Iris

A dependency-free TypeScript client that calls OpenAI's Codex (ChatGPT) backend using your ChatGPT subscription (OAuth device-code) instead of an OPENAI_API_KEY. It exposes a familiar Chat Completions API and translates it to the Responses API under the hood.

Quickstart

npm install
npm run build
npm run smoke      # verifies auth + a real call end-to-end (logs you in if needed)

Usage

import { chatCompletionsCreate } from "./dist/index.js";

const res = await chatCompletionsCreate({
  model: "gpt-5.5",
  messages: [{ role: "user", content: "Say hello in one sentence." }],
});
console.log(res.choices[0]?.message.content);

Streaming, device-code login, env vars (IRIS_AUTH_FILE, IRIS_BASE_URL, IRIS_MODEL, …), the ~/.iris/auth.json store, and how credentials are resolved are all documented in docs/iris-auth.md.

Requirements

• Node 18+
• A ChatGPT subscription with Codex access (Plus/Pro/Team/etc.) — this talks to the ChatGPT-Codex backend, so a plain platform OPENAI_API_KEY account won't work here.

Scripts

Command	What it does
npm run build	Compile TypeScript to dist/
npm test	Run the vitest suite
npm run check	Build + test
npm run smoke	Build, then make a real authenticated call to confirm it works

License

MIT — not affiliated with or endorsed by OpenAI. Use at your own risk.