What counts as a request on my ModelRiver plan?
Updated
Each production AI call counts as one request, whichever endpoint you use. Test mode, the Playground, webhook deliveries and callbacks don't count.
What counts?#
- Sync, async and OpenAI-compatible calls
- Streaming requests
- Responses served from the cache
- Each AI step in a backend pipeline
- Failed calls
What doesn't count?#
- Requests to a workflow in test mode
- Playground runs in the console
- Webhook deliveries to your server
- Callbacks from your backend
- Messages added to a conversation through the Sessions API without calling a model
If a request fails over to a backup model, is that two requests?#
No. A request counts once, however many backup models are tried.
Why do cache hits and failed calls count?#
ModelRiver still does the work: routing, checking, logging and responding. A cache hit saves you the provider's token cost, not the request.
Is the per-minute rate limit counted the same way?#
Not quite. The rate limit is per project and is about traffic, so it also includes your backend's callbacks. If you hit it often, spread requests out or move to a plan with a higher limit.
Where can I see my usage?#
The Billing & usage page in the console shows requests used this billing period, counted across all your projects.