How is ModelRiver different from a regular AI gateway?
A typical AI gateway sits between your app and AI providers and stops at the response: one API, routing, fallbacks, keys and logs. ModelRiver includes all of that, and also handles what happens around the AI call: delivering results to the browser over WebSocket, sending them through your backend with webhooks, chaining steps into pipelines and remembering conversations.
What does ModelRiver share with an AI gateway?#
The gateway part is all there, on every plan:
- One API for 13 providers, with your own provider keys
- OpenAI-compatible endpoints, so existing SDKs work
- Automatic failover to up to two backup models
- Response caching, spend limits and guardrails
- Request logs with tokens, timing and every failover attempt
What does ModelRiver add on top?#
| You need to… | With a gateway alone | With ModelRiver |
|---|---|---|
| Show the answer live in the browser | Build your own streaming relay or socket server | Async requests deliver to the browser over WebSocket; the client SDK handles reconnects |
| Process the result in your backend first | Write your own queue and callback handling | A backend pipeline sends the result to your webhook, waits for your callback, then pushes the final result to the user |
| Chain several steps | Orchestrate them in your own code | Up to 6 steps after the first AI call, mixing your backend steps with other AI workflows |
| Notify your servers | Poll or build it yourself | Signed webhooks (HMAC-SHA256) with automatic retries |
| Remember a conversation | Store and resend the history yourself | Sessions keep it, with live updates for your app or support dashboard |
| Get valid JSON | Validate in your code | Structured outputs are checked, and backups are tried if a model fails |
| Change models or prompts safely | Redeploy your app | Draft and Live versions, with publish and rollback |
Every step shows up in the same request Timeline: the AI attempts, webhook deliveries and backend callbacks.
Do I have to use all of it?#
No. Use ModelRiver as a plain gateway with sync requests or the OpenAI SDK, and add async delivery, webhooks, pipelines or sessions only when you need them. Every plan includes every feature, so you don't pay more to use them.