Hosted inference

A hosted LLM inference API for applications that do not want to run GPUs

The gateway handles access to configured upstream models. Your application still owns prompt design, output validation, retries and user-facing safety.

Planned recharge methods: Alipay and USDT. Collection is not enabled yet.

What the service handles

  • Authentication
  • Model routing within the configured catalog
  • Usage accounting
  • Rate and budget enforcement
  • Public model metadata

What your application handles

  • Input and output validation
  • Business authorization
  • Retry policy
  • Data minimization
  • User experience when a model is unavailable

Common questions

Is the API compatible with the OpenAI SDK?

The public gateway follows the OpenAI-compatible chat completions pattern. Confirm model-specific features in the live catalog and documentation before production rollout.

Where do model availability and prices come from?

Public model pages read the active catalog exposed by the service. A model is not presented as available merely because its name appears in an article.

Related resources