nRouter's API Reference documents the whole surface of its gateway: endpoints, authentication, and errors, in one page.
One key, two wire formats
Every request takes a scoped virtual key (sk-nrouter-...) as a Bearer token against https://api.nrouter.ai/v1. Standard SDKs connect by changing the base URL and the key. The same key speaks the OpenAI wire format at /v1/chat/completions (streamed or buffered, with guardrails and provider fallback) and the Anthropic Messages format at /v1/messages, which also has a count_tokens endpoint for counting input tokens before sending.
The rest of the endpoint set
- GET /v1/models lists models across every connected provider; GET /v1/models/{model_id} retrieves one model
- POST /v1/embeddings for semantic search, RAG and classification
- POST /v1/images/generations for image generation
- POST /v1/audio/speech, /v1/audio/transcriptions (timestamps, language detection) and /v1/audio/translations
- POST /v1/videos starts an asynchronous video job; GET /v1/videos/{id} polls it and GET /v1/videos/{id}/content downloads it — polling and download are unmetered
Cross-cutting behaviour
A four-phase preflight security check runs before dispatch, rates are pass-through list rates, and real-time latency and spend are reported back on x-nr- response headers for every call.
Full reference: https://nrouter.ai/docs/api-reference