Browse documentation

Hosting & data regions

See where your organization runs and stores its data, and where model calls are processed.

Investment operations data has to stay where your obligations say it does. SPARK shows, for your organization, where it runs and where its data is kept, and NGT routes the model calls it runs for you to EU endpoints. This page explains where to read that, and what it covers.

The organization's General page

Open Settings, then General under Organisation. The page describes the organization your workspaces belong to, in one Details section:

RowWhat it shows
Organization IDThe ID that identifies your organization. It can't be changed; quote it when you contact support.
HostingWhere the organization runs: Google Cloud for SPARK's cloud, or the name of a dedicated deployment.
Data regionWhere the organization's data is stored: EU for SPARK's cloud, or the dedicated deployment's own environment.
Help improve SparkWhether your users' feedback on agent replies, and the thread it relates to, is shared with NGT.

Hosting and Data region are facts about your deployment, set by NGT, not choices you make here. Everyone who can open Settings can read them.

GeneralDetails of the organization your workspaces belong to.DetailsOrganization IDIdentifies this organization. It can't be changed.org_Kc4TqN8mWz2RbV1xHostingWhere this organization runs.Google CloudData regionWhere this organization's data is stored.EUHelp improve SparkShare your users' feedback on agent replies, and the thread it relates to, with NGT.
The organisation's General page: where it runs and where its data is stored.

Help improve Spark

When people rate an agent's reply, the rating, its reason and the thread it is about can be reviewed by the NGT team to improve SPARK. Help improve Spark is on until an administrator turns it off; for everyone else the switch is shown but disabled. The change saves as soon as you flip it.

Model calls

The agents and features of SPARK call language models. Where those calls go depends on whose account they run on:

  • Calls NGT runs for you go to EU endpoints. This covers NGT Auto, the model NGT provides for agents, which runs on Google's Gemini models at the EU endpoint of Google Cloud's enterprise AI platform (Gemini Enterprise Agent Platform, formerly Vertex AI), along with thread titles and the document reading and entity extraction behind file extraction. Under NGT's Google Cloud agreement, Google does not train or fine-tune its models on your data.
  • Optical character recognition (OCR) of scanned documents and images, when a file is extracted, runs on Mistral OCR at Mistral AI's EU endpoint. It is the one step of an extraction that leaves Google Cloud.
  • Web searches an agent makes, when a workspace switches Web Search on: Enterprise grounding runs in the EU; Google Search may process the searches outside the EU, which is why switching it on asks first; Tavily runs on the workspace's own Tavily account.
  • Your own provider key: a workspace can bring its own key for a model provider and pick its models for an agent. Those calls go straight to that provider, on your account and under your agreement with them, so the EU commitments above do not apply to them. A Gemini model picked on your own key runs on Google's Gemini API, which makes no location commitment.

You can see this on the Models page in Agent Studio (Settings, then Models under Agent Studio). Its Enterprise guarantees section lists what NGT's Google Cloud agreement covers, Google Cloud, EU data residency and Enterprise security, no training on your data, each with Google's own pages. Providers that run there carry the Google Cloud and EU marks; hover a mark to read what it means. The OCR row sits under NGT's models, and the providers you can add with your own key are listed under Bring your own.

ModelsThe model providers and models your agents can use.Enterprise guaranteesGoogle Cloud, EU data residencyMarked providers run on Gemini Enterprise Agent Platform's EU endpoint, so your data stays in the EU.Data residencyEnterprise security, no training on your dataGoogle Cloud's security controls apply, and Google doesn't train or fine-tune its models on your data.Security controlsZero data retentionActive (1)NGT1
The Models page: NGT's guarantees, and the marks of the providers that run under them.

Calls on NGT's account count toward your organization's NGT tokens and document extraction limits; calls on your own key do not use NGT tokens. See Limits & Quotas.

Where to go next