Data residency

Compute runs in Sydney and the default language model is pinned to Australian regions. Speech synthesis follows the voice you pick: one of the four providers has an Australian endpoint and three do not. Everything else is either Australian by default or controlled by a runtime setting.

Australian by default

Speech recognition runs against our provider's Australian endpoint on every environment that serves calls. Speech synthesis is the one component whose region follows a choice you make. See Speech synthesis.

Language models

The default model runs on Amazon Bedrock using the Australian cross-region inference profile (Sydney and Melbourne). Every Bedrock model identifier the platform offers carries the Australian prefix.

Other selectable models do not share that property. The production catalogue:

"Global" means the provider routes the request itself and may serve it from outside Australia. The model picker labels each model's residency at selection.

If residency is a requirement

Four of the thirty-two available models are Australian: the default, the two Claude tiers above it, and Gemini 2.5 Flash on Google Vertex. A model's region is not the whole picture for a voice call -- synthesis and telemetry are covered separately below.

Contact support for written confirmation of the residency position for your workspace.

Speech synthesis

Four providers can synthesise the agent's voice, and the region follows the voice you pick.

On a Deepgram voice the agent's reply text never leaves Australia. The platform routes that synthesis to the Sydney endpoint on every environment that serves calls, the same endpoint the speech recogniser uses.

The other three receive the reply text in the United States and send audio back. They store none of it.

If you need synthesis onshore

Pick a Deepgram voice. Thirty-six of them, one Australian-accented: Sharon. See Voice to change an agent's voice.

This matters most when the agent reads personal details aloud, because those details are the reply text.

Paths that leave Australia

Both are platform-level settings, not per-agent ones.

Telemetry

Error tracking, product analytics, model tracing and log aggregation each use a provider outside Australia (three EU, one US). These receive metadata only by default.

Sent by default:      identifiers, counts, durations, model names,
                      error types, stack frames

Not sent by default:  transcripts, prompts, tool arguments,
                      personal information

Telemetry content levels

Set per workspace with an optional per-agent override. The level applies per call, so one workspace's setting cannot affect another's exports.

Capturing caller IP addresses and session replay are both off by default.

Telephony during a regional outage

The platform can reroute call media through an alternative carrier site if Sydney becomes unavailable. Every alternative site is outside Australia. This is currently off and is a platform setting, not a per-workspace one. Contact support if you need it disabled contractually.

What is stored, and where

See Privacy for retention and deletion details.

Verifying this for your workspace

Your telemetry content level is shown in workspace settings. For anything else, contact support -- we can provide written confirmation of data processing locations for a security review.

Related

• Privacy
• Compliance
• Security