What happens if the AI service is temporarily unavailable?

Two recovery mechanisms handle this without you: retries and reconnection.

When the model provider returns a temporary error, Simple Chat retries automatically, for customer-facing replies and supporting tasks alike. The retry happens behind the scenes; customers stay in the same thread and shouldn't need to know it happened. If retries are exhausted, the conversation stays intact and the shopper sees a recovery message inviting them to try again, rather than a thread stuck in limbo.

Live connections heal themselves

Real-time message delivery in the dashboard and the storefront chat runs over a WebSocket. Brief drops (a Wi-Fi blip, a laptop waking from sleep, a mobile network handoff) trigger automatic reconnection, so staff watching a live conversation should see updates resume without refreshing the page once connectivity returns.

During an incident

Mostly nothing to do. Expect temporary delays during provider or network trouble rather than permanent stuck states. If something looks wrong for more than a moment, refresh the dashboard or ask the customer to reload the storefront page, but in most cases the built-in retries and reconnection recover on their own.

You don't manage model routing or provider keys, so there's no failover runbook to write. After an incident, review conversations the way you would after any support spike: confirm replies stayed accurate, and escalate any thread where automation paused mid-issue.

Changelog releases

This topic appears in the following release notes: