OpenAI-compat: Custom remote provider with trailing-slash /v1/, not OpenAI card

Working config (Remote Providers → Custom, BASE_URL /v1/)

The OpenAI / Groq / Mistral cards whitelist first-party ids (gpt-4o, etc.). Custom is the OpenAI-compatible slot. BASE_URL is the /v1 root with a trailing slash so models resolves as {base}/models (QUrl::resolved). Origin-only posts /chat/completions (404). Pasting the chat path doubles /chat/completions. Host appends /chat/completions onto the path.

A Chat Completions host whose public catalog is GET https://api.pzero.studio/v1/models (no key) fits Custom if BASE_URL is https://api.pzero.studio/v1/ and MODEL_NAME is a catalog id.

Add Model → Remote Providers → Custom
BASE_URL: https://api.pzero.studio/v1/
API_KEY: <Bearer key for that host>
MODEL_NAME: deepseek-v4-flash

Keep embeddings off this host.