OneAPI + Replicate + OpenAI-compatible

Global token API demo for the current live route

This demo now follows the route already tested in production: overseas server, OneAPI gateway, Replicate upstream, and a single public model for first-stage rollout.

Single public model

First version is locked to one model only: meta/meta-llama-3-8b-instruct.

OpenAI-compatible call

Keep the usual client style and only change the base URL, key, and model name.

Live route already verified

The real tested path is OneAPI to Replicate, not the old domestic Qwen route.

Demo pages stay local

These pages are for explanation and demonstration. The real API test address is shown on the docs page.

Current rollout logic

1. Keep one path only

Do not mix domestic closed-model resale with the overseas open-model route.

2. Test one model first

Only keep one public model name and one upstream route until the first batch is stable.

3. Then add pages and billing

After the call path is stable, finish pages, payment, and compliance checks.