@galdor/provider-openai

The OpenAI (Chat Completions API) provider adapter for galdor. Implements the core Provider interface, so it drops into any graph, agent, or council.

Because it speaks the Chat Completions wire format, it also targets OpenAI-compatible endpoints (Groq, Together, MiniMax, Mistral, DeepSeek, vLLM, Ollama, …) via baseURL. Covers generate + streaming, tool calling, structured output, vision, reasoning effort, the typed error taxonomy with Retry-After, cancellation, and a stream-safe header timeout (60 s default).

Install

bun add @galdor/provider-openai   # or: npm install @galdor/provider-openai

Usage

import { newOpenAI } from "@galdor/provider-openai";
import { userMessage, messageText } from "@galdor/core/schema";

const provider = newOpenAI({ apiKey: process.env.OPENAI_API_KEY! });

const res = await provider.generate({
  model: "gpt-4o-mini",
  messages: [userMessage("Explain MCP in one sentence.")],
});
console.log(messageText(res.message), res.usage);

Point it at an OpenAI-compatible backend with baseURL:

const groq = newOpenAI({ apiKey: process.env.GROQ_API_KEY!, baseURL: "https://api.groq.com/openai/v1" });
const local = newOpenAI({ apiKey: "no-key", baseURL: "http://localhost:11434/v1" }); // Ollama

Tip: @galdor/providerset resolves these aliases by name (groq, ollama, …) with the right preset baseURL.

License

Apache-2.0