Processing regions

A security classification says how sensitive data is. Processing regions say where that data is allowed to be handled. They are two different questions, and a model has to satisfy both before Intric will offer it.

This is the control that turns a data-residency requirement — “personal data in this category stays inside the EU” — from a policy people have to remember into something the platform enforces on every request.

How the pieces fit together

Three things that already exist in Intric meet here. It’s worth being precise about which is which, because they’re easy to conflate.

The sub-processor — who processes the data

Whenever a model runs, some third party actually does the processing: the model vendor for a model reached through an external API, or Intric’s own infrastructure provider for a model Intric hosts. That party is the sub-processor. See Sub-processors.

The hosting region — where that party runs

Every sub-processor in the Sub-processors view carries a hosting region: the place its infrastructure physically sits. This is the column you already use for a data-residency review, and it’s a property of the sub-processor, not something you set.

The processing regions — where this data may go

A classification carries the set of regions its data may be processed in. This one you do set, per classification, and it’s the half of the equation that expresses your organization’s own requirement rather than a vendor’s fact.

What Intric does with them

When it decides whether to offer you a model, Intric looks at where that model’s sub-processor is hosted and checks it against the regions the Space’s classification permits. If the region isn’t on the list, the model isn’t available there — regardless of how the model itself is classified.

Note

Read it as two independent tests a model has to pass. Is it classified highly enough for this Space? and does it run somewhere this classification allows? A model can pass the first and fail the second, which is exactly the case that used to be invisible.

Setting them

Processing regions are configured per security classification, under Policies in Administration. You set them on the classification rather than on a Space, so every Space carrying that classification inherits the same restriction automatically — and a Space admin can’t loosen it.

What they apply to

The regions apply to every kind of model alike:

• Completion models — the model that reasons and writes the answer.
• Embedding models — the model that converts your documents into searchable form.
• Transcription models — the model that converts speech to text.
• Image models — the model that generates and edits images.

This matters more than it first appears. A Space can be entirely compliant on its completion model and still be sending audio to a transcription model hosted somewhere you never approved — different model types are configured separately, and before this control they were reviewed separately too.

What you see when a region blocks something

Building an assistant

A model hosted outside the permitted regions simply doesn’t appear in the picker in a Space with that classification, the same way a model classified too low doesn’t appear. See Security classification for the user-facing view of this.

Saving or asking

Where the model is named explicitly rather than chosen from a list, the action is refused and the region is given as the reason — not a generic error with an ID. The message tells you the model is hosted outside the regions this classification permits, so the person hitting it knows it’s a deliberate policy rather than a fault.

Reviewing your configuration

Two views answer the practical questions:

• Sub-processors — every party currently processing your organization’s data, with its hosting region. Start here to see what regions you actually depend on today.
• Resources — every assistant and Space with its model and classification, exportable to CSV. Use it to find what a new region restriction would affect before you set one.

Important

Set the regions on a classification before you rely on them for a compliance answer, and check the Sub-processors view afterwards to confirm nothing is still reaching a region you’ve excluded. A processing-region claim is the kind of statement that ends up in a procurement questionnaire or a DPIA, so it’s worth verifying rather than assuming.

Related

• Classifications — defining the levels themselves.
• Model classification — classifying each model type.
• Security classification — how a classification reaches a Space and what users see.
• Policies — where processing regions are configured, alongside your organization’s other enforced rules.