. Regional storage and processing are separate parts of data residency support; check the endpoint, model and region.
Your own inference machines The key's My nodes only setting restricts routing to the account's nodes. Review the current service agreement and the provider's documented operating model.
Control plane In the hosted service, requests pass through Porten's Hub even when your own node runs the model. Check the project's settings and which data is covered by the regional terms.
Storage and metadata Usage records, including model, token counts and timing, are handled separately from inference. Data residency does not cover all system data, such as some account, usage and billing information.

OpenAI documents European data residency and regional processing for supported combinations. It is therefore inaccurate to describe the service as having no EU options. Check eligibility, project configuration and feature restrictions in OpenAI's data controls. This comparison was checked on 28 October 2026; consult the source for updated terms.

Configure and test Porten

  1. Choose the key's allowed region in API keys.
  2. Select My nodes only if inference must run on your account's machines.
  3. Make sure the node's region is registered correctly and the selected model can run there.
  4. Send a harmless test request. Then take your allowed nodes offline and repeat the test. The request should fail when no allowed node can run the model; it should not move to other nodes to improve availability.

Region, owner and minimum node class are routing constraints, not model-quality settings. The exact error depends on whether the model is denied by policy or lacks available capacity. See regions and your own hardware.

Trust and the control plane

A node operator controls the machine that processes the prompt. Do not assume that a community node hides content from its operator. A higher node class does not replace your assessment of the operator, agreements and access. Hardware-protected inference with TEE and remote attestation is a roadmap feature, not a general guarantee for today's shared network.

My nodes only does not move the hosted Hub into your network. The Hub relays prompts and responses; usage and account data are still handled by the platform. For control over the Hub, database and operations as well, evaluate self-hosting Porten.

Before making a decision

Document your requirements for inference location, transport, retention, subprocessors, support access and backups. Verify them against the chosen deployment and applicable agreements. Technical routing is part of that assessment; it does not, by itself, establish GDPR compliance.

Once the requirements are clear, the migration guide explains how to move a Chat Completions client and test functionality, capacity and rollback.

📄 Reading as a machine? This page is available as raw Markdown at https://porten.ai/docs/data-residency.md — or grab the whole site via llms.txt / llms-full.txt.