Data that legallycannot leave
When residency is not a preference but a requirement.
The clean answer
Legal, health, finance: the cleanest possible answer to a data-residency question is that inference happens on premises, on owned hardware, with no third-party processor in the chain at all.
Auditable by inspection
No DPA negotiation, no sub-processor list, no cross-border transfer analysis. The audit is: here is the machine, here is the network, nothing leaves.
The stack we set up for this
Configured for data that cannot leave:
- Document pipelines — OCR, extraction, filing — fully on premises
- Dictation and drafting over privileged or patient records, offline
- Air-gap-capable: the stack runs with the network cable unplugged
What the audit becomes
No DPA negotiation, no sub-processor list, no cross-border transfer analysis. The audit is: here is the machine, here is the network, nothing leaves. When chat-scanning rules order providers to hand over conversations, there is no provider in this chain to order.