Choose your workspace AI region
Open Settings → Workspace → AI inference region, choose Global, EU or US, then save. Workspace owners and admins can change this setting.
The workspace data overview under Settings → Legal summarizes the selected AI region, hosting locations and recorded models from recent calls. Model regions show availability in the current catalog. Use Print / PDF to save a copy for your workspace records.

Settings → Workspace → AI inference region: choose Global, EU or US.
SIP trunks in the EU and US
Famulor SIP trunks are available in the EU and US. For assistant calls using phone numbers purchased through Famulor, saving the workspace region automatically switches inbound and outbound routing to the corresponding region: EU → EU, US → US. This automatic switch applies to Famulor-provided phone numbers used by assistants. Your own SIP trunks, verified caller IDs and Loop phone routes keep their separate routing settings.Infrastructure locations
Our realtime connection network includes the following locations:
Regions with multiple locations, such as the EU and US, offer automatic in-region redundancy for realtime traffic. If one location becomes unavailable, traffic reroutes to another available location within the same region.
Changing the workspace AI region does not move the database or application hosting out of Frankfurt. Choosing a workspace AI region does not change the storage region for session diagnostics or set the region for every part of a call. For named infrastructure providers, storage locations and data-processing details, see the Trust Center.
Supported models
Compare supported models by use case and regional endpoint. The assistant editor shows the options available for your workspace’s region, plan and engine mode. Yes marks an available regional endpoint. No means the model is unavailable when that region is selected. Global route and EU route identify options offered in US workspaces through those routes.Gemini 3.5 Transcribe Live and Muse Voice Transcribe 1.0 require the Global workspace setting.
Decision models
Decision models handle tasks such as checking consent or classifying messages. The regions below show their availability in the model catalog.
Open Settings → Legal → Workspace data overview to see the current default for call consent, based on your workspace’s AI region. Workspaces with EU or US use the built-in consent check. The overview also includes this information when you select Print / PDF.
Language models (LLM)
Speech recognition (STT)
For AI phone assistants, Speechmatics Linden-1 is the recommended Speechmatics model. Speechmatics Oak-1 is optimized for medical transcription, including drug names and clinical terminology. Speechmatics model guide
GPT-Live Transcribe and Inworld STT 1 are available for Pipeline in Global workspaces only. Inworld STT 1 recognizes English only. GPT-Live Transcribe is separate from the Full Duplex conversation engine.
Speech generation (TTS)
Soniox has EU and US endpoints for speech recognition and speech synthesis. In Global workspaces, recognition follows the voice-agent location: Frankfurt uses the EU endpoint, and Ashburn and Mumbai use the US endpoint. Synthesis defaults to the US endpoint in Global workspaces.
Gemini 3.8 Flash-Lite TTS and Gemini 3.8 Flash TTS are available in Global and US workspaces through unrestricted Global routing. Global routing in the US column means workspace availability, not a dedicated US endpoint or a guarantee of processing within the US. These two models are unavailable in EU workspaces.
Gradium TTS Beta is available in Global and EU workspaces for Pipeline and Half-Cascade, in English, French, German, Spanish and Portuguese. It is an additional model choice; the existing Gradium TTS default remains available.
Realtime speech models
Grok Voice 3.0 uses Realtime → Standard with 28 public voices. It is available in Global and US workspaces through a global route, with no dedicated EU endpoint. The US workspace setting does not establish a dedicated US endpoint for this model.
Rime Prism is available in Realtime → Standard in Global workspaces. Its four English voices are Astra, Aurora, Brando and Wawona.