createOpenAIModelProvider

import { createOpenAIModelProvider } from "@elie-laloum/outpost";

Purpose and behavior

Create a reusable provider for an OpenAI-compatible service over Chat Completions (default) or Responses, with tool calls, reasoning effort and streaming. Invalid options fail with code configuration. Each request is sent once, with no retry or protocol fallback.

Complete example and detailed rules.

Parameters and properties

optionsRequired
OpenAIModelProviderOptions
Base URL, protocol, API key and request bounds. Unknown options fail with code configuration.
options.apiOptional
"chat-completions" | "responses" | undefined
HTTP protocol: chat-completions by default, or responses. No automatic protocol fallback.
options.baseUrlRequired
string
Absolute HTTP(S) base URL including its version prefix, such as https://api.openai.com/v1; chat/completions or responses is appended. Credentials, query or fragment fail with code configuration, and redirects are refused.
options.apiKeyRequired
string | false
Explicit bearer API key, or false for an unauthenticated endpoint. No environment variable or account login is read automatically.
options.timeoutMsOptional
number | undefined
Request deadline in milliseconds, default 120000, at most 2147483647. While streaming it restarts on each received chunk; expiry rejects with code timeout.
options.maxResponseBytesOptional
number | undefined
Maximum response body in bytes after decompression, default 8388608 (8 MiB); a streamed response counts all its chunks. A larger response fails with code response.

Returns

ModelProvider

Signature

export declare function createOpenAIModelProvider(
  options: OpenAIModelProviderOptions,
): ModelProvider;

Purpose and behavior

Parameters and properties

Returns

Signature

Related contracts