OpenAI

Runtime provider id: openai. Defaults to the OpenAI Responses API.

Provider options

Field	Type	Notes
apiKey	string	Required (or set via environment).
baseURL	string	Optional. Override the API endpoint.
headers	Record<string, string>	Optional. Extra HTTP headers.
api	"responses" | "chat"	Optional. Which OpenAI-compatible API surface to use; defaults to "responses".
reasoning	object	Optional. See Reasoning. Only honored on api: "responses".

api: "responses" (the default) streams reasoning (response.reasoning_summary_text.delta) so it shows up in the UI. api: "chat" is for OpenAI-compatible MaaS/proxies that do not implement the Responses API — it does not surface reasoning, since the SDK’s Chat adapter drops reasoning_content. For a proxy that only exposes thinking via chat reasoning_content (e.g. Kimi/Moonshot), route it through moonshot or openrouter instead.

Reasoning

Reasoning controls only apply to the Responses API (api: "responses"):

• effort: reasoning effort. Some models (e.g. GPT-5) only emit a reasoning summary at medium or high; low/minimal yield no visible thinking.
• summary: summary verbosity. Defaults to auto. Set to none to disable summaries (e.g. for non-reasoning models that reject the reasoning param).
• display is not used by OpenAI.

Example configs

{
  "name": "OpenAI",
  "provider": "openai",
  "options": {
    "apiKey": "sk-..."
  }
}

{
  "name": "GPT-5",
  "provider": "OpenAI",
  "options": {
    "model": "gpt-5",
    "reasoning": { "effort": "high", "summary": "detailed" }
  },
  "default": true
}

Chat Completions mode against an OpenAI-compatible proxy that doesn’t implement Responses:

{
  "name": "OpenAI Proxy",
  "provider": "openai",
  "options": {
    "apiKey": "sk-...",
    "baseURL": "https://litellm.internal/v1",
    "api": "chat"
  }
}

{
  "name": "GPT-4o (Chat API)",
  "provider": "OpenAI Proxy",
  "options": {
    "model": "gpt-4o"
  }
}