How automatic failover works

When you're running production traffic through a model gateway, the last thing you want is a single provider outage taking down your app. Ada AI handles that at the proxy layer.

Matching the request to upstreams

When a request arrives, Ada AI looks at the requested model and finds every enabled upstream that exposes that model. If only one upstream matches, it gets the request. If several match, they're ordered by priority number, highest first.

Retrying on failure

If the first upstream returns a 5xx, Ada AI immediately tries the next matching upstream. The response shape stays the same, so your client doesn't need any retry logic. We only retry on server errors; 4xx responses are passed through as-is so you can fix bad requests locally.

Transparent in the dashboard

Every attempt is logged with provider, status, tokens, and latency. That means you can see exactly when a failover happened and which upstream ultimately served the request.

Read more in the models and routing docs.