NVIDIA and domestic silicon pooled together, exposing exactly one thing to the outside: an OpenAI-compatible token endpoint. Registration mints a default API key — no approval queue, no instance types to pick.
Language, multimodal, embedding, reranking and speech all share one base_url — switching models is a single field.
Every response carries its routing record: silicon family, compute centre and time-to-first-token.
Rate limits, quotas and per-call records show up in the console in real time. Split keys per workload and they never crowd each other out.
Use the email and password you registered with. You will land in the console.
Registration mints a default API key, so you can make a real call right away.