Reference
Frequently asked questions
Short answers to the questions we hear most. Can't find it? Reach support@neuralrouter.ai.
Is Neural Router OpenAI-compatible?
Yes. The API mirrors the OpenAI Chat Completions schema. Point an existing OpenAI SDK at https://api.neuralrouter.ai/v1 with your nr- key and most code works unchanged.
Do you train on my prompts or completions?
No. We never use your prompts or completions to train models, and we don't sell personal data. Workspaces can enable zero-retention so content is never stored.
How does billing work?
You pay for what you route, plus your plan. Set hard monthly budget caps and alerts per org or workspace under Guardrails; routing stops when a cap is reached. See pricing.
How are models chosen?
Set model: "auto" and the router scores eligible models against your objective and policies. Read Model routing and Provider routing.
Can I use my own provider accounts?
Yes — add provider keys under BYOK and route through your own credits and rates.
What about rate limits?
Rate limits are per workspace and reported via x-ratelimit-* headers. Upstream provider limits are absorbed by failover. See the API reference.
Can data stay in a specific region?
Yes. Pin inference to a jurisdiction and enforce zero-retention — see Zero-retention & residency.
How do I list my own models on the marketplace?
Apply through provider onboarding: submit your endpoint, pass automated validation, and your models become eligible for routing. Reputation and SLA then influence your traffic share.