docs: hosted gptel-make-openai host/endpoint gotchas vs Responses API

[cursor]

IF your feature request is related to a problem

gptel-make-openai treats api.openai.com as Responses API (/v1/responses). A hosted Chat Completions host must use a different :host and keep :endpoint "/v1/chat/completions". :host is hostname only (no https://). Easy to paste a full URL and get a bad request.

Describe the solution you'd like

Working backend for a live Chat Completions host (PZERO). Public catalog: GET https://api.pzero.studio/v1/models. Default text id: deepseek-v4-flash. Prepaid Bearer key (pzero_…).

(setq gptel-model 'deepseek-v4-flash
      gptel-backend
      (gptel-make-openai "PZERO"
        :host "api.pzero.studio"
        :protocol "https"
        :endpoint "/v1/chat/completions"
        :stream t
        :key (getenv "PZERO_API_KEY")
        :models '(deepseek-v4-flash)))

Do not set :host "api.openai.com" (that selects Responses). Do not put https:// in :host. Endpoint here is the chat path (gptel concatenates protocol+host+endpoint). That is the opposite of Avante/LiteLLM /v1 roots.

Describe alternatives you've considered

The Groq/Mistral examples in the README already use gptel-make-openai with a custom host. This is that shape with the Responses-dispatch gotcha spelled out.

Additional context

https://github.com/karthink/gptel/blob/master/gptel-openai.el (host matching api.openai.com switches constructor)