Reliability

Failover

What happens when the AI model or the voice fails during a call, and how a workspace's residency setting decides whether a fallback may run outside Australia.

When the AI model fails, the caller's turn is answered again on the same conversation context, so the conversation is not lost. When the voice fails, the call can switch to a fallback voice you set. A workspace's data residency setting decides whether either fallback may run outside Australia (see Data residency).

SubsystemWorkspace set to AustraliaOverseas fallback allowed
AI model, defaultThe failed turn is retried in AustraliaThe call can switch to an overseas provider of the same model
VoiceA fallback voice is used only if it runs in AustraliaAny fallback voice you set

AI model

A retry or a switch happens when a reply fails or stalls, and the caller's turn is then answered again. Both run on a shared conversation context, so neither loses the conversation.

If the model stalls before its first word, a watchdog bounds the wait and treats the stall as a failure.

Workspace set to Australia

The default model has no second provider. A failed turn is retried in Australia, up to twice. If it still fails, the agent says a short line and transfers a phone caller to the agent's overflow number, or to the number in its transfer_number variable when it has no overflow number. With neither, or on a call from a browser, it ends the call cleanly.

Overseas fallback allowed

A failed turn switches to the model's overseas provider. Both paths are configured identically, including prompt, token limits, temperature and caching, so behaviour cannot drift between them. Once the primary provider recovers, traffic returns to it automatically.

Agents on a single provider

Some models have no second provider. For those agents, the first error makes the agent say a short line, then transfer a phone caller to the agent's overflow number or transfer_number, or end the call cleanly. The underlying error is recorded and never spoken.

Voice

Set a fallback voice from a different provider on the Voice page. A fallback voice from the same provider as the main voice is not used. The fallback voice connects at the start of the call and stays idle, so it is ready the moment it is needed.

The first time the voice fails, the reply is tried again on the same voice. If it fails again, the call switches to the fallback voice and stays on it, rather than returning the caller to a provider that was failing.

On this page