OpenAI-compatible

{
  "kind": "openai",
  "name": "openai",
  "baseUrl": "https://api.openai.com/v1",
  "apiKeySecret": "openai.key"
}

Field	Meaning
baseUrl	required; requests go to <baseUrl>/chat/completions
apiKeySecret	secret reference, sent as authorization: Bearer …
headers	extra static headers
headerSecrets	header name to secret reference, e.g. Azure’s api-key
query	query parameters, e.g. Azure’s api-version

Azure OpenAI

{
  "kind": "openai",
  "name": "azure",
  "baseUrl": "https://my-resource.openai.azure.com/openai/deployments/gpt-4o",
  "headerSecrets": { "api-key": "azure.openai.key" },
  "query": { "api-version": "2024-10-21" }
}

vLLM and LM Studio

Point baseUrl at the server’s OpenAI-compatible endpoint, for example http://vllm:8000/v1. If the server needs no key, leave out apiKeySecret. Raise the clearance to confidential or restricted for models that run on your own hardware.