Regions
Where decisions run, how to pin a region, and where records are kept.
Planned
Region choice is not live yet. This page describes the plan; the hosting regions, the default region and the sub-processors are not confirmed.
Hosted decisions are planned to run in three regions: US, EU and India. Vault runs wherever you deploy it.
| Region | Notes |
|---|---|
| US | Planned. The default region for new organisations is not decided yet. |
| EU | For EU data residency. |
| India | For Indian data residency. Request metadata is also kept in India (see below). |
Pinning
An organisation can be pinned to a region by contract. Requests are then served only from that region, with no fallback to another region: if the model is not available there, the request fails with 529 (503 on /v1/decisions). GET /v1/models with the header Sieve-Extras: true lists the regions in which each model is served. An empty list means every region. Pinning is set by Sieve; there is no console setting.
Auto and 31B
Sieve-31B, and so the escalation step of Auto, may run in fewer regions than the smaller models. Where 31B is not served, Auto answers from 4B and sets the response header Sieve-Escalation: unavailable.
What stays where
- Content (state, questions, answers) is processed in memory in the serving region and is not stored. See Data retention.
- Request metadata (no content) has a copy in India as Indian law requires, whatever region served the request.
- Books of account (ledger, invoices, payments) have a daily backup in India.
- Vault: nothing leaves your environment.
The list of sub-processors and the region of each is not published yet. Ask support@sthanika.ai if you need it for a review.
Next steps
Last updated