Data residency

Restrict inference and dedicated training for your Fireworks Enterprise account to a selected region such as US-only, with regional API endpoints, SDK configuration, and compliance.

Data residency restricts inference and dedicated training on your account to a single region. Once a region is set, every request must use that region’s API endpoint and a model served in that region, and every dedicated training job must run in that region.

It does not change how your requests are routed. Turning it on will not move existing traffic into the region. This setting enforces all requests to be in the set region, and rejects the ones that are not.

Enterprise feature. Data residency is available on Enterprise accounts. Contact your Fireworks representative if you need this enabled.

Only account Admins can change data residency. Other roles can view the current setting but cannot change it.

Residency applies account-wide: every API key on the account is restricted to the selected region.

​

Available regional restrictions

Regional restrictions	API endpoint	Serverless	Dedicated deployments
None	api.fireworks.ai	Any model	Any region
US	us.api.fireworks.ai	US-only Serverless	US multi-region, or a single US region such as US_IOWA_1

None is the default. For a region that is not listed, contact sales.

​

Before you switch it on

A residency change takes effect as soon as you save it, and anything that does not match the region is rejected.

​

Serverless

Point your clients at the region’s API endpoint and change every request to use a model from that region. See US-only Serverless for the US endpoint and model IDs.

inference_geo is deprecated in favor of data residency. As you switch over, remove the inference_geo field from your request bodies and the Fireworks-Inference-Geo header from your requests.

​

Dedicated deployments

Every dedicated deployment must run in the selected region. Create it with --region set to that region, or to a single region inside it. See Regions.

Check what you already have before you save the setting:

firectl deployment list

• New deployments outside the region are rejected when you create them.
• Existing deployments are not checked, moved, or stopped when you switch the setting on. One outside the region keeps running and keeps costing you money, but requests to it are rejected. Replace it first.
• You can move a deployment to another region inside your residency region, but not out of it.

​

Dedicated training

Supervised fine-tuning, DPO, reinforcement fine-tuning, and RLOR trainer jobs must set an explicit training region inside the selected region, for example with firectl’s --region US_IOWA_1 flag. See Regions.

• Jobs without a region are rejected, including jobs created from the console, which does not set one.
• Jobs with a region outside the selected region are rejected, both when you create them and when you resume them.
• Only where the job runs is enforced. Where its checkpoints and outputs are stored is not.

​

Configure in the console

Account admins can set data residency in the Fireworks console at Settings → Governances → Data Residency (open in console).

Select the region and save. The console asks you to confirm that your clients already use the endpoint and models shown for that region.

​

Configure with firectl

Set the region:

firectl policy residency set US

Inspect the current setting:

firectl policy residency get

Remove the restriction and return the account to unrestricted serving:

firectl policy residency clear

​

Configure with the API

Set the region with Update Policy Settings, using an API key that belongs to an account Admin and naming residency in updateMask:

curl -X PATCH \
  "https://api.fireworks.ai/v1/accounts/${ACCOUNT_ID}/policySettings?updateMask=residency" \
  -H "Authorization: Bearer ${FIREWORKS_API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{"residency": "US"}'

To remove the restriction, send the same request with an empty body (-d '{}').

Get Policy Settings returns the current setting. An absent residency means serving is unrestricted.

​

Pricing

Regional models are priced at a premium over the base serverless price for the same model. See Serverless pricing.

​

Limitations

The following are not supported while a region is set: calls are rejected, and stay rejected until you clear the setting.

• Serverless training. Serverless Training sessions, runs, and their samplers are blocked. Support is coming soon. Dedicated training is supported.
• Nexus Auto outside the US. With US residency, auto works and serves open-model turns from US-only Serverless. Anthropic and OpenAI turns run on your own provider account, where that provider’s data policies apply. With other regions, auto is rejected. See Enterprise Controls.
• BYOC. BYOC deployments run in your own cloud account, so Fireworks cannot enforce where they run.

​

Related

• US-only Serverless — endpoint and model IDs for the US region
• Enterprise features — overview of Enterprise administration capabilities
• Managing users — account roles and permissions
• Data Security — encryption, retention, and access controls