EU residency is the only mode.
Every Azure-backed model on Kataleptic runs in Sweden Central, and the self-hosted models run on our own EU machine — no toggle to find, no enterprise tier, same models and prices. We retired the US plate in August 2026 rather than keep a second region that nobody used. The one exception is named below, in bold, rather than buried: three native realtime voice tiers are globally routed by Azure.
How it works
There is nothing to configure, because there is nothing to choose. Every chat, embedding, image, audio and video request on your key is dispatched to the Sweden Central pool for that model. No separate endpoint, no separate key, no separate price, and no header to remember — a call you make with a stock OpenAI SDK is already EU-resident.
POST /v1/chat/completions
Authorization: Bearer dg_…
{
"model": "deepseek-v3-2",
"messages": [ … ]
}Your account page states the region it is served from; it reads European Union · Sweden Central for every account, because that is the only place we run.
When EU degrades, we don't pretend it didn't
Automatic cross-region failover is the silent default at most inference vendors. The customer never sees the dip; the uptime number stays clean; the residency commitment quietly evaporates. Kataleptic cannot do that even by mistake: there is no second region to fail over to. If Sweden Central is unhealthy for a model you requested, the request fails fast and says so:
HTTP/1.1 503 Service Unavailable
Retry-After: 30A 503 you can retry is a worse afternoon than a silent reroute, and a better compliance story. Status for each model is on the status page.
What about the price?
Same prices as the default pool. There is no EU premium, no "compliance tier," no minimum spend, and no procurement conversation. Residency is part of the platform; it is not upsold separately.
The honest reason this is feasible: Sweden Central is not a compliance add-on we stood up for you, it is our production environment. Charging for residency would mean charging you for where our servers already are.
What's covered
Every model whose backend we operate in Azure — the bulk of the catalogue — runs in Sweden Central. For passthrough models that we resell from upstream providers, residency follows the upstream provider's policy; we document this on each model's page in the catalogue.
Self-hosted open-weight models on our dedicated GPUs (Whisper V3 Turbo, embeddings) are served from our own EU machine.
The exception: native realtime voice
Three tiers of /v1/realtime —
gpt-realtime-2,
gpt-realtime-2.1 and
gpt-realtime-2.1-mini — are Azure's own
native speech-to-speech models, and Azure routes them
globally. They are not EU-pinned, and we will not
claim otherwise to keep this page tidy.
The other realtime tiers are. Our cascade and HD tiers compose STT, a chat model and TTS from EU-resident parts and speak the same WebSocket protocol, so an EU-residency requirement costs you a model choice, not an integration. Each tier's routing is stated on the model cards and in the realtime docs.
Five dollars of free credit. No card, no call. Get a key