Use your ChatGPT subscription via Codex (no API key)

Dimmy can run OpenAI's official codex CLI as a subprocess and use it as the LLM provider for style rewrites and meeting recaps. The CLI owns your login session (your ChatGPT Plus / Pro / Team / Business / Enterprise plan), and Dimmy only talks to its stdin/stdout. Dimmy never reads your credentials file.

Prerequisites

• An active paid ChatGPT plan: Plus, Pro, Team, Business, or Enterprise. A free ChatGPT account won't work for Codex.
• The codex CLI installed. It ships as a native binary, so Node.js is optional.
• A browser, for the sign-in flow.

Install the CLI

If you have Node.js, the fastest path is npm:

bash
npm install -g @openai/codex

No Node.js? Grab the native installer from OpenAI (developers.openai.com/codex/cli). Open a NEW terminal so PATH refreshes, then verify it's there:

bash
codex --version

Sign in

Easiest path: let Dimmy do it. Open the ChatGPT (Codex) card in Dimmy (see below) and click Sign in with ChatGPT. Dimmy launches codex login in a terminal, you complete the ChatGPT login in your browser, and Dimmy detects when it's done. You can also just run codex login in a terminal yourself.

Turn it on in Dimmy

The connection card lives on Settings → Integrations in the "OpenAI subscription" group. Where you flip the provider on differs slightly per platform.

1. 1Open the ChatGPT (Codex) card on Settings → Integrations. If nothing is installed yet, click Set up wizard for the guided install + sign-in walkthrough.
2. 2If it's not signed in, click Sign in with ChatGPT and finish the browser login.
3. 3Click Test (Test connection on Windows) to send a tiny ping through codex and confirm it answers end-to-end.
4. 4Flip on the subscription toggle as in the table above.
5. 5Press your hotkey and dictate a sentence with an LLM style active. The result now comes from your ChatGPT subscription.

‹›

Which model runs

When you use the card's toggle, Dimmy points the LLM at codex://default, so Codex uses its own configured default model. If instead you keep the OpenAI provider selected and turn on the subscription toggle, Dimmy passes the OpenAI model you picked to codex. There's no separate Dimmy model dropdown on the Codex card itself.

What this affects

• Style rewrites (Correct, Summarize, Elaborate, Prompt, and friends). Every dictation that passes through an LLM style now uses your subscription.
• Meeting recaps, when the subscription toggle is on (one toggle on macOS, the separate recap toggle on Windows).
• Custom prompts you defined in Custom prompts. They keep working, only the provider changes.

What this does NOT affect

• STT (speech-to-text). That still runs locally (Whisper) or via your chosen STT provider. This integration is LLM-only.
• Telemetry. Dimmy records aggregate events (status probed, login spawned, invocation success / failure) with no prompt or response content. See Telemetry.
• Other AI tools on your machine. Dimmy spawns its own subprocess; it does not interfere with codex when you use it manually.

Cost and quotas

Dimmy makes one CLI call per LLM action (one rewrite, one recap). Each call counts against your ChatGPT plan's usage. There is no separate Dimmy quota: the CLI talks to OpenAI with your account.