Baseten: data residency & retention

BaseTen Labs, Inc., 201 Spear St, Suite 1600, San Francisco, CA (per privacy policy). Runs a multi-cloud inference platform but does not publicly enumerate the regions serving its Model APIs; docs say compliance policies define 'the geographic regions where your workloads can run' and are read-only, managed by Baseten. Privacy policy notes data may be transferred internationally.

Geographic workload policies exist but are managed by Baseten (not self-serve); organizations needing infrastructure control can use self-hosted deployments in their own VPC.

Docs: 'Baseten does not store model inputs, outputs, or weights by default' — zero-data-retention posture for synchronous inference. Async inference inputs are temporarily stored until processing completes (docs cite up to 24 hours in a secure queue); outputs are not stored.

Not explicitly addressed in the privacy policy or security docs reviewed; Baseten positions itself as inference infrastructure and stores no inputs/outputs by default, but publishes no express no-training clause.

ZDR is the default for synchronous inference ('does not store model inputs, outputs, or weights by default'), not a special tier.