Routing & Failover

Each model may be served by multiple upstream channels (pools). Requests are routed as follows:

• Cost-first: automatically selects the cheapest available channel
• Automatic failover: if an upstream times out or degrades, traffic switches in seconds without interrupting you
• High availability: multi-region bare-metal clusters; average added latency under 100ms

You can see the active channel for each model on its detail page.