Skip to main content
This guide assumes you already have:
  • A virtual key with the sk-corvex- prefix. If you don’t, ask your team admin or see Authentication.
  • The host of your Corvex inference gateway (defaults shown below use http://localhost:8080; replace with your environment’s gateway URL).

​
Send a chat completion

Already on the OpenAI SDK? Swap 3 lines.

The base-URL/API-key swap shown on the drop-in page is validated in CI against staging on every docs PR.
curl https://your-gateway.example.com/v1/chat/completions \
  -H "Authorization: Bearer sk-corvex-YOUR_VIRTUAL_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "zai-org/GLM-5.1-FP8",
    "messages": [
      {"role": "user", "content": "Hello, world."}
    ]
  }'
If your client library defaults to OpenAI’s hosted endpoint, set base_url (Python) / baseURL (TypeScript) — pointing it at your Corvex gateway is the only change needed.

​
Model naming

Corvex models use bare lab-canonical Hugging Face repo ids — <hf-namespace>/<model-name> (e.g. zai-org/GLM-5.1-FP8, deepseek-ai/DeepSeek-V4-Flash). There is no Corvex prefix; legacy corvex/<…> and corvex-pd/<…> ids return HTTP 404 since the Day-1 dogfooding cutover (see ADR product-024 / RD-651, RD-651). Topology — standard vs prefill-decode disaggregated — is decided server-side from the catalog, not from the request shape. Get the current list of available models for your virtual key: 
curl https://your-gateway.example.com/v1/models \
  -H "Authorization: Bearer sk-corvex-YOUR_VIRTUAL_KEY"

​
What’s next

The OpenAI SDK drop-in examples on the drop-in page are compiled and run against staging on every docs PR (REQ-081). The broader samples on this page are illustrative.