Regions and data residency
Where models process your requests, what the gateway does and does not do about regions, the countries where Safeguard's own models are offered, and where Safeguard stores your API Platform records.
Where inference runs
A third-party model runs on its provider's infrastructure, so a request to it leaves Safeguard for that provider. Where the provider processes the request is set by the provider's terms. Each model's page on safeguard.sh/models shows it under Data policy › Region.
What the catalog records for each provider:
| Provider | Region |
|---|---|
| OpenAI | Global by default. OpenAI offers data residency in the US, the EU and other regions, with its approval and at a higher price for newer models. |
| Anthropic | Global by default. US-only inference is available at 1.1 times the price, for Claude 4.6 and later. |
| Google (Gemini API) | No region guarantee. Data may be stored briefly or cached in any country where Google operates. |
| Mistral AI | Global by default, with optional EU or US regional endpoints at 1.1 times the price. |
| xAI | No region guarantee on the default endpoint. A US regional endpoint is available for Grok 4.7 and Grok 4.6 at 1.1 times the price. |
| Cohere | Not stated. |
| DeepSeek | China (PRC). |
| Qwen (Alibaba Cloud) | Singapore, the international deployment whose prices the catalog lists. Other regions have separate price lists. |
| Meta | Depends on the host for its open-weight models. Not stated for the others. |
| Open-weight models served by a host | Set by the host. |
| Safeguard's models | The region your Safeguard workspace runs in. |
The catalog lists more models than the gateway can serve. GET /gpt/v1/models returns the ones you can call.
The gateway does not choose a region
This version has no region setting. You can't choose or restrict the region for a request, a key or a workspace, and the providers' regional options in the table above can't be selected through the gateway.
The catalog can also give a model a region value. GET /gpt/v1/models returns it as region, and your request log records it with each request. It describes the model. It does not route requests.
Where Safeguard stores your records
| Record | Stored in |
|---|---|
| Request log, credit ledger and gateway settings | Safeguard's billing service |
| Prompt and output text, if Store prompts and outputs is on | Safeguard's billing service |
| API key hashes, owners and limits | Safeguard's authentication service |
Both services run in the Safeguard deployment that hosts your workspace. Safeguard does not send these records to model providers. For where Safeguard's hosted service runs, see the FAQ on Safeguard's Trust Center.
Countries where Safeguard's own models are offered
Griffin, Eagle, Lion and Zero are available in 245 countries and territories (the ISO 3166-1 list minus the exclusions). They are not offered in Cuba, Iran, North Korea or Syria, nor in the Crimea region of Ukraine or the so-called Donetsk People's Republic (DNR) and Luhansk People's Republic (LNR) regions of Ukraine. None of them can be called through the API yet. See Safeguard models.
Keeping inference in one place
The Safeguard platform can run Safeguard's models and open-weight models inside an environment you control. Through the API Platform, that is not available yet: catalog models are served by their providers or hosts, and Safeguard's own models cannot be called yet. See Deployment options.
Related
Zero data retention
How the gateway sends a request only to models whose provider offers zero data retention, and which models the catalog marks that way.
Security and compliance
How the API Platform protects requests and keys: TLS in transit, hashed single-purpose API keys, checks that fail closed, an audit trail, and where to find Safeguard's compliance reports.