Docs
FAQ
Questions that come up most often during evaluation and integration.
- With routing on, is it impossible to predict which model gets called?
- The decision log records candidates, scores and exclusion reasons. For paths where prediction is essential, pin the model and leave only the rest to routing.
- Do we have to rewrite existing prompts?
- Mostly they work unchanged. Instruction-following does vary by model though, so if quality swings widely, narrow the candidates with the evaluation feature.
- Does data leave the country?
- Models on a domestic inference region never leave it. Block cross-border transfer in your organization policy and only those models become candidates.
- Do you store full request and response bodies?
- Only when the organization enables audit log storage. The default is off, and permission to read full bodies is granted separately.
- What happens when credits run out?
- Requests stop with 402. You are emailed before that once the low-balance threshold is crossed, so many organizations prefer alerts over auto top-up.