One model name. The platform chooses per request. nrouter/auto, charged at the served model's list price
Send nrouter/auto as the model and the platform chooses a model for each request. A decision model, TypeSafe Jev System One, picks from your organization's candidates within a 400ms budget; if it has no confident answer in time, signals in the request choose instead. You are charged from your credit balance at the list price of the model that served the request, with 0% token markup.
What a request to nrouter/auto carries
- Decision budget
- < 400ms
- Token markup
- 0%
- Fee on credit purchases
- 0%
- Served model
- Shown
After that, request signals choose the model
Charged at the served model’s list price
A purchase charges exactly the credits bought
Named in the response and in your logs
Set the model to nrouter/auto
It works on the chat completions, messages and responses endpoints with the SDK you already use. Replace the model name with nrouter/auto; nothing else in the request changes.
| 1 | curl https://api.nrouter.ai/v1/chat/completions \ |
| 2 | -H "Content-Type: application/json" \ |
| 3 | -H "Authorization: Bearer $NROUTER_API_KEY" \ |
| 4 | -d '{ |
| 5 | "model": "nrouter/auto", |
| 6 | "messages": [ |
| 7 | { |
| 8 | "role": "user", |
| 9 | "content": "Analyze our Q3 cloud infrastructure spend and suggest 3 cost optimizations." |
| 10 | } |
| 11 | ], |
| 12 | "temperature": 0.2 |
| 13 | }' |
nrouter/auto and OpenRouter
What each charges and who chooses the model. OpenRouter's figures are the ones it publishes.
| What | nRouter (nrouter/auto) | OpenRouter |
|---|---|---|
| Fee on credit purchases | 0%: a purchase charges exactly the credits bought | Publishes a 5.5% fee with a $0.80 minimum |
| Token price | Exact list price of the served model, 0% markup | Provider price, plus the purchase fee above |
| Who chooses the model | A decision model within a 400ms budget, then request signals | See its auto router documentation |
| Narrowing the candidates | Owners and Org Admins exclude or pin models per organization | See its auto router documentation |
The full comparison is on the nRouter vs OpenRouter page.
What happens to a request for nrouter/auto
Five steps, in order, from the moment the request arrives to the moment it is charged.
Checks before any spend
Your key, model access, limits and guardrails are checked first. A request refused here spends $0 and holds $0.
Candidates narrowed
The candidates are the published models your organization may call, minus any your Owner or Org Admin excluded in Router Settings.
Model chosen (<400ms)
TypeSafe Jev System One picks one candidate within a 400ms budget. With no confident answer in time, signals in the request choose instead.
Credit held
Before the provider is called, enough credit is held to cover the most expensive model in the chain the request may try.
Served and settled
The provider answers. The request settles at the list price of the model that served it and the rest of the hold is released.
No markup, no routing fee
You pay the list price of the model that served the request. nrouter/auto adds nothing on top.
Questions about nrouter/auto
How the model is chosen, how a request is charged and how to limit the candidates.
How does nrouter/auto decide which model to call?
When a request arrives with the model nrouter/auto, the gateway asks a decision model, TypeSafe Jev System One, which of your organization’s candidates fits the request. It has a 400ms budget. If it answers confidently in time, that model serves the request. Otherwise the gateway picks from the same candidates using signals in the request itself, such as its size and the features it uses. A request is never refused because the decision was slow.
How is a request to nrouter/auto charged?
Like a request to any named model: from your organization’s credit balance, at the list price of the model that served it. nrouter/auto adds no routing fee and no token markup. Organizations still on a subscription plan keep that plan’s nrouter/auto allowance.
Can I limit which models nrouter/auto may pick?
Yes. In the dashboard under Router Settings, the nRouter/Auto Models tab lists every published model nrouter/auto may pick from. Owners and Org Admins can exclude models or pin an explicit list. Members and viewers see the list read-only. A limit only narrows the candidates; it never grants access to a model your organization could not otherwise call.
How much credit is held while the request runs?
Before the provider is called, the gateway holds enough credit to cover the most expensive model in the chain it may try. When the response completes, the request settles at the list price of the model that actually served it and the rest of the hold is released.
How do I know which model answered?
The response names it. The x-nr-model response header and the model field in the response body carry the served model, and the same name appears in your request logs.
Get started
Send your first request to nrouter/auto
Buy credits from $5, create a key and set the model to nrouter/auto.