Router and OpenWeight: route the task inside your data policy
How AiAx Router matches model effort to the work, while OpenWeight keeps local, cloud and EU hosting choices explicit, with zero retention only where required.
Choose the model and reasoning effort around the task, instead of asking people to pick a model every time. AiAx Router frames routine work, difficult analysis and restricted tasks. AiAx OpenWeight covers the deployment choices behind those routes: local hosting, cloud, EU-hosted options, and zero retention only where required.
Route the work, then respect the deployment
A routine request may use a lighter route. Harder analysis can justify a stronger model or higher reasoning effort within configured provider and budget limits. Restricted work must stay on deployments approved for that information.
EU hosting and zero-retention requirements depend on the selected service and configuration. Workspace deployment policy must constrain eligible routes before a model is chosen. Seeing a Router configuration in one environment does not by itself prove the same policy is enforced in every workspace.
OpenWeight separates model choice from hosting choice. Start with straightforward cloud access, or choose more control over location and retention when the workload requires it. Zero retention is configured only where it is a customer requirement and the selected setup supports it. EU hosting alone is not a GDPR compliance guarantee.
For provider data controls, OpenAI documents that Zero Data Retention and Modified Abuse Monitoring are approval-based controls with endpoint-specific eligibility, and that EU data residency uses eu.api.openai.com with regional storage and processing subject to those requirements. Check the exact endpoint and agreement before treating a route as zero-retention or EU-resident.
Primary source: OpenAI data controls.
A concrete illustration
A workspace policy marks a service-report classification task as eligible for an open-weight route in an approved environment. An agent receives:
Classify this incoming service report and extract the work to be done. Use only the deployment approved for this information.
Router selects within the permitted endpoints. If the case is ordinary, a lighter approved model may be enough. If the text is ambiguous, the route can escalate to a stronger eligible model or a person. If the customer requires zero retention for that inference path, only an endpoint and configuration that actually support it are allowed. Workspace history and AiAxBrain memory keep their own retention rules.
Illustrative workflow using sample information. Not a certification claim or a guarantee that every model is deployable everywhere.
Decision checklist
Before you approve a restricted route for production traffic, confirm:
- Data policy is written first (what the model may receive, where processing may happen)
- Eligible endpoints are listed for that sensitivity class
- Zero retention is claimed only for endpoints/configs that actually support it
- EU hosting (if required) is an explicit endpoint choice, not assumed from branding
- Workspace history, files and AiAxBrain retention rules are separate from inference ZDR
- Escalation to a stronger eligible model or a person is defined for ambiguous cases
