What happens when your primary model goes down? API providers have outages. Rate limits hit. You need a fallback plan. The agents.defaults.model.fallbacks field takes an ordered list of models to try.
If Claude is down, fall back to GPT. If GPT is also down, fall back to an open-source model via Ollama. The Gateway tries each model in order until one responds. You can also set fallbacks per-agent so your code review agent falls back to a different model than your personal assistant does. Without fallbacks, a single provider outage makes your entire Gateway go silent. Don't let that happen.