Skip to main content

Go multi-model

The whole point of a gateway: one integration, every model. Switching providers is a string change.

Swap the model field​

 {
-  "model": "gpt-5.1",
+  "model": "claude-opus-4-7",
   "messages": [{"role": "user", "content": "Summarize this…"}]
 }

No new SDK, no new key, no new billing account. The response comes back in the same wire format you called with.

What the gateway does for you​

  • Routing — each model maps to a weighted group of upstream providers; healthy upstreams get the traffic.
  • Failover — if an upstream degrades or opens its circuit breaker, calls shift to the fallback chain automatically. You see one successful response, not a provider incident.
  • Unified accounting — every call lands in the same usage dashboard and wallet regardless of provider, with per-project quotas if you use them.

Picking models​

The model catalog lists every available model with live pricing, modalities (chat, image, video, audio, embeddings) and context windows. The id shown in the catalog is exactly what you put in the model field.

Streaming works everywhere​

Set "stream": true and consume SSE chunks in the format of the surface you called — OpenAI-style chat.completion.chunk deltas or Anthropic-style message events. Mid-stream failover is handled before the first token; once a stream starts it sticks to its upstream.